The Evolution of a High Mutation Rate and Declining Fitness in Asexual Populations

University of Pennsylvania ScholarlyCommons

Publicly Accessible Penn Dissertations

2012

Christopher Francis Gentile University of Pennsylvania, [email protected]

Follow this and additional works at: https://repository.upenn.edu/edissertations

Part of the Biology Commons, and the Evolution Commons

Recommended Citation Gentile, Christopher Francis, "The Evolution of a High Mutation Rate and Declining Fitness in Asexual Populations" (2012). Publicly Accessible Penn Dissertations. 510. https://repository.upenn.edu/edissertations/510

This paper is posted at ScholarlyCommons. https://repository.upenn.edu/edissertations/510 For more information, please contact [email protected]. The Evolution of a High Mutation Rate and Declining Fitness in Asexual Populations

Abstract Simulations of asexual populations undergoing continual adaptation present a definite prediction: mutator hitchhiking should drive the mutation rate upwards in an asexual population until it reaches an intolerable level, at which point the population will be driven extinct. Experimental studies have shown that a mutator allele can readily hitchhike to fixation with beneficial mutations in an asexual population having a low, wild-type mutation rate. Using Eshcerichia coli, I show that a genotype bearing two mutator alleles can supplant an asexual population already fixed for one mutator allele. My results provide experimental support for recent theory predicting that mutator alleles will tend to accumulate in asexual populations by hitchhiking with beneficial mutations, causing an ever-higher genomic mutation rate. Interestingly, results from recent simulations also suggest that the deleterious mutational load which accompanies an increased mutation rate is not immediately incurred. The delay in accruing a mutational load may potentially allow a succession of mutator hitch hiking events to occur before the cost of the initial hitchhiking event is incurred. Ultimately, mutation rates in asexual populations may evolve upwards to a rate that threatens extinction due to overwhelming mutational pressure. To test this prediction, I established independent populations of E. coli bearing either a single- or double-mutator genotype and propagated them for several thousand generations. At the conclusion of the experiment, the growth rate of the single-mutator populations had risen substantially, consistent with the substitution of beneficial mutations; in contrast, the growth rate of the double-mutator populations had fallen significantly, consistent with erosion of fitness by deleterious mutations. In addition, both sets of populations had maintained the mutation rate present at the outset of the experiment. Strikingly, the experiment seems to have provided evidence of mutation-driven fitness decline in the double-mutator populations despite the presence of beneficial mutations. I address the experimental results in the context of theoretical predication of fitness decline at a high mutation ates.r

Degree Type Dissertation

Degree Name Doctor of Philosophy (PhD)

Graduate Group Biology

First Advisor Paul Sniegowski

Keywords asexual, decline, deleterious, evolution, fitness, mutation ater

Subject Categories Biology | Evolution

This dissertation is available at ScholarlyCommons: https://repository.upenn.edu/edissertations/510 THE EVOLUTION OF A HIGH MUTATION RATE AND DECLINING FITNESS IN ASEXUAL POPULATIONS

Christopher F. Gentile

A DISSERTATION in Biology Presented to the Faculties of the University of Pennsylvania in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy 2012

Paul Sniegowski, Associate Professor, Department of Biology, University of Pennsylvania Dissertation Supervisor

Doris Wagner, Associate Professor, Department of Biology, University of Pennsylvania Graduate Group Chairperson

Dissertation Committee: Paul Schmidt, Associate Professor, Department of Biology, University of Pennsylvania Dustin Brisson, Assistant Professor, Department of Biology, University of Pennsylvania Joshua Plotkin, Associate Professor, Department of Biology, University of Pennsylvania Carlo Maley, Associate Professor, Department of Surgery, University of California, San Francisco Philip J. Gerrish, Associate Professor, Department of Biology, University of New Mexico

ii ACKNOWLEDGMENTS

Many people have influenced me, and I am grateful to each one. This dissertation could not have been possible without my advisor, Paul Sniegowski, whose enthusiasm for science and creativity are inspiring. Paul is a great advisor - a critical thinker, a caring mentor, and an excellent role model. Through his patient advice and encouragement, I have grown as a scientist, a writer, and a person.

Everyone in the Sniegowski lab contributed in some way to the work presented here. I was fortunate to work with a fantastic group of labmates over the past six years, including Szi-Chieh Yu, Sebastian Akle, Matthew Gazzara, Dave Katzianer, Kate Kerpen, Emily Parke, Angela Halstead, Dan Balk, Tanya Singh, and Mitra Eghbal. I am particularly indebted to my fellow graduate student, Eugene Raynes, for his scientific advice, contributions, and occasional clarity of thinking. My committee - Dustin Brisson, Josh Plotkin, Carlo Maley, Paul Schmidt, and, in particular, Phil Gerrish- has provided excellent advice and suggestions.

Most importantly, my family has provided encouragement and support. I thank my parents and siblings for showing me the value of curiosity, hard work, and lifelong learning. My wife, Amanda, has been extraordinarily patient with me and all the time I have spent in lab. She has been a constant source of love and encouragement since we first met. My daughter Lily has been a fantastic helper in lab: building castles out of tape rolls, shooting darts at me and Eugene, drawing on the incubator, and reminding me that there was ALWAYS time to go see the ducks, turtles, and fish at the pond. Finally, my daughter Violet has been a joy to come home to every day and constant reminder that the best parts of life are found outside the lab. I am eternally grateful for all the love and support my family has given me.

iii ABSTRACT

THE EVOLUTION OF A HIGH MUTATION RATE AND DECLINING FITNESS IN ASEXUAL POPULATIONS Christopher F. Gentile

Paul Sniegowski

Simulations of asexual populations undergoing continual adaptation present a definite prediction: mutator hitchhiking should drive the mutation rate upwards in an asexual population until it reaches an intolerable level, at which point the population will be driven extinct. Experimental studies have shown that a mutator allele can readily hitchhike to fixation with beneficial mutations in an asexual population having a low, wild-type mutation rate. Using Eshcerichia coli, I show that a genotype bearing two mutator alleles can supplant an asexual population already fixed for one mutator allele.

My results provide experimental support for recent theory predicting that mutator alleles will tend to accumulate in asexual populations by hitchhiking with beneficial mutations, causing an ever-higher genomic mutation rate. Interestingly, results from recent simulations also suggest that the deleterious mutational load which accompanies an increased mutation rate is not immediately incurred. The delay in accruing a mutational load may potentially allow a succession of mutator hitch hiking events to occur before the cost of the initial hitchhiking event is incurred. Ultimately, mutation rates in asexual populations may evolve upwards to a rate that threatens extinction due to overwhelming mutational pressure. To test this prediction, I established independent populations of E. coli bearing either a single- or double-mutator genotype and propagated them for several iv thousand generations. At the conclusion of the experiment, the growth rate of the single- mutator populations had risen substantially, consistent with the substitution of beneficial mutations; in contrast, the growth rate of the double-mutator populations had fallen significantly, consistent with erosion of fitness by deleterious mutations. In addition, both sets of populations had maintained the mutation rate present at the outset of the experiment. Strikingly, the experiment seems to have provided evidence of mutation- driven fitness decline in the double-mutator populations despite the presence of beneficial mutations. I address the experimental results in the context of theoretical predication of fitness decline at a high mutation rates.

v TABLE OF CONTENTS

Chapter 1. Introduction ...... 1 1.1. Mutation rate evolution ...... 1 1.2. Dissertation Structure ...... 3 1.3. Escherichia coli and experimental evolution ...... 4 Chapter 2. Competition between high- and higher-mutating strains of Escherichia coli 6 2.1. Introduction ...... 6 2.2. Methods and Materials ...... 8 2.2.1. Bacterial strains and media ...... 8 2.2.2. Mutation rate assay ...... 10 2.2.3. Competitions under soft selection ...... 10 2.2.4. Competitions under hard selection ...... 11 2.3. Results and Discussion ...... 12 2.3.1. Competitions under soft selection ...... 12 2.3.2. Competitions under hard selection ...... 15 2.3.3. Discussion ...... 18 Chapter 3. Declining fitness in a higher-mutating strain of Escherichia coli ...... 20 3.1. Introduction ...... 20 3.2. Methods and Materials ...... 23 3.2.1. Bacterial strains ...... 23 3.2.2. Long-term propagation of populations ...... 25 3.2.3. Fitness assay...... 25 3.2.4. Temperature sensitivity assay ...... 26 3.2.5. Competitive fitness assay ...... 27 3.2.6. Fitness variance assay ...... 28 3.2.7. Extinction risk assay ...... 29 3.2.8. Mutation rate assay ...... 30 3.2.9. Sequencing assay ...... 30 3.3. Results ...... 31 3.3.1. Fitness decreased in double-mutator populations ...... 31 3.3.2. The risk of extinction increased in double-mutator populations ...... 34

vi 3.3.3. Competitive fitness decreased in double-mutator populations ...... 36 3.3.4. Double-mutator populations evolved an increased sensitivity to high ambient temperature ...... 39 3.3.5. Double-mutator populations maintained their mutator genotype and phenotype ...... 41 3.3.6. Fitness variance assay ...... 41 3.4. Discussion...... 44 3.4.1. Limitations of previous work on mutation-driven fitness decline ...... 44 3.4.2. Fitness diverged and ultimately declined in double-mutator populations ... 46 3.4.3. The mutation rate did not decrease in the double-mutator populations ...... 50 3.4.4. The consequences of a very high mutation rate ...... 51 Chapter 4. General Discussion ...... 59 4.1. Introduction ...... 59 4.2. Primary results ...... 59 4.3. Limitations of the experimental approach and future directions ...... 60 4.4. Implications ...... 63 Bibliography ...... 66

vii LIST OF TABLES

Table 3-1: A list of strains used in this work and relevant genotype information...... 24 Table 3-2: Results of extinction risk assay. Rows report the transfer outcome, Columns report the genotypes ...... 35

viii LIST OF FIGURES

Figure 2-1: Estimates of the per base pair mutation rate to nalidixic acid resistance in the single- and double-mutator strains. Error bars represent 95% confidence intervals. 9 Figure 2-2: Competitions under soft selection. Blue lines: populations propagated in DM25; red lines: populations propagated in DM1000. Ten populations in DM25 in which the double-mutator began at low frequency (~0.03) and was lost within 70 generations are not shown...... 14 Figure 2-3: Competitions under hard selection. Grey boxes indicate periods of antibiotic exposure. The number of generations reported in the figure is a minimal estimate because of sharp reductions in population size and subsequent regrowth upon exposure to antibiotics...... 17 Figure 3-1: Evolution of fitness (as measured by the inverse of doubling time). The doubling time of each population was measured at each archived time point. The inverse of the average doubling time of 5 replicates is plotted. Blue lines: single- mutator populations; red lines: double-mutator populations...... 33 Figure 3-2: Competitions between evolved double-mutator and ancestral single-mutator genotypes. Solid lines: populations propagated in DM25; dashed lines: populations propagated in DM1000...... 38 Figure 3-3: Results of temperature sensitivity assay. – denotes no visible growth and + denotes visible growth after 24 hours...... 40 Figure 3-4: The variance in fitness (inverse of doubling time) assayed over time in a random, representative single-mutator (top) and double-mutator (bottom) population plotted over the course of propagation. The predicted variance threshold ( ) and error threshold ( ) for each population are also plotted...... 43

ix Chapter 1. Introduction

1.1. Mutation rate evolution

The genomic mutation rate is influenced by multiple genetic pathways and numerous loci

(Friedberg et al. 1995) and can reasonably be considered a polygenic quantitative trait.

Alleles at loci affecting the mutation rate may be considered ―mutation rate modifier alleles‖ and are presumably subject to natural selection. There are potentially two types of natural selection on mutation rate modifier alleles, direct and indirect (Sniegowski et al. 2000). Direct selection acts via the immediate fitness effects a mutation rate modifier allele may have. For example, if a mutation rate modifier allele coincidentally increases the speed of DNA replication and allows faster cell growth and division, it may have an immediate beneficial fitness effect and may be subject to direct selection. In contrast, indirect selection acts on the fitness of a mutation rate modifier allele‘s genetic background. For example, one or more beneficial mutations may arise on the same background as a mutator allele (a mutation rate modifier allele which increases the genomic mutation rate). Indirect selection via these beneficial mutations may drive the fixation of both the beneficial mutations and the mutator allele. This example also serves to illustrate the phenomenon of mutator hitchhiking, whereby mutator alleles rise in frequency towards fixation with associated beneficial mutations, which results in an increased genomic mutation rate across the population.

1 The strength of indirect selection is strongly influenced by the level of linkage disequilibrium between mutation rate modifier alleles and fitness-affecting loci, which is eroded by recombination. In recombining populations, mutation rate modifier alleles can be separated from the genetic background in which they arose, thus limiting indirect selection via any potentially linked deleterious or beneficial mutations. This suggests that sexual populations have relatively weak indirect selection to modify mutation rates, and that they may evolve to a potential equilibrium rate which balances the cost of deleterious mutations with the direct fitness costs of increased replication (Kondrashov

1995; Sniegowski et al. 2000; but see Lynch 2008). In contrast to sexual populations, asexual populations are likely to be subject to strong indirect selection of mutation rate modifiers because of linkage disequilibrium. The notion that indirect selection should act to set an optimal equilibrium mutation rate in an asexual population was first proposed by

Fisher (1930) and has since received significant theoretical and experimental treatments

(see chapter III).

Several relatively recent observations suggest that the theory that mutation rates of asexual populations evolve to an optimum may not be adequate. That is not to suggest that there is no optimal mutation rate, but rather that populations do not necessarily maintain it. Many of the models predicting evolution of an optimal mutation rate assume that genetic load changes concurrently and immediately with changes in the mutation rate. Johnson (1999), however, noted that there is a delay between an increase in mutation rate and a corresponding increase in mutational load. Thus, the full deleterious effects of an increased mutation rate are not immediately subject to selection. Theory 2 (Andre and Godelle 2005) and simulations (Gerrish et al. 2007) suggest that this delay can allow the occurrence of recurrent mutator hitchhiking events in an asexual population, resulting in an upward bias in mutation rate evolution. In theory, the mutation rate of an asexual population may even evolve upwards to a rate that increases the risk of extinction due to the overwhelming accumulation of deleterious mutations

(Gerrish et al. 2007).

1.2. Dissertation Structure

The rest of chapter I reviews some of the important context required to understand the results presented in the dissertation. This includes a review of previous relevant experimental evolution studies and the advantages of Escherichia coli as a model organism.

Chapter Two present the results of work addressing the potential for recurrent mutator hitchhiking in an asexual population. This work investigates the conditions under which mutator hitchhiking is probable and addresses previous reports of mutators in natural and experimental microbial populations. The results provide some experimental support for recent theory predicting that mutator alleles will tend to accumulate in asexual populations by hitchhiking with beneficial mutations, causing an ever-higher genomic mutation rate.

Chapter Three investigates the long-term consequences of bearing a very high mutation rate, using a long-term propagation and experimental evolution approach. The results

3 presented in this chapter analyze the fitness consequences of a very high mutation rate and examine how the mutation rate affects the probability that extinction may occur. The results are examined in the context of theories predicting extinction of an asexual population that carries a very high mutation rate.

Chapter Four reviews the primary results of the previous chapters and examines them in the context of future research directions and opportunities.

1.3. Escherichia coli and experimental evolution

In the work presented here, I investigate the dynamics of mutator hitchhiking and the long-term fitness consequences of carrying a high mutation rate using an experimental evolution approach with E. coli. There are several practical benefits to using E. coli that make it ideal for investigating the evolution of mutation rates in asexual populations: E. coli has a short generation time, it is easy to propagate, and it is easy to maintain a range of population sizes that spans several orders of magnitude. In addition, the genomes of several strains of E. coli -including the strain used in this study- have been completely sequenced (Blattner et al. 1997; Perna et al. 2001); as whole-genome sequencing becomes more readily available, genomic analysis of experimental E. coli populations will be much easier.

There is a rich body of work investigating the genetic basis of mutation rate in E. coli

(Treffers et al. 1954; Lederberg 1965; Liberfarb and Bryson 1970; Siegel and Ivers 1975;

Sevastopoulos and Glaser 1977; Quinones and Piechocki 1985; Schaaper and Dunn 1987;

4 Schaaper and Radman 1989; Fijalkowska et al. 1993; Krishnaswamy et al. 1993;

Schaaper 1993; Oller and Schaaper 1994; Fijalkoswka and Schaaper 1995; Fijalkowska and Schaaper 1996; Schaaper 1996; Notley-McRobb et al. 2002; Szidic and Petranovic

2003). Previous work has also investigated the evolution of mutation rates using E. coli

(Chao and Cox 1983; Fijalkowska et al. 1993; Krishnaswamy et al. 1993; Sniegowski et al. 1997; Notley-McRobb et al. 2002; Shaver et al. 2002). There is also a history of using

E. coli in experimental evolution and propagation experiments (Lenski 1988; Lenski et al.

1991; Travisano et al. 1995; Sniegowski et al. 1997; Holloway et al. 2007; Blount et al.

2008; Khan et al. 2011). Here, I use E. coli to investigate theory suggesting that the mutation rate of an asexual population may evolve upwards to a rate that causes an overwhelming number of deleterious mutations.

5 Chapter 2. Competition between high- and higher-mutating strains of Escherichia coli

This chapter was adapted from:

Gentile, C. F., S.-C. Yu, S. A. Serrano, P. J. Gerrish, and P. D. Sniegowski. 2011.

Competition between high- and higher-mutating strains of Escherichia coli. Biology

Letters 7:422-424.

2.1. Introduction

Theories for the evolution of mutation rates have emphasized the contrasting influence of beneficial mutations in sexual and asexual populations. In sexual populations, recombination erodes linkage disequilibrium and prevents mutator alleles from hitchhiking to fixation with beneficial mutations; in asexual populations, complete linkage enables hitchhiking of mutator alleles (reviewed in Sniegowski et al. 2000).

Classical theories predicted that the trade-off between mutator hitchhiking and increased deleterious mutational load would enforce optimal equilibrium mutation rates in asexual populations (eg. Fisher 1930). Recent theory, however, predicts that the genomic mutation rate of an asexual population will be evolutionarily unstable with a strong upward bias as long as beneficial mutations are occurring, because deleterious mutational

6 load accumulates slowly (Johnson 1999) whereas mutator hitchhiking is relatively rapid

(Andre and Godelle 2005; Gerrish et al. 2007).

Competition experiments between isogenic wild-type and mutator strains in E. coli and

Saccharomyces cerevisiae have shown that mutators will fix by hitchhiking if present at sufficiently high initial frequencies (Chao and Cox 1983; Thompson et al. 2006). Long- term evolution experiments have shown that spontaneously arising mutator alleles can hitchhike to fixation in wild-type E. coli populations (Shaver et al. 2002). Mutators have also been observed at substantial frequencies in natural populations of bacteria (LeClerc et al. 1996; Matic 1997; Boe et al. 2000), cancer cell populations (Loeb 2001), and influenza virus populations (Suarez et al. 1992). Evolved mutator phenotypes have persisted robustly for tens of thousands of generations in experimental E. coli populations

(Woods et al. 2006; Elena et al. 2007), consistent with the idea that the increased deleterious load associated with mutators takes considerable time to accumulate.

Here, I test whether a second mutator allele can hitchhike to fixation in an asexual population already fixed for one mutator allele, thereby raising the genomic mutation rate of the population even further. My approach echoes earlier experiments on the fate of genotypes bearing single mutator alleles in E. coli populations (Chao and Cox 1983; Mao et al. 1997). Here, however, I compete isogenic strains carrying one and two mutator alleles (―single-mutator‖ and ―double-mutator‖ strains). I assess the fate of double- mutators under contrasting regimes: selection primarily affecting growth rate (soft selection) and selection primarily affecting survival (hard selection).

7 2.2. Methods and Materials

2.2.1. Bacterial strains and media

An asexual (F-) strain of E. coli bearing the mutL13 allele, which confers mismatch repair deficiency, was obtained from the E. coli Genetic Stock Center at Yale University. (See

Ch. III, Table 1) The dnaQ905 allele, which confers DNA proofreading deficiency, was cotransduced with a tetracycline resistance determinant from donor strain CSH116

(Miller 1992) into the single-mutator (mutL13) strain to produce an isogenic double- mutator (mutL13 dnaQ905) strain. The point mutation rate of the single-mutator strain was ~100-fold higher than that of an isogenic wild-type strain, and the point mutation rate of the double-mutator strain was a further ~45-fold higher than that of the single- mutator strain (see Fig. 2-1).

Figure 2-1: Estimates of the per base pair mutation rate to nalidixic acid resistance in the single- and double-mutator strains. Error bars represent 95% confidence intervals.

9 2.2.2. Mutation rate assay

I performed fluctuation tests (Luria and Delbruck 1943) on individual random clones from single- and double-mutator populations at the initial and final time point to estimate the rate of mutation to nalidixic acid. Strains were inoculated from frozen stock, grown, and transferred daily for two days in DM1000. For each fluctuation test, 25 cultures were grown from inocula of ~500 cells. Cultures were grown to stationary phase before selective plating. I estimated final population sizes by sampling and plating an appropriate dilution from five cultures on permissive LB plates and counting the number of colonies after 24 hours. Nalr mutants per culture were enumerated on LB agar containing 20 µg/ml of nalidixic acid. Mutation rates and 95% confidence limits were calculated with the ft computer program (Shaver and Sniegowski 2003) which generates a maximum likelihood estimation of mutation rate using the entire distribution of mutant numbers per culture.

2.2.3. Competitions under soft selection

To assess the possibility of double-mutator hitchhiking under soft selection, I carried out competitions in a relatively benign environment. I established 25 populations combining the isogenic single- and double-mutator strains in Davis minimal broth (Carlton and

Brown 1981) supplemented with 25 mg l-1 of glucose (hereafter ―DM25‖) and 15 similar populations in Davis minimal broth supplemented with 1 g l-1 of glucose (hereafter

―DM1000‖). The effective sizes of the populations propagated in DM25 and DM1000 were ~1.7x107 and ~6.6x108. In DM25, I carried out five replicate competitions at each

10 of the following frequencies: high frequency (~0.5), high intermediate frequency (~0.3), and low intermediate frequency (~0.1); I also carried out ten replicate competitions with the double-mutator starting at low frequency (~0.03). In DM1000, I carried out five replicate competitions each with the double-mutator starting at high (~0.5), high intermediate (~0.3) and low intermediate (~0.1) frequency; I did not carry out competitions in DM1000 at low double-mutator starting frequency because the results of competitions in DM1000 at the higher frequencies indicated that such competitions would be uninformative. Competitions were initiated by aliquoting appropriate numbers of cells from a single pair of parent cultures. All populations were maintained in 20 ml culture tubes at 37°C with shaking at 120 rpm: every day, 100 µl from each overnight culture was transferred to 9.9 ml of fresh medium. During the competitions, samples from each population were diluted appropriately and plated on permissive LB agar

(Miller 1992). After 24 h of growth, permissive plates were replica plated to LB agar supplemented with 15 µg/ml of tetracycline to estimate double-mutator frequency.

2.2.4. Competitions under hard selection

To assess the possibility of double-mutator hitchhiking under hard selection, I carried out competitions in an environment that imposed a series of novel lethal selective events. I established 20 populations combining the isogenic single- and double-mutator strains in

DM1000 as described above. I carried out five replicate competitions with the double- mutator starting at an intermediate frequency (~0.5), and fifteen replicate competitions with the double-mutator starting at low frequency (~0.03-0.1). All populations were propagated in 20 ml culture tubes at 37°C with shaking at 120 rpm: every day for seven 11 days, 1 ml from each overnight culture was transferred to 9 ml of fresh medium. To impose lethal selection, I supplemented the DM1000 with a series of single antibiotics as follows: 100 µg/ml of rifampicin on days 2-3, 100 µg/ml of streptomycin on days 4-5, and 100 µg/ml of novobiocin on days 6-7. Estimation of double-mutator frequency was carried out by replica plating.

2.3. Results and Discussion

2.3.1. Competitions under soft selection

Figure 2-2 (blue lines) shows the results of competitions under soft selection in DM25.

These experiments produced outcomes consistent with the results of early work competing single-mutator and wild-type E. coli strains (Chao and Cox 1983). When the double-mutator began at low intermediate frequency or higher, it always rose toward fixation. However, when the double-mutator began at low frequency it was always lost

(data not shown). Double-mutators were detectable at the outset of the low-frequency

DM25 experiments at frequencies of ~3%; however, they fell to undetectable frequencies within a few tens of generations. The success of the double-mutator only when initially present at intermediate or higher frequency is best explained as a consequence of hitchhiking with beneficial mutations. The alternative interpretation—that the dnaQ905 allele or its associated flanking DNA confers a direct fitness advantage on the double- mutator strain—would predict that the double-mutator should also prevail from a low starting frequency.

12 Figure 2.2 (red lines) shows the results of competitions under soft selection at higher effective population size, in DM1000. Here, the double-mutator was lost in all competitions except at the highest starting frequency (~0.5), where some double-mutator subpopulations increased in frequency after a lag. A large effective population size increases the likelihood that single- and double-mutator subpopulations will acquire beneficial mutations contemporaneously, because the beneficial mutation supply rate of each subpopulation (the product of its beneficial mutation rate and population size) is high. Under such a circumstance, a double-mutator would have a net advantage in acquiring beneficial mutations only when at a high starting frequency, as observed in our experiments. Indeed, the dynamics of double-mutators in the high-frequency DM1000 populations are quite variable, consistent with a shifting selective advantage caused by beneficial mutations arising on both backgrounds. Similar dependence of mutator hitchhiking on the beneficial mutation supply rates of competing clonal genotypes has been observed previously (de Visser and Rozen 2005).

1.00

0.80

mutator -

0.60

0.40 Frequency of double 0.20

0.00 0 50 100 150 200 250 300 350 400 450 Generation

Figure 2-2: Competitions under soft selection. Blue lines: populations propagated in

DM25; red lines: populations propagated in DM1000. Ten populations in DM25 in which the double-mutator began at low frequency (~0.03) and was lost within 70 generations are not shown.

14 To address directly the possibility that intrinsic fitness differences, rather than mutator hitchhiking, could explain the results of the soft-selection experiments, we assayed maximal growth rates of the single- and double-mutator strains in DM1000. Doubling times for the two strains were closely similar (single-mutator: 1.70 ± 0.05 h; double- mutator: 1.72 ± 0.03 h) and statistically indistinguishable (two-tailed P = 0.90) and thus provided no evidence of intrinsic fitness differences sufficient to explain the dynamics observed in the soft-selection experiments.

2.3.2. Competitions under hard selection

Figure 2-3 shows the results of competitions under hard selection in which populations were exposed to a series of lethal selective events; this situation might arise, for example, in a pathogen population faced with host immune surveillance. Here, the double-mutator strain rose to apparent fixation in every experiment in which it began at intermediate frequency (~0.5). Similarly, in 7 of 15 competitions in which the double-mutator began at a lower frequency (0.03-0.1), its frequency increased substantially over the course of the experiment. However, in the remaining eight competitions in which the double- mutator began at the lower frequency the single-mutator prevailed instead. The mixed success of the double-mutator at the lower starting frequency is consistent with roughly equivalent mutation supply rates in the single- and double-mutator subpopulations: the double-mutator genotype has a 45-fold greater mutation rate than the single-mutator, but its initial frequency in these experiments was about 50-fold lower than that of the single- mutator. Consequently, mutations conferring resistance to a particular antibiotic were about equally likely to arise in the single- and double-mutator subpopulations. Overall, 15 the results of the hard selection experiments are similar to those of experiments in which large cultures of E. coli were shown to be enriched for single-mutator frequency after repeated exposure to antibiotics (Mao et al. 1997).

Figure 2-3: Competitions under hard selection. Grey boxes indicate periods of antibiotic exposure. The number of generations reported in the figure is a minimal estimate because of sharp reductions in population size and subsequent regrowth upon exposure to antibiotics.

17 2.3.3. Discussion

My results demonstrate that there are circumstances under which recurrent mutator hitchhiking can occur in an asexual population as predicted by recent theory

(Andre and Godelle 2005; Gerrish et al. 2007). An obvious limitation of our approach is that I seeded the double-mutator genotype into populations at a higher frequency than would have applied had it arisen spontaneously, as was done in previous work with single-mutators (Chao and Cox 1983). This raises the question of how a genotype with a higher mutation rate would attain the frequency necessary for it to hitchhike in a natural population. For single-mutators, this question was answered by the observation that spontaneously arisen mutators can, over time, acquire chance associations with beneficial mutations and hitchhike (Taddei et al. 1997; Shaver et al. 2002). Similar hitchhiking of spontaneously arisen double- and higher-order mutator genotypes has been predicted theoretically and observed in simulations of asexual populations (Andre and Godelle

2005; Gerrish et al. 2007), but the phenomenon has not been confirmed experimentally and remains an important subject for future research.

The observation that double-mutators can prevail over single-mutators is perhaps not surprising because a higher mutation rate confers a per capita advantage in acquiring beneficial mutations. However, an extremely high mutation rate can cause extinction of even the largest asexual population, whether through erosion of genomic information

(Eigen and Schuster 1977) or excessive deleterious mutational load (Bull et al. 2007; Bull and Wilke 2008). Our double-mutator strain has a high estimated genomic deleterious mutation rate of Ud = 0.9 (see Ch. III) which exceeds the value of Ud = 0.69 theoretically 18 predicted to cause extinction of an asexual bacterial population in the absence of beneficial mutations {Bull, 2008 #292}. Chapter III presents experimental results that address whether a very large population that has fixed the double-mutator genotype is destined for extinction in the relatively short term.

19 Chapter 3. Declining fitness in a higher- mutating strain of Escherichia coli

3.1. Introduction

Mutation is the ultimate source of the genetic variation upon which selection acts and the genomic mutation rate is, itself, a quantitative, polygenic trait that is potentially under some form of selection (see Sniegowski et al. 2000). Numerous loci in many organisms are known that effect genomic and locus-specific mutation rates (e.g. Friedberg et al.

1995). Mutators (alleles that increase the rate of mutation) have been found at high frequency in virus populations (Suarez et al. 1992; Chikova and Schaaper 2006), mammalian cancer cell populations (Loeb 1991; Tucker 1997; Stoler et al. 1999; Loeb

2001), natural microbial populations (Treffers et al. 1954; Miyake 1960; Kimura 1967;

Liberfarb and Bryson 1970; Boe et al. 2000; Loeb 2001), and experimental microbial populations (Cox 1976; Fijalkowska 1993; Miller et al. 2002; Notley-McRobb et al.

2002; Shaver and Sniegowski 2003). Although mutation is the ultimate source of all beneficial mutations, it simultaneously serves to erode fitness through the generation of deleterious mutations; if we consider that the majority of phenotype-affecting mutations are deleterious (Fisher 1930; Kneightley and Eyre-Walker 1999; Lynch et al. 1999), the abundance of mutators may seem paradoxical. Indeed, given the expected preponderance of deleterious mutations, A. H. Sturtevant (1937) asked: ―Why does the mutation rate not become reduced to zero?‖ The factors that influence the evolution of mutation rate in 20 both sexual and asexual populations have been well-reviewed (see Sniegowski et al.

2000; Baer et al. 2007; Lynch 2008). There are two factors worth noting that theoretically establish a minimal limit to mutation rate. First, physiological limitations of

DNA repair and replication probably prevent reduction of the genomic mutation rate to zero (Drake 1990, 1993; Kondrashov 1995); it could be energetically costly or physicochemically impossible to prevent all replication errors. A second factor that might prevent complete reduction of the mutation rate is the necessity of variation; selection requires variation to act upon, and a mutation rate of zero would render adaptation impossible for lack of novel genetic variation. Some combination of these two factors, and perhaps others as well (Andre and Godelle 2005; Lynch 2008; Ancliff and

Park 2009), may be expected to establish a theoretical minimum mutation rate.

As mentioned above, it is not merely the generation of deleterious mutations that influences the evolution of mutation rate; the production of beneficial mutations is integral as well. Fisher (1930) first proposed that indirect selection might favor an equilibrium mutation rate which optimized the balance of deleterious and beneficial mutations in an asexual population. Kimura refined this optimum theory, calculating the mutation rate predicted to maximize the long-term fitness of an asexual population

(Kimura 1967). Multiple models have since been proposed to further develop the theory that mutation rates evolve to an optimum in asexual populations (Kimura 1967; Leigh

1970; Holsinger and Feldman 1983; Ishii et al. 1989; Charlesworth 1990; Orr 2000;

Nilsson and Snoad 2002; Clune et al. 2008). Experimental evidence, however, suggests that mutator hitchhiking (see chapter II) in an asexual population results in relatively 21 large increases in the mutation rate as opposed to small alterations that fine-tune the rate to an optimal level (Chao and Cox 1983; Mao et al. 1997; Sniegowski et al. 1997; Wylie et al. 2009; Gentile et al. 2011). Concurrent with experimental results, recent theory suggests that the evolution of mutation rate in an adapting asexual population may actually be upwardly biased (Andre and Godelle 2005; Gerrish et al. 2007).

Most ‗optimal‘ mutation rate models predict that the strength of purifying selection will be sufficient to eliminate new deleterious mutations from the population, thus preserving the fitness and genomic integrity of the population. This prediction is partly based on the assumption that the mutational load, the fitness cost of deleterious mutation in a population (see Discussion), increases immediately with any increase in the mutation rate. Johnson (1999) noted, however, that there is a delay between an increase in mutation rate and the accrual of increased genetic load, hence the cost associated with an increased mutation rate is not immediately paid. Both simulations and analytical approaches suggest that the delay in accruing genetic load may potentially allow a succession of mutator hitchhiking events to occur, thus yielding an ever-increasing mutation rate in an asexual population (Andre and Godelle 2006; Gerrish et al. 2007).

Although the aforementioned simulations, experiments, and theoretical results suggest that natural selection may act to continually increase mutation rate in an asexual population, the short- and long-term consequences of bearing a very high mutation rate are poorly understood and have received limited experimental investigation. Intuitively, the cost of increasing the mutation rate beyond the power of purifying selection seems

22 clear: fitness should ultimately decline, perhaps to a level sufficient to cause extinction.

If purifying selection is unable to remove deleterious mutations from the population we might predict the steady accrual of deleterious mutations in the genome, the cost of which can be observed as a progressive reduction of fitness. Various theories have been proposed that may apply to the dynamics of a population at such an overwhelmingly high mutation rate (e.g. Bull, 2008; Eigen, 1979). Here, I experimentally investigate the consequences of a very high mutation rate in asexual populations through the long-term propagation of populations of Escherichia coli simultaneously bearing defects in both

DNA mismatch repair and proofreading (Gentile et al. 2011).

3.2. Methods and Materials

3.2.1. Bacterial strains

An asexual (F-) strain of E. coli bearing the mutL13 allele (see Table 3-1), which confers mismatch repair deficiency, was obtained from the E. coli Genetic Stock Center at Yale

University. The dnaQ905 allele, which confers DNA proofreading deficiency, was cotransduced with a tetracycline resistance determinant from donor strain CSH116

23 Strain Designation Genotype Reference

CSH116 - F-, ara-600, yafC502::Tn10, dnaQ905, Δ(gpt- Miller 1992

lac)5, λ-, relA1?,spoT1?, thi-1

ES568 "single- F-, fhuA2, lacY1, tsx-1 or tsx- Liberfarb and

mutator" 70, glnV44(AS), gal-6, λ-, xyl- Bryson 1970

7,mtlA2, mutL13

PS2534 "double- F-, fhuA2, lacY1, tsx-1 or tsx- this work

mutator" 70, glnV44(AS), gal-6, λ-, xyl-

7,mtlA2, mutL13, yafC502::Tn10, dnaQ905

Table 3-1: A list of strains used in this work and relevant genotype information.

24 3.2.2. Long-term propagation of populations

I established six populations of the double-mutator strain and six populations of the single-mutator strain in 2.5 ml of Davis minimal broth (Carlton and Brown 1981) supplemented with 1 g/l of glucose (hereafter, DM1000). Double-mutator populations were grown in a medium supplemented with 15 µg/ml of tetracycline as a guard against cross-contamination. Populations were established by aliquoting samples of ~2500 cells from a single parent culture. All populations were maintained in 13 ml culture tubes at

37°C with shaking at 120 rpm: every 24 hours, 100 µl from each overnight culture was diluted 1:100,000 and 250 µl was transferred into 2.25 ml of fresh DM1000. The dilution

6 factor permitted 20 generations of binary fission per day (log2 10 = 20), thus the effective population size was approximately ~5x104 (~2,500 individuals per bottleneck *

20 generations between bottlenecks) (Wahl and Gerrish 2001). Populations were propagated for 125 days (~2500 generations) and samples of each population were archived at -80°C every ~140 generations in 15% glycerol. In instances when a population did not grow (as judged by visual inspection) within a 24 hour period, 100 µl of inoculum from the previous day‘s growth was used to reestablish this population and continue the propagation.

3.2.3. Fitness assay

I characterized the evolution of fitness in all 12 experimental populations using maximal growth rate as a proxy for absolute fitness (see Discussion). Frozen archives were partially thawed, and 100 µl (representing standing genetic variation in each population)

25 was sampled and inoculated in 2.5 ml of DM1000 and grown overnight at 37°C with shaking at 120 rpm. 100µl of the overnight culture was inoculated into 2.4 ml of fresh

DM1000 in a 13 mm tube identical to those used in the course of the propagation. Each tube was incubated at 37°C with shaking at 120 rpm and read periodically during the log growth phase for absorbance at 600 nm in a Spectronic 20D+ spectrophotometer. I assayed five replicates of each population from every archived time point. The growth rate (x = doubling time in hours) was calculated using the formula:

, where A0 is the absorbance at time 0, At is the absorbance at time t, and t is the time elapsed in hours. Analysis of the initial growth rate assays suggested that absorbance measurements between 0.05 and 0.4 were typically representative of the log growth phase; therefore, the time it took for absorbance to increase from ~0.05 to ~0.4 was used as t in the formula above.

3.2.4. Temperature sensitivity assay

I assayed the ancestral and evolved single- and double-mutator populations for growth sensitivity to high ambient temperatures. As above, frozen archives of the single- and double-mutator ancestors and each archived population at the final time point were partially thawed, and 100 µl (representing standing genetic variation in each population) was sampled and inoculated in 2.5 ml of DM1000 and grown overnight. A single

1:100,000 dilution was then prepared for each sampled time point and 50 µl was inoculated into 200 µl of DM1000 in the wells of a 96-well microplate. The microplate was then incubated and read periodically for absorbance at 600 nm for 24 h with shaking

26 at both 44°C and 44.5°C in a ThermoScientific Multiscan GO microplate spectrophotometer. Pilot experiments suggested a divergence between the single- and double-mutator populations in the evolution of sensitivity to these temperatures.

Although the abiotic environment in this assay differed from that of the long-term propagation, the 96 well microplate allowed greater replication and temperature control.

It is conceivable that the smaller DM1000 volume may have had a minor effect on the growth rate, therefore the populations were qualitatively, not quantitatively, scored as either sensitive to elevated temperature (no growth after 24 h) or tolerant (some growth after 24 h) by inspection of growth profiles.

3.2.5. Competitive fitness assay

To assess the fitness of the evolved double-mutator populations relative to their ancestors, I carried out competitions in environments similar to that in which the populations had evolved. I established six populations combining the ancestral single- mutator strain and the evolved double-mutator strain archived at the final time point (one competition for each evolved population) in DM1000. Additionally, I established six identical populations in Davis minimal broth supplemented with 25 mg/l of glucose

(hereafter ―DM25‖). The effective sizes of the populations propagated in DM25 and

DM1000 were estimated to be ~1.7x107 and ~6.6x108; these population sizes are identical to those previously described in chapter II. In all competitions, the evolved double-mutator started at a frequency between ~0.3 and ~0.5. Competitions were initiated by aliquoting appropriate numbers of cells from a single pair of parent cultures.

All populations were maintained in 20 ml culture tubes at 37°C with shaking at 120 rpm: 27 every day, 100 µl from each overnight culture was transferred to 9.9 ml of fresh medium.

During the competitions, samples from each population were diluted appropriately and plated on permissive LB agar (Miller 1992). After 24 h of growth, these permissive plates were replica plated to LB agar supplemented with 15 µg/ml of tetracycline to assay double-mutator frequency.

3.2.6. Fitness variance assay

The distribution of absolute fitness amongst individuals, relative to genomic mutation rate, in a population is likely to determine whether fitness increases, is stable, or declines

(see Discussion). Frozen archives from each time point from one single- and one double- mutator population were partially thawed, and 100 µl (representing standing genetic variation in each population) was sampled and inoculated in 2.5 ml of medium and grown overnight at 37°C with shaking at 120 rpm. 100 µl of the overnight culture was diluted appropriately, spread on a permissive LB agar plate, and incubated overnight at 37°C in order to isolate individual clones. 30 colonies (representing individual clones) from each population were picked at random from the LB agar plates, inoculated into 2 ml of

DM1000, and incubated overnight at 37°C with shaking at 120 rpm. The growth rate of each clone was assayed as previously described. The assay was replicated five times for five of the clones from each population in order to estimate the experimental variance of the growth rate estimates. Maximum fitness, was calculated in the following manner: of the 30 growth rate measurements at a given time point, 5 measurements were chosen at random and the maximum of those 5 were recorded. This process was repeated

28 5000 times to get a sample of 5000 maxima from samples of size 5. The mean of the maxima was recorded. The process was then repeated for samples of all sizes between 6 and 29. We then calculated a regression of mean maxima on log sample size. was then calculated by extrapolating mean maximum value from an assumed sample size of

50,000. The average fitness of individuals in the population, , is simply the mean of the

30 growth rate measurements.

3.2.7. Extinction risk assay

Frozen archives of each population at the final time point were partially thawed, and 100

µl (representing standing genetic variation in each population) was sampled and inoculated in 2.5 ml of DM1000 and grown overnight at 37°C with shaking at 120 rpm.

After 24 hours of incubation, the overnight culture was diluted and inoculated into three tubes of 2.25 ml of DM1000 and grown overnight again. After another 24 hours, 100 µl from each overnight culture was diluted 1:100,000 fold and 250 µl was transferred into

2.25 ml of fresh DM1000. This procedure was identical to the daily transfer protocol in the long-term propagation. 50 replicates of this transfer and inoculation were conducted for each population. The replicated inoculations were incubated overnight at 37°C with shaking at 120 rpm. The cultures were qualitatively scored as either transfer failures (no visible turbidity after 24 h) or transfer successes (visible turbidity after 24 h) by visual inspection. The results were tabulated in a 2x2 contingency table and a two-tailed

Fisher‘s exact test was used to test for a significant difference between the transfer failure rates of the single- and double-mutators.

29 3.2.8. Mutation rate assay

I performed fluctuation tests (Luria and Delbruck 1943) on individual random clones from single- and double-mutator populations at the initial and final time point to estimate the rate of mutation to nalidixic acid. We have previously shown the mutation rate to nalidixic acid is a reliable proxy for the genomic mutation rate (see Chapter II).

Strains were inoculated from frozen stock, grown, and transferred daily for two days in

DM1000. For each fluctuation test, 25 cultures were grown from inocula of ~500 cells.

Cultures were grown to stationary phase before selective plating. We estimated final population sizes by sampling and plating an appropriate dilution from five cultures on permissive LB plates and counting the number of colonies after 24 hours. Nalr mutants per culture were enumerated on LB agar containing 20 µg/ml of nalidixic acid. Mutation rates and 95% confidence limits were calculated with the ft computer program (Shaver and Sniegowski 2003) which generates a maximum likelihood estimate of mutation rate using the entire distribution of mutant numbers per culture.

3.2.9. Sequencing assay

I sequenced the dnaQ905 allele from the evolved double-mutator populations in order to confirm that the double-mutator still carried the second mutator allele at the final time point and to ensure that the double-mutator populations were not contaminated by any other strain of E. coli. A random clone was picked from each population of double- mutator and genomic DNA was prepared for sequencing. The dnaQ locus was amplified using flanking primers dnaQF (5‘ATGAGCACTGCAATTACACGC3‘) and dnaQR

30 (5‘TTTAGCGCCTTTCACAGGTAT3‘) and the Bio-Rad High-Fidelity DNA polymerase (Bio-Rad cat#172-5302). Sequencing was performed with the forward primer dnaQampF (5‘AGTCTGACATAAATGACCGCT3‘).

3.3. Results

3.3.1. Fitness decreased in double-mutator populations

Figure 3-1 plots the inverse of doubling time (a good proxy for absolute fitness) of each population over the course of propagation. The fitness of each population was measured with five-fold replication at every archived time point. All six of the single- mutator populations (blue lines) increased in fitness over 2500 generations. Independent two-tailed two-sample t-tests (8 d.f. per test) comparing initial and final fitness were conducted for each single-mutator population and revealed that the fitness increase was significant (p< .05) in every population. These results are unsurprising and consistent with previous work reporting increased fitness in evolution experiments with mutators of low or intermediate strength (Lenski et al. 1991; Shaver et al. 2002; Gentile et al. 2011).

Five of the double-mutator populations (red lines) peaked in fitness within the first 1000 generations. Independent two-tailed two-sample t-tests (8 df) comparing initial and peak fitness were conducted for each double-mutator population and revealed that the fitness increase was significant (p< .05) in two of the five populations. Interestingly, the sixth double-mutator decreased in fitness significantly within the first 600 generations (p

<.001) of its founding. Among the five double-mutator populations that initially rose in fitness, all suffered from declining fitness over the remainder of the propagation. 31 Independent two-sample t-tests (8 d.f.) comparing initial and final fitness were conducted for each double-mutator population and revealed that this fitness decrease was significant

(p< .05) in five of the six populations; the sixth double-mutator population had no significant change in fitness (p=.13) although its final fitness was significantly lower than its peak fitness (two-tailed, two sample, independent t test, 8 df, p< .001) and showed a strong downward trend over the final ~1200 generations.

32 0.8

0.75

0.7

0.65

0.6

0.55

Inverse Inverse Doublingof Time 0.5

0.45

0.4 0 500 1000 1500 2000 2500 3000 Generations

Figure 3-1: Evolution of fitness (as measured by the inverse of doubling time). The doubling time of each population was measured at each archived time point. The inverse of the average doubling time of 5 replicates is plotted. Blue lines: single-mutator populations; red lines: double-mutator populations.

33 3.3.2. The risk of extinction increased in double-mutator populations

During the course of the propagation experiment, there were occasions when a population failed to grow. In such situations, populations were restarted with undiluted inoculum from the last successful day of growth. Proximate causes of culture growth failure could have been either experimental error or declining culture density, a symptom of reduced fitness, in the experimental populations. If experimental error was the true cause, I might predict that the single- and double-mutator populations had an equal likelihood of transfer failure. To distinguish between these alternatives, I assayed every single- and double- mutator population at the initial and final time point for the risk of extinction, which we define as a failure to transfer. Each population was assayed with 50-fold replication. The results from each evolved genotype (i.e. all six evolved single-mutator populations) were pooled separately, thus yielding 300-fold replication (6 populations * 50 replicates/population) for both evolved genotypes. The results are shown in table 3-2. A two-tailed Fisher‘s exact test comparing the extinction risk between the evolved single- and double-mutator populations revealed that the double-mutator populations evolved a significantly greater risk of extinction (p=.001) over the course of the experiment.

Genotype

single mutator double-mutator

success 298 284 Transfer failure 2 16

Table 3-2: Results of extinction risk assay. Rows report the transfer outcome, Columns report the genotypes

35 3.3.3. Competitive fitness decreased in double-mutator populations

The ancestral single- and double-mutator populations are nearly identical in fitness as measured by growth rate (see Chapter II). I previously reported that the ancestral double- mutator will successfully hitchhike to fixation over the ancestral single-mutator when introduced at a sufficient frequency in DM25 (Gentile et al. 2011). If the double-mutator decreased in fitness during propagation in the current experiment, I might predict that it would no longer successfully hitchhike when introduced at a similar frequency. To test this, I introduced the evolved double-mutator populations into populations of single- mutators at a frequency of ~.5 and tracked the frequency of the double-mutator until it was lost or reached fixation. The results are shown in Figure 3-2. In contrast with the fate of the ancestral double-mutator, the evolved double-mutator genotypes were rapidly lost in competition with the ancestral single-mutator. This outcome suggests that double- mutator populations declined in fitness over the course of the propagation and is qualitatively consistent with the reduced absolute fitness reported above.

I can crudely estimate the strength of selection acting on the genetic background of the mutator genotype from the change in mutator frequency during the course of the propagation (Crow and Kimura 1970). Using the competition results, I roughly estimated the average selection coefficient of the evolved double-mutator genotype using the asexual, haploid selection model:

36 where s is the selection coefficient, t is time in generations, and p is the frequency of the gene): -.030 in both DM25 and DM1000. The competitive fitness assay reported here is analogous to competition experiments reported in chapter II; in those experiments, I carried out competitions between the ancestral single- and double-mutator genotypes with approximately equal starting frequencies. From these ancestral competitions, I estimated the selection coefficient of the genetic background of the ancestral double-mutator genotype in the same media and competing against the same strain; the average selection coefficient was .014 in DM25 and ~0 in DM1000. It is important to note that the estimated selection coefficients estimated only offer a rough comparison because they are based on deviations from the initial double-mutator frequency and not on the dynamics of the actual fitness-affecting mutations that arose in the background of the single- and double-mutator subpopulations. Nevertheless, the reduced selection coefficient of the evolved double-mutator genetic background suggests that selection was acting against the genomic mutations that had been acquired by the double-mutator populations by the end of propagation.

0.60

mutator - 0.40

0.20 Frequency of double

0.00 0 50 100 150 200 250 Generation

Figure 3-2: Competitions between evolved double-mutator and ancestral single-mutator genotypes. Solid lines: populations propagated in DM25; dashed lines: populations propagated in DM1000.

38 3.3.4. Double-mutator populations evolved an increased sensitivity to high ambient temperature

Previous work suggests that the deleterious effects of high mutation rate could increase a population‘s sensitivity to temperature (Zeldovich et al. 2007; Chen and Shakhnovich

2009). Although the single- and double-mutator populations were propagated continuously in a stable, 37° environment in this experiment, I was interested in assaying whether their ability to grow at higher temperatures had evolved as a byproduct over the course of the propagation. To test this, I assayed populations for growth at 44° and 44.5°, temperatures above what they had experienced during propagation. Interestingly, while the ancestors of both the single- and double-mutator populations were capable of growing at 44°, the evolved double-mutator populations were not (figure 3-3). In contrast, the evolved single-mutator populations were able to grow at 44° and also at the higher temperature of 44.5°, a temperature that the ancestral single-mutator population could not tolerate. I discuss some possible implications of these results below.

Figure 3-3: Results of temperature sensitivity assay. – denotes no visible growth and + denotes visible growth after 24 hours.

40 3.3.5. Double-mutator populations maintained their mutator genotype and phenotype

Single random clones taken from each of the six double-mutator populations were assayed for mutation rate and mutator genotype. Every double-mutator clone assayed for mutation rate was found to have a mutation rate statistically indistinguishable from that of the ancestral double-mutator population. Likewise, all six single-mutator clones assayed for mutation rate were found to have the same mutation rate as the ancestral single-mutator population. Although I only assayed individual clones from each population, the results are consistent with the maintenance of the single- and double- mutator genotypes at high frequencies in these populations. In addition, sequencing of one isolate from each double-mutator population revealed that, as expected, they all still carried the dnaQ905 allele, indicating that there was no reversion of the second mutator allele and no contamination. Subsequently, whole-genomic sequencing of clones from several evolved single- and double-mutator populations also shows that neither mutL nor dnaQ have any SNPs relative to their respective ancestors, further confirming that there was no reversion of either mutator allele.

3.3.6. Fitness variance assay

Recent theory suggests that the relationship between fitness variance and mutation rate allows quantitative predictions above the fitness trajectory of a population. Thirty random clones taken from each archived time point of a random, representative single- mutator and double-mutator population were assayed for fitness (the inverse of doubling

41 time). Five of the clones were assayed with five-fold replication to estimate the experimental variance; which was then subtracted from the the average fitness variance.

The results of the fitness variance assay are shown in figure 3-4. The implications of the observed fitness variance are discussed below.

Figure 3-4: The variance in fitness (inverse of doubling time) assayed over time in a random, representative single-mutator (top) and double-mutator (bottom) population plotted over the course of propagation. The predicted variance threshold ( ) and error threshold ( ) for each population are also plotted. The theory of error catastrophe predicts that in order to maintain fitness the fitness variance must be above and

must be above

43 3.4. Discussion

I established independent populations of E. coli bearing either a single- or double-mutator genotype and propagated them for several thousand generations. At the conclusion of the experiment, the fitness of the single-mutator populations had risen substantially, consistent with the substitution of beneficial mutations; in contrast, the fitness of the double-mutator populations had fallen significantly, consistent with erosion of fitness by deleterious mutations. I discuss the results here in light of previous experimental theoretical work regarding high-mutation rate evolution and address theories regarding mutation-driven fitness decline in asexual populations.

3.4.1. Limitations of previous work on mutation-driven fitness decline

Some previous experiments have employed chemical mutagens to induce a high mutation rate in viral populations (Schaaper 1998; Loeb et al. 1999; Crotty et al. 2002; Negishi et al. 2002; Graci et al. 2007; Graci et al. 2008; Perales et al. 2009); unfortunately, the results are inconclusive in addressing whether a population can maintain, and perhaps benefit from in the short-term, a high mutation rate while ultimately declining in fitness.

Extinction of viral populations has been attributed solely to lethal chemical mutagenesis despite evidence that chemical mutagens may have directly inhibitory effects on viral replication (Bull et al. 2007). Several efforts at lethal chemical mutagenesis in virus populations have actually failed when the populations evolved resistance to chemical mutagens (Vignuzzi et al. 2005a; Springman et al. 2010). Previous work in E. coli suggests that deoxyribosyldihydropyrimido[4,5-c][1,2]oxazin-7-one (dP), a potent base 44 analogue shown to increase genomic mutation rate and decrease cell viability, could be capable of inducing a lethal rate of mutation (Negishi et al. 2002). Work in our lab however, has shown that the single- and double-mutator E.coli can evolve resistance to dP and persist for at least several hundred generations in the presence of the mutagen (D.

Balk et al in prep). Although there is potential for resistance to evolve, previous work has reported successful mutagen-induced extinction (Negishi et al. 1997; Graci et al.

2007; Graci et al. 2008); interestingly, the cause of extinction has often been attributed solely to an increased mutation rate, whereas the perceived contribution of potential direct inhibitory effects have been marginalized. Unfortunately, quantitative analysis of theories of mutation-driven fitness decline and extinction is rendered more difficult (or perhaps impossible) if both mutagenic and inhibitory effects contributed to the previously reported viral extinctions.

Similarly to the methods utilized in this work, some previous experiments have attempted to generate a high mutation rate by genetically constructing mutators with one or more mutator alleles (Fijalkowska and Schaaper 1996; Negishi et al. 2002; Herr et al.

2011). The immediately lethality of very high-mutation rate genotypes has been assumed to be a byproduct of the high predicted mutation rate. While this assumption is certainly plausible, the inviability of these theoretical mutator strains renders quantitative analysis of mutation-driven fitness decline impossible. In contrast, the results presented here allow a useful analysis of populations that have a very high mutation rate precisely because the double-mutator genotype is viable under certain conditions.

45 3.4.2. Fitness diverged and ultimately declined in double-mutator populations

It is immediately apparent that fitness decreases in five of the six double-mutator populations and increases in all six single-mutator populations (fig. 2). In populations propagated by batch culture, absolute fitness is really the sum of at least three components: speed of recovery out of stationary phase, maximal growth rate, and survival in stationary phase. Here, I use maximal growth rate as a proxy for absolute fitness because the time in stationary phase is minimized by the experimental design. In theory, significant reductions in growth rate should result in an increased risk of failure to recoup the daily dilution, thereby increasing the probability that a culture would fail to transfer successfully. This risk is discussed in detail below.

Interestingly, the fitness trajectories seemed to differ amongst the double-mutator populations; the populations attained peak fitness at different time points. This is unsurprising, given that divergence in clonal populations adapting to identical environments has been reported in microbial (Lenski et al. 1991; Travisano et al. 1995;

Woods et al. 2006), phage (Bull and Molineux 1992; Yin 1993), Drosophila melanogaster (Cohan and Hoffmann 1986), and cancer cell populations (Tsao et al.

1999). While it is unclear what contributed to the divergence in double-mutator fitness, we may consider several possibilities. The fitness landscape could have played a significant role in adaptation and fitness divergence (Tenaillon et al. 1999). If the populations existed on a relatively rugged fitness landscape, then the high mutation rate

46 of the double-mutator may have allowed the populations to rapidly explore a wider range of the fitness landscape (Desai et al. 2007; Weissman et al. 2009) than the single-mutator.

The order in which mutations, both beneficial and deleterious, arise is also an important factor in determining the overall fitness trajectory of a population (Mani and Clarke

1990). The order of mutation has been shown to potentially impose constraints on the future evolutionary potential of experimental microbial populations (Blount et al. 2008;

Woods et al. 2011). Given the high mutation rate of the double-mutator population, it is also conceivable that deleterious mutations of differing effect hitchhiked to fixation with the first beneficial mutations to fix in each population. Thus, the fitness of the populations could have differed greatly even if identical beneficial mutations had been fixed (Ferenci 2008).

The probable key to overall fitness decline observed in the double-mutator populations is the relative rates, magnitudes, and interactions between beneficial and deleterious mutations over the course of propagation. In an experiment of this kind, the supply of highly beneficial mutations is expected to be highest at the onset of the experiment and to decline as large-effect mutations are discovered and fixed (de Visser et al. 1999). As noted above, two of the six double-mutator populations increased significantly in fitness over the first several hundred generations, confirming the availability of large effect beneficial mutations at the onset of propagation. (In addition, see chapter II.) Fitness increases in the four remaining double-mutator populations could have been limited by deleterious mutations that hitchhiked with the earliest beneficial mutations that fixed. In contrast to the dwindling supply of beneficial mutations, the supply of deleterious 47 mutations is likely to have remained high over course of propagation; the combined effect of a steady accrual of deleterious mutations and a decreasing supply of beneficial mutations could have retarded adaptation and contributed to the fitness decline of double- mutator populations. Whole-genome sequencing of the double-mutator populations might reveal the prevalence of potentially beneficial and deleterious nonsynonomous fixed mutations and polymorphisms. We have sequenced the ancestral and three evolved single- and double-mutator populations as well as clones isolated from each population using SOLiD sequencing with 300X coverage. Currently, preliminary analysis of clonal isolates suggests that evolved double-mutator populations have more SNPs than the single-mutators; the reported number of SNPs is likely be greater once the data from the populations are analyzed.

In addition to estimating the absolute fitness of each population, I used direct competitions between strains to generate a measure of competitive fitness that incorporates all aspects of growth. Competitions between the evolved double-mutator and the ancestral single-mutator populations indicate that competitive fitness does indeed decline in all the double-mutator populations (Fig. 3). These results are in agreement with the growth rate measurements and imply that the double-mutator populations actually decreased in competitive fitness. These competitions were necessarily done in the absence of tetracycline. Efforts are currently underway to measure the fitness of the evolved double-mutator populations by competing the evolved and ancestral double- mutator populations in an environment with tetracycline. It should be noted, however,

48 that all of the results presented here indicate that the fitness of the double-mutator populations does indeed ultimately decline over time.

An interesting outcome in the present experiment was the apparent evolution of increased temperature sensitivity in the double-mutator, but not the single-mutator, populations

(Fig. 3). Zeldovich et al. (2007) proposed a model of lethal mutagenesis suggesting that erosion of protein stability, via the accumulation of mutations, would eventually lead to the loss of essential cellular functions and protein aggregation. Subsequently, genotypes with a high mutation rate tend to have lower protein stability than genotypes with a low mutation rate (Chen and Shakhnovich 2009). It is possible that the double-mutator populations evolved greater temperature sensitivity than their ancestor as the byproduct of an increased genetic load; thus, temperature sensitivity may be symptomatic of an overall decline in fitness. Interestingly, the single-mutator populations evolved resistance to a temperature higher than the single-mutator ancestor could tolerate, 44.5°C. This surprising result might indicate that overall protein stability, and perhaps fitness, increased in the single-mutator populations. If protein stability is indeed eroding in the double-mutator populations, then we may predict that they would be less tolerant of colder temperatures as well. These results are very preliminary, but they suggest further intriguing avenues of research with these populations.

49 3.4.3. The mutation rate did not decrease in the double-mutator populations

It has long been theorized that asexual populations should evolve to an optimal, equilibrium mutation rate which balances the effects of deleterious and beneficial mutations (Fisher 1930; Kimura 1967; Leigh 1970; Orr 2000). As previously noted, the primary assumptions of most optimal mutation rate models include continual adaptation, complete linkage, and an immediate change in the deleterious mutational load when the mutation rate changes. When these assumptions are violated, then theory suggests there need not be evolution towards an optimal mutation rate; indeed, there might be no theoretical bounds on how high a mutation rate may then evolve (Andre and Godelle

2005; Gerrish et al. 2007). Previous experimental work has demonstrated that a high mutation rate can evolve in an asexual population and be maintained for tens of thousands of generations (Sniegowski et al. 1997; Shaver et al. 2002; Cooper et al. 2003).

Here, fluctuation testing and sequencing of the dnaQ locus in clones from the double- mutator populations suggest that they still maintain their mutator phenotype and genotype at the final time point, despite the consequences discussed below. If we consider that beneficial mutations may have been arising even as the population‘s fitness declined, then it is unsurprising that no mutator-revertants arose and reached a detectable frequency; an individual with a lower mutation rate is both less likely to produce a beneficial mutation and more likely to be lost by drift than the dominant mutator- genotype.

50 3.4.4. The consequences of a very high mutation rate

As noted previously, the majority of mutations that affect phenotype are deleterious

(Kneightley and Eyre-Walker 1999; Lynch et al. 1999), and thus any mutation rate may impose a cost on a population that can be measured by reductions in fitness due to mutation (Muller 1950; Haldane 1957; Lynch and Gabriel 1990); the cost has been termed the mutational load and is described as the difference between the maximum potential fitness, the theoretical fitness of an individual that carries the fittest available allele at each locus, and the average fitness in a population (Haldane 1930). This difference is increased when any phenotype-affecting mutation arises, therefore an increasing mutation rate also increases the genetic load carried by a population. Barring any restrictions on the potential size of the mutational load, it is conceivable that it may increase, in conjunction with an upwardly evolving mutation rate, to the point where it reduces fitnesses to a level sufficient to cause extinction. Many stochastic and deterministic processes have been proposed that address the fitness consequences of bearing a mutation rate that is exceptionally high. These processes are reviewed below and discussed in the context of the results reported here.

Muller‘s ratchet is a process that describes the irreversible accumulation of deleterious mutations in the genomes of an asexual population (Muller 1964; Haigh 1978; Gessler

1995; Gordo and Charlesworth 2000; Goyal et al. 2011); it is important to note that

Muller‘s ratchet and most models derived from it assume that that no beneficial mutations are available. The ratchet was first proposed to explain the advantage of sexual reproduction (Muller 1932, 1964), but it has since been invoked more generally as 51 a fitness-eroding process in asexual populations (Chao 1990; Duarte et al. 1992;

Andersson and Hughes 1996; Charlesworth and Charlesworth 1997). There are both deterministic and stochastic regimes of Muller‘s ratchet that may concurrently operate to reduce fitness. If we assume a Poisson distribution of the number of mutations, then the size of No, fittest class of individuals (those with the a mutation free genome) is expected to be given by (Haigh 1978)

where N is the effective population size, Ud is the deleterious genomic mutation rate, and sd is the average fitness effect of a deleterious mutation. If we ignore beneficial mutations for now, we may estimate No in the single- and double-mutator populations.

Kibota and Lynch (1996) inferred sd≈.02 and Ud ≥ 0.0002 per generation in E. coli growing in a minimal medium similar to DM1000; assuming that changes in this genomic deleterious mutation rate would be proportional to changes in the point mutation rate, my double-mutator strain is expected to have a deleterious genomic mutation rate of

~.9 (Recall that the double-mutator has a 4,500 fold increase in mutation rate over wild- type E. coli, thus 4,500*0.0002 = 0.9) while the single mutator is estimated to have a rate of ~.02 (Recall that the single-mutator has a 100 fold increase in mutation rate over wild- type E. coli, thus 100*0.0002 = 0.02) deleterious mutations per genome per generation.

-15 No is predicted to be ~1.4*10 ( in the double-mutator and ~18,400

( in the single-mutator populations. Gessler (1995) noted that a

52 finite asexual population may deterministically decrease in fitness if the size of the fittest class(es) in every successive generation is much less than 1 individual; it seems that the double-mutator populations meet this condition while the single-mutators do not.

As noted previously, there is a stochastic regime of Muller‘s ratchet that can also act to reduce the fitness of an asexual population. Drift-driven loss of a population‘s mutation- free fitness class irreversibly reduces the average fitness of the population; the ‗mutation- free‘ genotype is lost forever and those individuals with one mutation become the new

‗mutation-free‘ class. Repeated turns of Muller‘s ratchet result in a population with reduced fitness. The stochastic regime of Muller‘s ratchet is slowed in large populations, but a sufficiently high mutation rate may still deterministically drive even large populations extinct. Loss of fitness due to Muller‘s ratchet has been experimentally demonstrated (Chao 1990; Andersson and Hughes 1996). In theory, any mutation rate above zero may drive the process and ultimately reduce the absolute fitness of an asexual population to less than one, although beneficial mutations are known to retard the ratchet

(Bachtrog and Gordo 2004; Goyal et al. 2011).

Lynch et al (1993) proposed a variant of Muller‘s ratchet that incorporates population demographics: mutational meltdown. Under this regime, the accumulation of deleterious mutations under Muller‘s ratchet results in decreased population size, which further facilitates the accumulation of even more deleterious mutations through drift.

Consequently, fitness and population size concurrently enter a downward spiral which may ultimately result in extinction (Lynch et al. 1993; Lynch et al. 1995). Loss of fitness

53 due to mutational meltdown has been demonstrated in laboratory strains of yeast (Zeyl et al. 2001). As with the initial formulation of Muller‘s ratchet, beneficial mutations may inhibit mutational meltdown (Goyal et al. 2011).

Bull et al (2007; 2008) proposed a model of mutation-driven fitness decline that allows for growing populations: lethal mutagenesis. Unlike like the previously discussed models of Muller‘s ratchet, lethal mutagenesis is a deterministic process. Models of lethal mutagenesis predict that a sufficiently high genomic mutation rate will generate an overwhelming number of deleterious mutations that erode a population‘s equilibrium absolute fitness to below one (Bull et al. 2007; Bull and Wilke 2008). Notably, extinction by lethal mutagenesis is theoretically independent of population size but is likely to be slowed or prevented by the availability of beneficial mutations (Bull and

Wilke 2008). The theory of lethal mutagenesis assumes that that a population is continually growing and the condition for extinction is defined as

where b is the average number of offspring for each individual and Ud is the deleterious mutation rate. If we relax the assumption that the population is continually growing and fix b as 1 (i.e. every individual, on average, replaces itself in the populations and the population size remains constant), then it becomes apparent that any mutation rate is sufficient to drive an asexual population extinct and the lethal mutagenesis model is merely a reformulation of Muller‘s ratchet (Colato and Fontanari 2001). The relevancy

54 of the lethal mutagenesis model to experimental populations of fixed size is questionable.

Although lethal mutagenesis has been implicated in the extinction of viral populations

{Graci, 2008 #290; Graci, 2007 #389}, several attempts at chemically inducing lethal mutagenesis have either been unsuccessful (Springman et al. 2010, see below) or have failed to distinguish between theories of lethal mutagenesis and error catastrophe, a separate model of extinction (Crotty et al. 2002).

Models of Muller‘s ratchet, including mutational meltdown and lethal mutagenesis, assume that beneficial mutations are unavailable; all three processes would be slowed or halted completely if beneficial mutations were available (Bachtrog and Gordo 2004;

Goyal et al. 2011). In contrast, the error catastrophe model operates in the presence of available beneficial mutations (Eigen 1971; Eigen and Schuster 1979). In its original formulation, error catastrophe occurs when the mutation rate of a population surpasses an

―error threshold‖, the point at which the accumulation of mutations overwhelms the strength of purifying selection to maintain the fittest genotype (Eigen and Schuster 1979) and any sequence is equally likely to be present, regardless of fitness. Theoretical work regarding error catastrophe is based on quasispecies models with infinite population size, soft selection, and a simplistic fitness landscape with a single fit genotype (Brumer et al.

2006; Summers and Litwin 2006). By definition, a population of infinite size cannot go extinct, but surpassing the error threshold is usually assumed to be akin to extinction.

Recent theoretical work (Gerrish and Sniegowski, in prep.) suggests a new derivation of the ―variance threshold‖ mutation rate that can be applied to experimental populations

55 and gives a quantitative prediction about the relationship between genomic mutation rate and fitness variation in a population that will lead to extinction:

Where U is the genomic mutation rate, is the average effect of all mutations, and is the additive genetic variance in fitness within the population. can be defined as the product of the fraction and mean effect of beneficial mutations minus the product of the fraction and mean effect of deleterious mutations, thus . Intuitively, this formulation makes sense, as represents the overall deleterious effects of new

mutations and represents the strength of natural selection (Fisher 1930); therefore, if the above condition is met, then the mutation rate of a population is above the variance threshold and fitness should deterministically decline. A key advantage of this formulation, relative to previous iterations of the error catastrophe model, is that all of the variables can be estimated from experimental results. Drake (1991) conservatively estimated a U≈.0019 in wild-type E.coli; assuming that changes in this genomic mutation rate would be proportional to changes in the point mutation rate, the double-mutator strain is expected to have a genomic mutation rate of ~8.55 (4,500*0.0019 = 8.55) while the single mutator is estimated to have a rate of .19 (100*.0019=.19) mutations per genome per generation. If we assume that beneficial mutations are negligible (fbmb=0), then is equivalent to fdmd; As noted previously, Kibota and Lynch (1996) used mutation accumulation experiments to conservatively estimate md≈.02 and Ud ≈ 0.0002 for wild-

type E. coli growing in minimal media, thus . Given these estimates and

56 assuming that beneficial mutations are negligible, fitness variance must exceed ~.018

( ))) in the double-mutator population and 5.3*10-4 (

))) in the single-mutator population to prevent fitness from declining.

The fitness variance of the assayed double-mutator population does not exceed .018 at any time point while the fitness variance of the single-mutator is consistently well above the predicted variance threshold (Fig. 3-4), thus it appears that the mutation rate of the assayed double-mutator population had surpassed the variance threshold from the onset of propagation while the single-mutator population never approached it. It is important to note that these estimates of the threshold are based on the assumption that beneficial mutations are negligible. However, it is possible that the mutation rate of the double- mutator population is well beyond the predicted error threshold and beneficial mutations may merely slow the decline in fitness, not stop it. Recent theoretical work (Gerrish and

Sniegowski, in prep.) suggests a second novel derivation of the ―error threshold‖ that gives a quantitative prediction about the relationship between deleterious genomic mutation rate and the fate of new beneficial mutations in a population that will lead to extinction:

Where is the deleterious mutation rate, is the fitness of the fittest individual in

the population, and , where N is population size, is the fitness of

individual i , and S is the set of all suboptimal individuals (with fitness less than ).

57 represents the number of incoming deleterious mutations and represents a measure of mutational load, which increases when a new beneficial mutation arises but decreases as the fittest genotype accrues deleterious mutations (see Muller 1950; Haldane

1957) but. The probability and rate of fixation of a novel beneficial mutation is dependent on the mutation‘s fitness effect relative to the average fitness of the population. If we consider a population with a high mutation rate, the fitness advantage of genotype with a beneficial mutation may be quickly eroded by arising deleterious mutations, thus the probability and rate of fixation may decline to zero. A population that cannot fix beneficial mutations may ultimately be destined for extinction if the strength of deleterious mutations is sufficiently powerful. As noted above, is estimated to be

~0.9 in the double-mutator and .02 in the single-mutator genotypes. The results shown in

Fig. 3-4 clearly show that the single-mutator population never approaches the error threshold while the double-mutator surpasses the threshold at all but one assayed time point.

The current results demonstrate that there are circumstances under which a sufficiently high mutation rate can erode the fitness of a population as predicted by theory (Eigen and

Schuster 1977; Haigh 1978; Gessler 1995). It remains unclear which theoretical model or combination of models best describes the fitness decline observed in this experiment. In addition, the relationship between beneficial mutations and fitness decline has received insufficient theoretical and experimental treatment and remains an important subject for future research.

58 Chapter 4. General Discussion

4.1. Introduction

Recent theory (Andre and Godelle 2005) and simulation results (Gerrish et al. 2007) suggest that recurrent mutator hitchhiking in an asexual population should cause the evolution of mutation rate to be upwardly biased. Theory also suggests that a sufficiently high mutation rate might contribute to fitness decline and extinction in an asexual population (Muller 1964; Eigen 1971; Bull and Wilke 2008). The work presented in chapters II and III of this thesis experimentally investigated both of these theoretical predictions. In this chapter, I review the primary results of the experiments, discuss their possible limitations and implications, and suggest future directions of study.

4.2. Primary results

The results presented in chapter II support the notion that there are circumstances under which recurrent mutator hitchhiking can indeed occur in an asexual population. A mutator genotype bearing a mutation rate 4,500-fold greater than the wild-type rate fixed in replicate populations of genotypes that already carried a 100-fold increase in mutation rate over the wild-type genotype. The results presented in chapter III show that a 4,500- fold increase in mutation rate is sufficient to cause a decline in fitness in replicate asexual populations of E. coli over 2,500 generations, whereas a 100-fold increase in mutation rate is not. Taken together, the results presented in both chapters are consistent with the

59 idea that natural selection may drive the fixation of a very high mutation rate which, paradoxically, is sufficient to drive fitness down.

4.3. Limitations of the experimental approach and future directions

Although my results support the theory that a high mutation rate can evolve by natural selection, the approaches used in chapter II do have obvious limitations. The double- mutator genotype was seeded into populations at a higher frequency than would have applied had it arisen spontaneously, as was done in previous work with single-mutators

(Chao and Cox 1983). This approach does not answer the question of how a genotype with a higher mutation rate would attain the frequency necessary for it to hitchhike in a natural population. Previously, mutators have been observed to arise spontaneously in experimental populations with a wild-type mutation rate and, over time, acquire chance associations with beneficial mutations and hitchhike towards fixation (Sniegowski et al.

1997; Shaver et al. 2002). Spontaneously arising mutators have arisen in natural populations as well and have been observed at high frequency (Matic 1997; Boe et al.

2000; Loeb 2001).

Theory predicts that recurrent spontaneous mutator hitchhiking (ie the sequential fixation of more than one mutator allele in a population) should be observed in a single population. A good experimental test of this prediction has recently been proposed

(Wylie et al. 2009) and is under development in our laboratory: propagate populations of

60 single-mutator E. coli and observe whether a novel mutator allele arises and hitchhikes towards fixation. As noted previously, spontaneous mutator hitchhiking has occurred in experimentally propagated population of wild-type E. coli. Such an observation in experimentally propagated populations of mutator E. coli would certainly lend further experimental support to theory predicting recurrent mutator-hitchhiking.

The fitness decline observed in the propagation experiment (Ch. III) is consistent with various theories of mutation-driven erosion of fitness. It is possible, or even probable that some variations of Muller‘s ratchet and error catastrophe contribute to fitness decline; my results do not address, however, to what degree each model might contribute to the fitness decline. It might be possible to experimentally address this gap by manipulating the size of the double-mutator populations. For example, if the double- mutator populations were operating under the stochastic regime of Muller‘s ratchet then increasing the population size might prevent extinction in the short-term. Unfortunately, such a test would be complicated because the double-mutator populations are likely also operating under the deterministic regime of Muller‘s ratchet and it would require a

21 population size on the order of 10 individuals , an extraordinarily large population size, to have any chance of producing a mutational class that does not have at least one new mutation every generation. It should be noted again, however, that estimates of assume that beneficial mutations are negligible, thus it is possible that a smaller population could escape the deterministic regime of fitness decline by generating a sufficient number of beneficial mutations (Goyal et al. 2011).

61 More detailed and more replicated data regarding population fitness may also provide insight into the processes that erode fitness. The labor-intensive approach utilized here does not allow for a very high degree of replication; I only propagated six double- and six single-mutator populations and was limited to five-fold replication when estimating absolute fitness of each population at each time point. Propagating large numbers of populations in microplates would allow daily estimates of population fitness (data would be available for every facet of fitness: the speed of recovery out of stationary phase, maximal growth rate, and survival in stationary phase), easy archiving of evolving populations at shorter intervals, and a high degree of replication.

A significant complication in quantitative analysis of mutation-driven extinction is the effect of beneficial mutations. Most models of Muller‘s ratchet assume that no beneficial mutations are available and my estimates of the error threshold assume that beneficial

mutations are negligible (recall that and I estimated

. In contrast, beneficial mutations are known to be available to the

double-mutator populations: all of the single-mutator and some of the double-mutator populations increased in fitness early in the propagation. If beneficial mutations are considered when calculating the error threshold, then the mutation rate necessary to surpass the error threshold is predicted to be greater. Remarkably, the availability of beneficial mutations was ultimately insufficient in preventing fitness decline in the double-mutator populations which suggests that the mutation rate of the double-mutator was, indeed, well above the error threshold. Although some models of Muller‘s ratchet

62 and error catastrophe attempt to include beneficial mutations (Bachtrog and Gordo 2004;

Goyal et al. 2011), current theory does not provide an adequate framework for analysis of the current results.

The results presented here generally support the theory that mutation rates may evolve upwards to a rate that threatens the fitness of the population. However, the experiments do not demonstrate both phenomena (hitchhiking and extinction) in one continuously propagated population. I suggested above that the experimental propagation of mutator

E. coli populations of mutator E. coli might demonstrate whether a novel mutator allele can spontaneously arise and hitchhike towards fixation; assuming mutator hitchhiking is observed in such a population, then these populations might be propagated continually and observed for fitness decline. Such an observation would certainly lend significant experimental support to theory predicting recurrent mutator-hitchhiking and fitness decline.

4.4. Implications

There are several basic and practical implications of the current work. It has been noted previously that many asexual lineages derived from sexual populations (i.e., parthenogens) are evolutionarily short-lived (e.g. Haigh 1978; Maynard Smith 1982). As noted in Ch. III, models of Muller‘s ratchet were initially proposed to explain an advantage of sexual (or a disadvantage of asexual) reproduction. The current results, in accordance with previous theory (Andre and Godelle 2005) and simulations (Gerrish et al. 2007), suggest that the upward bias of mutation rate evolution, not solely deleterious 63 mutations, may be influential in determining the ultimate fate of an asexual population.

In contrast, recombination in sexual populations may prevent mutator hitchhiking, thus a sexual population might rarely be threatened by a high mutation rate. It is important to note that many populations that are considered asexual, including natural populations of

E. coli, do undergo some form recombination. It remains an open question as to the level of recombination necessary to prevent recurrent mutator hitchhiking.

It has been observed that RNA viruses have exceptionally high genomic mutation rates relative to DNA based microbes (Drake et al. 1998; Drake and Holland 1999). With such a high basal genomic mutation rate, a feasible route to treating viral infections might be the induction of a mutation rate that surpasses a critical threshold. Current mutation- based approaches of treating viral infections involve the application of mutagens to increase the genomic mutation rate of a virus (Loeb et al. 1999; Crotty et al. 2002;

Vignuzzi et al. 2005b; Graci et al. 2007; Graci et al. 2008). However, viruses are likely to evolve resistance to mutagens in the same manner that they can evolve resistance to many other antiviral drugs, often through loss-of-function mutations (Springman et al.

2010). Indeed, as noted in chapter III, preliminary experiments in our lab with E. coli that have a viral-like mutation rate suggest that resistance to base analogue mutagens can evolve rapidly, negating any potential long-term mutagenic effects (Balk et al. in prep)

Conversely, simulations of organisms with RNA virus-level mutation rates that have been treated with a recombination-inhibiting drug suggest that the evolution of resistance to this drug type does not occur and populations are likely to go extinct via increased mutation rates and consequent error catastrophe (P. Gerrish, personal communication). 64 This suggests that anti-recombinatorial drugs may be a potential candidate for the treatment of viral infections.

Bibliography

Ancliff, M., and J.-M. Park. 2009. Maximum, minimum, and optimal mutation rates in dynamic environments. Physical Review E 80:061910.

Andersson, D. I., and D. Hughes. 1996. Muller's ratchet decreases fitness of a DNA- based microbe. Proceedings of the National Academy of Science 93:906-907.

Andre, J. B., and B. Godelle. 2005. The evolution of mutation rates in finite asexual populations. Genetics 172.

Andre, J. B., and B. Godelle. 2006. The evolution of mutation rates in finite asexual populations. Genetics 172.

Bachtrog, D., and I. Gordo. 2004. Adaptive Evolution of Asexual Populations Under Muller's Ratchet. Evolution 58:1403-1413.

Baer, C. F., M. M. Miyamoto, and D. R. Denver. 2007. Mutation rate variation in multicellular eukaryotes: causes and consequences. Nature Reviews Genetics 8:619-631.

Blattner, F. R., G. P. III, C. A. Bloch, N. T. Perna, V. Burland, M. Riley, J. Collado- Vides, J. Glasner, C. K. Rode, G. F. Mayhew, J. Gregor, N. W. Davis, h. A. Kirpatrick, M. A. Goeden, D. J. Rose, B. Mau, and Y. Shao. 1997. The Complete Genome Sequence of Escherichia coli K-12. Science 277:1453-1462.

Blount, Z. D., C. Z. Borland, and R. E. Lenski. 2008. Historical contingency and the evolution of a key innovation in an experimental population of Escherichia coli. Proceedings of the National Academy of Science 105:7899-7906.

Boe, L., M. Danielson, S. Knudsen, J. B. Petersen, J. Maymann, and P. R. Jensen. 2000. The frequency of mutators in populations of Escherichia coli. Mutat. Res. 448:47- 55.

66 Brumer, Y., F. Michor, and E. I. Shakhnovich. 2006. Genetic instability and the quasispecies model. Journal of Theoretical Biology 241:216-222.

Bull, J. J., and I. J. Molineux. 1992. Molecular Genetics of Adaptation in an Experimental Model of Cooperation. Evolution 46:882-895.

Bull, J. J., R. Sanjuan, and C. O. Wilke. 2007. Theory of lethal mutagenesis for viruses. J. Virology 81:2930-2939.

Bull, J. J., and C. O. Wilke. 2008. Lethal Mutagenesis of Bacteria. Genetics 180:1061- 1070.

Carlton, B. C., and B. J. Brown. 1981. Gene Mutation. Pp. 222-242 in P. Gerhardt, ed. Manual of Methods for General Bacteriology. ASM Press, Washington, DC.

Chao, L. 1990. Fitness of RNA virus decreased by Muller's ratchet. Nature 348:454-455.

Chao, L., and E. C. Cox. 1983. Competition between high and low mutating strains of Escherichia coli. Evolution 37:125-134.

Charlesworth, B. 1990. Optization Models, Quantitative Genetics, And Mutation. Evolution 44:520-538.

Charlesworth, B., and D. Charlesworth. 1997. Rapid fixation of deleterious alleles can be caused by Muller's ratchet. Genetics Research 70:63-73.

Chen, P., and E. Shakhnovich. 2009. Lethal Mutagensis in Viruses and Bacteria. Genetics 183:639-650.

Chikova, A. K., and R. M. Schaaper. 2006. Mutator and Antimutator Effects of the Baceriophage P1 hot Gene Product. Journal of Bacteriology 188:5831.

Clune, J., D. Misevic, C. Ofria, R. E. Lenski, S. F. Elena, and R. Sanjuan. 2008. Natural Selection Fails to Optimize Mutation Rates for Long-Term Adaptation on Rugged Fitness Landscapes. Public Library of Science Computational Biology 4.

Cohan, F. M., and A. A. Hoffmann. 1986. Genetic Divergence Under Uniform Selection. II. Different Responses To Selection For Knockdown Resistence To Ethanol Among Drosophila Melanogaster Populations And Their Replicate Lines. Genetics 114:145-163.

67 Colato, A., and J. F. Fontanari. 2001. Soluble Model for the Accumulation of Mutations In Asexual Populations. Physical Review Letters 87:238102.

Cooper, T. F., D. E. Rozen, and R. E. Lenski. 2003. Parallel changes in gene expression after 20,000 generation of evolution in Escherichia coli. Proceedings of the National Academy of Science 100:1072-1077.

Cox, E. 1976. Bacterial Mutator Genes and the Control of Spontaneous Mutation. Annual Review of Genetics 10:135-156.

Crotty, S., C. Cameron, and R. Andino. 2002. Ribavirin's antiviral mechanism of action: lethal mutagenesis? Journal of Molecular Medicine 80:86-95.

Crow, J. F., and M. Kimura. 1970. An Introduction To Population Genetics Theory. Harper and Row, New York, NY. de Visser, J. A. G. M., and D. E. Rozen. 2005. Limits to adaptation in asexual populations. J. Evolution. Biol. 18:779-788. de Visser, J. A. G. M., C. W. Zeyl, P. J. Gerrish, J. L. Blanchard, and R. E. Lenski. 1999. Diminishing returns from mutation supply rate in asexual populations. Science 283:404-406.

Desai, M. M., D. S. Fisher, and A. W. Murray. 2007. The Speed of Evolution and Maintenance of Variation in Asexual Populations. Current Biology 17:385-394.

Drake, J. 1991. A constant rate of spontaneous mutation in DNA-based microbes. Proceedings of the National Academy of Science 88:7160-7164.

Drake, J. W. 1990. Evolving mutation rates and prospects for antimutagenesis. Plenum, New York.

Drake, J. W. 1993. General Antimutators are Improbable. Journal of Molecular Biology 229:8-13.

Drake, J. W., B. Charlesworth, D. Charlesworth, and J. F. Crow. 1998. Rates of Spontaneous Mutation. Genetics 148:1667-1686.

Drake, J. W., and J. J. Holland. 1999. Mutation rates among RNA viruses. Proceedings of the National Academy of Science 96:13910-13913.

68 Duarte, E., D. Clarke, A. Moya, E. Domingo, and J. Holland. 1992. Rapid fitness losses in mammalian RNA virus clones due to Muller's ratchet. Proceedings of the National Academy of Science 89:6015-6019.

Eigen, M. 1971. Self organization of matter and the evolution of biological macromolecules. Naturewissenschaften 58:465-523.

Eigen, M., and P. Schuster. 1977. A principle of natural self-organization. Naturwissenschaften 64:541-565.

Eigen, M., and P. Schuster. 1979. The Hypdercycle - A Principle of Natural Self- Organization. Springer Verlag, Berlin.

Elena, S. F., C. O. Wilke, C. Ofria, and R. E. Lenski. 2007. Effects of population size and mutation rate on the evolution of mutational robustness. Evolution 61:666-674.

Ferenci, T. 2008. The spread of a beneficial mutation in expermiental bacterial populations: the influence of the environment and genptype on the fixation of rpoS mutations. Heredity 100:446-452.

Fijalkoswka, I. J., and R. M. Schaaper. 1995. Effects of Escherichia coli dnaE antimutator alleles in a proofreading-deficient mutD5 strain. Journal of Bacteriology 177:5979-5986.

Fijalkowska, I. J., & Schaaper, R.M. 1993. Antimutator Mutations in the α Subunit of Eschericia coli DNA Polymerase III: Identification of the Responsible Mutations and Alignment With Other DNA Polymerases. Genetics 134:1039 - 1044.

Fijalkowska, I. J., R. L. Dunn, and R. M. Schaaper. 1993. Mutants of Escherichia coli with Increased Fidelity of DNA Replication. Genetics 134:1023-1030.

Fijalkowska, I. J., and R. M. Schaaper. 1996. Mutants in the Exo I motif of Escherichia coli dnaQ: Defective proofreading and inviability due to error catastrophe. Proceedings of the National Academy of Science 93:2856-2861.

Fisher, R. A. 1930. The Genetical Theory of Natural Selection. Oxford University Press, Oxford.

Friedberg, E. C., G. C. Walker, and W. Siede. 1995. DNA Repair and Mutagenesis. ASM Press, Washington, DC.

69 Gentile, C. F., S.-C. Yu, S. A. Serrano, P. J. Gerrish, and P. D. Sniegowski. 2011. Competition between high- and higher-mutating strains of Escherichia coli. Biology Letters 7:422-424.

Gerrish, P. J., A. Colato, A. S. Perelson, and P. D. Sniegowski. 2007. Complete genetic linkage can subvert natural selection. PNAS 104:6266-6271.

Gessler, D. D. G. 1995. The constraints of finite size in asexual populations and the rate of the ratchet. Genetical Research 66:241-253.

Gordo, I., and B. Charlesworth. 2000. On the Speed of Muller's Ratchet. Genetics 152:2137-2140.

Goyal, S., D. J. Balick, E. R. Jerison, R. A. Neher, B. I. Shraiman, and M. M. Desai. 2011. Rare beneficial mutations can halt Muller's ratchet. Proceedings of the National Academy of Science.

Graci, J. D., D. A. Harki, V. S. Korneeva, J. P. Edathil, K. Too, D. Franco, E. D. Smidansky, A. V. Paul, B. R. Peterson, D. M. Brown, D. Loakes, and C. E. Cameron. 2007. Lethal Mutagenesis of Poloiovirus Mediated by a Mutagenic Pyrimidine Analogue. Journal of Virology 81:11256-11266.

Graci, J. D., K. Too, E. D. Smidansky, J. P. Edathil, E. W. Barr, D. A. Harki, J. E. Galarraga, J. J. Martin Bollinger, B. R. Peterson, D. Loakes, D. M. Brown, and C. E. Cameron. 2008. Lethal Mutagenesis of Picornaviruses with N-6-Modified Purine Nucleoside Analogues. Antimicrobial Agents and Chemotherapy 52:971- 979.

Haigh, J. 1978. The accumulation of deleterious genes in a population -- Muller's Ratchet. Theorertical Population Biology 14:251-267.

Haldane, J. B. S. 1930. A methematical theory of natural and artificial selection (part VI. Isolation). Mathematical Proceedings of the Cambridge Philosophical Society 26:220-230.

Haldane, J. B. S. 1957. The Cost of Natural Selection. Journal of Genetics 55:511-524.

Herr, A. J., M. Ogawa, N. A. Lawrence, L. N. Williams, J. M. Eggington, M. Singh, R. A. Smith, and B. D. Preston. 2011. Mutator Suppression and Escape from Replication Error - Induced Extinction in Yeast. PLoS Genetics 7:e1002282.

70 Holloway, A. K., T. Palzkill, and J. J. Bull. 2007. Experimental Evolution of Gene Duplicates in a Bacterial Plasmid Model. Journal of Molecular Evolution 64:215- 222.

Holsinger, K. E., and M. W. Feldman. 1983. Modifiers of mutation rate: Evolutionary optimum with complete selfing. Proceedings of the National Academy of Science 80:6732-6734.

Ishii, K., H. Matsuda, Y. Iwasa, and A. Sasaki. 1989. Evolutionaty Stable Mutation Rate in a Periodically Changind Environment. Genetics 121:163-174.

Johnson, T. 1999. The approach to mutation-selection balance in an infinite asexual population, and the evolution of mutation rates. P. Roy. Soc. of. Lond. B Bio. 266:2389-2397.

Khan, A. I., D. M. Dinh, D. Schneider, R. E. Lenski, and T. F. Cooper. 2011. Negative Epistasis Between Beneficial Mutations in an Evolving Bacterial Population Science 332:1193-1196.

Kibota, T. T., and M. Lynch. 1996. Estimate of the genomic mutation rate deleterious to overall fitness in E. coli. Nature 381:694-696.

Kimura, M. 1967. On the evolutionary adjustment of spontaneous mutation rates. Genetical Research 9:23-34.

Kneightley, P. D., and A. Eyre-Walker. 1999. Terumi Mukai and the Riddle of Deleterious Mutation Rates. Genetics 153:515-523.

Kondrashov, A. S. 1995. Modifiers of mutation-selection balance: general approach and the evolution of mutation rates. Genetical Research 66:53-69.

Krishnaswamy, S., J. A. Rogers, R. J. Isbell, and R. G. Fowler. 1993. The higher mutator activity of the dnaQ49 allele of Escherichia coli is medium-dependant and results from both defective 3' --> 5' proofreading and methyl-directed mismatch repair. Mutation Research 288:311-319.

LeClerc, J. E., B. Li, W. L. Payne, and T. A. Cebula. 1996. High mutation frequencies among Escherichia coli and Salmonella pathogens. Science 274:1208-1211.

71 Lederberg, S. 1965. Genetics of Host-Controlled Restriction and Modification of Deoxyribonucleic Acid in Escherichia coli. Journal of Bacteriology 91:1029- 1036.

Leigh, E. G. 1970. Natural Selection and Mutability. The American Naturalist 104:301- 305.

Lenski, R. E. 1988. Experimental Studies of Pleiotropy and Epistatis in Escherichia coli. I. Variation in Competitive Fitness Among Mutants Resistant to Virus T4. Evolution 42:425-432.

Lenski, R. E., M. R. Rose, S. C. Simpson, and S. C. Tadler. 1991. Long-Term Experimental Evolution In Escherichia Coli. I. Adaptation And Divergence During 2,000 Generations. The American Naturalist 138:1315-1341.

Liberfarb, R. M., and V. Bryson. 1970. Isolation, Characterization, and Genetic Analysis of Mutator Genes in Escherichia coli B and K-12. Journal of Bacteriology 104:363-375.

Loeb, L. A. 1991. Mutator Phenotype May Be Required for Multistage Carcinogenesis. Cancer Research 51:1075-3079.

Loeb, L. A. 2001. A mutator phenotype in cancer. Cancer Res. 61:3230-3239.

Loeb, L. A., J. M. Essigmann, F. Kazazi, J. Zhang, K. D. Rose, and J. I. Mullins. 1999. Lethal mutagenesis of HIV with mutagenic nucleoside analogs. Proceedings of the National Academy of Science 96:1492-1497.

Luria, S. E., and M. Delbruck. 1943. Mutations of Bacteria from Virus Sensitivity to Virus Resistance. Genetics 28:491-511.

Lynch, M. 2008. The Cellular, Developmental and Population-Genetic Determinants of Mutation-Rate Evolution. Genetics 180:933-943.

Lynch, M., J. Blanchard, D. Houle, T. Kibota, S. Schultz, L. Vassilieva, and J. Willis. 1999. Perspective: Spontaneous Deleterious Mutation. Evolution 53:645-663.

Lynch, M., R. Burger, D. Butcher, and W. Gabriel. 1993. The Mutational Meltdown in Asexual Populations. Journal of Heredity 84:339-344.

72 Lynch, M., J. Conery, and R. Burger. 1995. Mutational Meltdowns In Sexual Populations. Evolution 49:1067-1080.

Lynch, M., and W. Gabriel. 1990. Mutation Load and the Survival of Small Populations. Evolution 44:1725-1737.

Mani, G. S., and B. C. Clarke. 1990. Mutational Order: A Major Stochastic Process in Evolution. Proceedings of the Royal Society of London B 240:29-37.

Mao, E. F., L. Lane, J.Lee, and J. H. Miller. 1997. Proliferation of mutators in a cell population. Journal of Bacteriology 179:417-422.

Matic, I. 1997. Highly variable mutation rates in commensal and pathogenic Escherichia coli. Science 277:1833-1834.

Maynard Smith, J. 1982. Evolution and the theory of games. Cambridge University Press, Cambridge.

Miller, J. H. 1992. A Short Course in Bacterial Genetics: A Laboratory Manual for Escherichia coli and Related Bacteria. Cold Spring Harbor Laboratory, Cold Spring Harbor, NY.

Miller, J. H., P. Funchain, W. Clendenin, T. Huang, A. Nguyen, E. Wolff, A. Yeung, J.- H. Chiang, L. Garibyan, M. M. Slupska, and H. Yang. 2002. Escherichia coli Strains Lacking Nucleoside Diphosphate Kinase (ndk) Are Powerful Mutators for Base Substitutions and Frameshifts in Mismatch-Repair-Deficient Strains. Genetics 162:5-13.

Miyake, T. 1960. Mutator factor in Salmonella typhimurium. Genetics 45:11-14.

Muller, H. J. 1932. Some Genetic Aspects of Sex. The American Naturalist 66:118-138.

Muller, H. J. 1950. Our load of mutations. American Journal of Human Genetics 2:111- 176.

Muller, H. J. 1964. The Relation of Recombination to Mutational Advance. Mutation Research 106:2-9.

Negishi, K., D. Loakes, and R. M. Schaaper. 2002. Saturation of DNA Mismatch Repair and error Catastrophe bya Base Analogue in Escherichia coli. Genetics 161:1363- 1371. 73 Negishi, K., D. M. Williams, Y. Inoue, K. Moriyama, D. M. Brown, and H. Hayatsu. 1997. The mechanism of mutation induction by a hydrogen bong ambivalent, bicyclic N4-oxy-2'-deoxyctidine in Escherichia coli. Nucleic Acids Research 25:1548-1552.

Nilsson, M., and N. Snoad. 2002. Optimal Mutation Rates in Dynamic Environments. Bulletin of Mathematical Biology 64:1033-1043.

Notley-McRobb, L., S. Seeto, and T. Ferenci. 2002. Enrichment and Elimination of mutY Mutators in Escherichia coli Populations. Genetics 162:1055-1062.

Oller, A. R., and R. M. Schaaper. 1994. Spontaneous Mutation in Escherichia coli Containing the dnaE911 DNA Polyermase Antimutator Allele. Genetics 138:263- 270.

Orr, H. A. 2000. The rate of adaptation in asexuals. Genetics 155:961-968.

Perales, C., R. Agudo, and E. Domingo. 2009. Counteracting quasispecies adaptability: extinction of a ribavirin-resistant virus mutant by an alternative mutagenic treatment. PLoS One 4.

Perna, N. T., G. Plunkett, V. Burland, B. Mau, J. D. Glasner, D. J. Rose, G. F. Mayhew, P. S. Evans, J. Gregor, H. A. Kirkpatrick, G. Posfai, J. Hackett, S. Klink, A. Boutin, Y. Shao, L. Miller, E. J. Grotbeck, N. W. Davis, A. Lim, E. T. Dimalanta, K. D. Potamousis, J. Apodaca, T. S. Anantharaman, J. Lin, G. Yen, D. C. Schwartz, R. A. Welch, and F. R. Blattner. 2001. Genome sequence of enterohaemorrhagic Escherichia coli O157:H7. Nature 409:529-533.

Quinones, A., and R. Piechocki. 1985. Isolation and characterization of Escherichia coli antimutators. Molecular and General Genetics 201:315-322.

Schaaper, R. M. 1993. The Mutational Specificity of Two Escherichia coli dnaE Antimutator Alleles as Determined From lacI Mutation Spectra Genetics 134:1031-1038.

Schaaper, R. M. 1996. Suppressors of Escherichia coli mutT: antimutators for DNA replication errors. Mutation Research 350:17-23.

Schaaper, R. M., & Dunn, R.L. 1998. Effects of Escherichia coli dnaE antimutator mutants on mutagenesis by the base analog N4-aminocytidine. Mutation Research 402:23-28. 74 Schaaper, R. M., and R. L. Dunn. 1987. Escherichia coli mutT Mutator Effect during in Vitro DNA Synthesis. The Journal of Biological Chemistry 262:16267-16270.

Schaaper, R. M., and M. Radman. 1989. The extreme mutator effect of Escherichia coli mutD5 results from saturation of mismatch repair by excessive DNA replication errors. The European Molecular Biology Organization Journal 8:3511-3516.

Sevastopoulos, C. G., and D. A. Glaser. 1977. Mutator action by Escherichia coli strains carrying dnaE mutations. Proceedings of the National Academy of Science 74:3947-3950.

Shaver, A. C., P. G. Dombrowski, J. Y. Sweeney, T. Treis, R. M. Zappala, and P. D. Sniegowski. 2002. Fitness evolution and the rise of mutator alleles in experimental Escherichia coli populations. Genetics 162:557-566.

Shaver, A. C., and P. D. Sniegowski. 2003. Spontaneously Arising mutL Mutators in Evolving Escherichia coli Populations Are the Result of Changes in Repeat Length. Journal of Bacteriology 185:6076-6082.

Siegel, E. C., and J. J. Ivers. 1975. mut-25, a Mutation to Mutator Linked to purA in Escherichia coli. Journal of Bacteriology 121:524-530.

Sniegowski, P. D., P. J. Gerrish, T. Johnson, and A. Shaver. 2000. The evolution of mutation rates: separating causes from consequences. BioEssays 22:1057-1066.

Sniegowski, P. D., P. J. Gerrish, and R. E. Lenski. 1997. Evolution of high mutation rates in experimental populations of E. coli. Nature 387:703-705.

Springman, R., T. Keller, I. Molineux, and J. J. Bull. 2010. Evolution at a High Imposed Mutation Rate: Adaptation Obscures the Load in Phage T7. Genetics 184:221- 232.

Stoler, D. L., N. Chen, M. Basik, M. S. Kahlenberg, M. A. Rodriguez-Bigas, N. J. Petrelli, and G. R. Anderson. 1999. The onset and extent of genomic instability in sporadic colorectal tumor progression. Proceedings of the National Academy of Science 96:15121-15126.

Sturtevant, A. H. 1937. Essays on Evolution. I. On the Effects of Selection on Mutation Rate. The Quarterly Review of Biology 12:464-467.

75 Suarez, P., J. Valcarcel, and J. Ortin. 1992. Heterogeneity of the mutation rates of Influenza A viruses: isolation of mutator mutants. J. Virology 66:2491-2494.

Summers, J., and S. Litwin. 2006. Examining The Theory of Error Catastrophe. Journal of Virology 80:20-26.

Szidic, S., and M. Petranovic. 2003. Mismatch repair in the antimutator Escherichia coli mud. Mutation Research 522:27-32.

Taddei, F., M. Radman, J. Maynard Smith, B. Toupance, P. H. Gouyon, and B. Godelle. 1997. Role of mutator alleles in adaptive evolution. Nature 387:700-702.

Tenaillon, O., B. Toupance, H. L. Nagard, F. Taddei, and B. Godelle. 1999. Mutators, Population Size, Adaptive Landscape and the Adaptation of Asexual Populations of Bacteria. Genetics 152:485-493.

Thompson, D. A., M. M. Desai, and A. W. Murray. 2006. Ploidy controls the success of mutators and nature of mutations during budding yeast evolution. Curr. Bio. 16:1581-1590.

Travisano, M., F. Vasi, and R. Lenski. 1995. Long-Term Experimental Evolution in Escherichia coli. III. Variation Among Replicate Populations in Correlated Responses to Novel Environments Evolution 49:189-200.

Treffers, H. P., V. Spinelli, and N. O. Belser. 1954. A factor (or mutator gene) influencing mutation rates in Escherichia coli Proceedings of the National Academy of Science 40:1064-1071.

Tsao, J., S. Tavare, R. Salovaara, J. R. Jass, L. A. Aaltonen, and D. Shibata. 1999. Colorectal Adenoma and Cancer Divergence: Evidence of Multilineage Progression. American Journal of Pathology 154:1815-1824.

Tucker, M. S. 1997. Estimation of mutation frequencies in normal mammalian cells and the development of cancer. Seminars in Cancer Biology 8:407-419.

Vignuzzi, M., J. K. Stone, and R. Andino. 2005a. Ribavirin and lethal mutagenesis of poliovirus: molecular mechanisms, resistance and biological implications. Virus Res. 107:173-181.

76 Vignuzzi, M., J. K. Stone, and R. Andino. 2005b. Ribavirin and lethal mutagenesis of poliovirus: molecular mechanisms, resistance, and biological implications. Virus Research 107:173-181.

Wahl, L. M., and P. J. Gerrish. 2001. The probability that beneficial mutations are lost in populations with periodic bottlenecks. Evolution 55:2606-2610.

Weissman, D. B., M. M. Desai, D. S. Fisher, and M. W. Feldman. 2009. The rate at which asexual populations cross fitness valleys. Theorertical Population Biology 75:286-300.

Woods, R., D. Schneider, C. L. Winkworth, M. A. Riley, and R. E. Lenski. 2006. Tests of parallel molecular evolution in a long-term experiment with Escherichia coli. PNAS 103:9107-9112.

Woods, R. J., J. E. Barrick, T. F. Cooper, U. Shrestha, M. R. Kauth, and R. E. Lenski. 2011. Second-Order Selection for Evolvability in a Large Escherichia coli Population. Science 331:1433-1436.

Wylie, C. S., C.-M. Ghim, D. Kessler, and H. Levine. 2009. The Fixation Probability for Rare Mutators in Finite Asexual Populations Genetics 181:1595-1612.

Yin, J. 1993. Evolution of bacteriophage T7 in a growing plaque. Journal of Bacteriology 175:1272-1277.

Zeldovich, K. B., P. Chen, and E. I. Shakhnovich. 2007. Protein stability imposes limits on organism complexity and speed of molecular evolution. Proceedings of the National Academy of Science 104:16152-16157.

Zeyl, C., M. Mizesko, and J. A. G. M. De Visser. 2001. Mutational Meltdown in Laboratory Yeast Populations. Evolution 55:909-917.