D

M

S

Aa

b

5

a

ARR1AA

KMMEM

C

1

ihpg

S

h1

ARTICLE IN PRESSG ModelNAREP-2004; No. of Pages 7

DNA Repair xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

DNA Repair

j ourna l ho me pa ge: www.elsev ier .com/ locate /dnarepai r

ini Review

pontaneous mutation rates come into focus in Escherichia coli

shley B. Williamsa,b,∗

Institute for Genome Stability in Ageing and Disease, Medical Faculty, University of Cologne, Joseph-Stelzmann-Str. 26, 50931 Cologne, GermanyCologne Excellence Cluster for Cellular Stress Responses in Aging-Associated Diseases (CECAD), University of Cologne, Joseph-Stelzmann-Str. 26,0931 Cologne, Germany

r t i c l e i n f o

rticle history:eceived 25 August 2014eceived in revised form5 September 2014ccepted 20 September 2014vailable online xxx

a b s t r a c t

Although the long-term outcome of mutagenesis is evolution by natural selection, it can also have pro-found immediate effects even on the level of individual organisms. In humans, the accumulation ofmutations can cause many types of cancer; in bacteria, mutations can lead to dangerous antibioticresistance and other phenotypic changes; and in viruses, mutations can cause drastic changes in thepathogenesis or modes of transfer. For these reasons, among others, a thorough understanding of muta-

eywords:utagenesisutation rate

scherichia coliutational topology

genesis is extremely important. One of the fundamental properties of the mutagenesis is its rate—theprobability of a mutation occurring within a defined time frame. Despite the lengthy history of studieson mutagenesis and mutation rates, new and exciting findings continue to emerge. This review brieflysummarizes the state-of-the-art in mutation rate analysis and continues with a discussion of some recentcompelling discoveries on the mutational topology of the E. coli chromosome.

© 2014 Elsevier B.V. All rights reserved.

ontents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 002. Techniques for mutation rate analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

2.1. Estimation of mutation rates by fluctuation analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 002.2. Estimation of mutation rates by comparative genomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 002.3. Estimation of mutation rates using mutation accumulation experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

2.3.1. Estimation of mutation rates from long-term evolution experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 002.3.2. Estimation of mutation rates using MA lines with bottlenecks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

3. New insight into the mutational topology of the E. coli chromosome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 004. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

Conflict of interest statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

. Introduction

Mutagenesis is a double-edged sword—at the same time thatt provides the fuel for evolution by natural selection, it can also

making mutagenesis an important research topic since early inthe development of modern genetics as a discipline. In his 1928paper “The measurement of gene mutation rate in Drosophila, itshigh variability, and its dependence upon temperature” [1], Nobel

Please cite this article in press as: A.B. Williams, Spontaneous mutathttp://dx.doi.org/10.1016/j.dnarep.2014.09.009

ave deleterious effects on genome stability that can reduce fitness,romote tumorigenesis and cancer formation, or cause heritableenetic disorders. Some of these issues have long been appreciated,

∗ Correspondence to: CECAD Research Center, University of Cologne, Joseph-telzmann-Str. 26, 50931 Cologne, Germany. Tel.: +49 221 478 84206.

E-mail address: awilliam@uni-koeln.de

ttp://dx.doi.org/10.1016/j.dnarep.2014.09.009568-7864/© 2014 Elsevier B.V. All rights reserved.

laureate Hermann Joseph Muller framed one of the central chal-lenges in the study of spontaneous mutagenesis and the empiricalmeasurement of mutation rates: that because mutagenesis is sucha rare process under normal conditions, innovative and exceed-ingly sensitive techniques would be required to generate data

ion rates come into focus in Escherichia coli, DNA Repair (2014),

leading to interpretable “positive knowledge” (i.e. non-negativeresults) above the level of experimental error. In this seminalpaper, Muller went on to describe his method of “balanced lethals,”

ARTICLE IN PRESSG ModelDNAREP-2004; No. of Pages 7

2 A.B. Williams / DNA Repair xxx (2014) xxx–xxx

Table 1Examples of spontaneous mutation rates calculated by different methods.

Mutation rate (per base pairper generation × 10−10)

Mutation rate (per genomeper generation × 10−3)

Mutational target Method Reference

4.10 1.90 lacI Fluctuation analysis [4]6.90 3.30 lacI Fluctuation analysis [4]5.10 2.40 hisGDCBHAFE Fluctuation analysis [4]0.45 0.1–0.2 – Comparative genomics [17]7.90 3.70 lacI Fluctuation analysis [49]0.89 0.41 – Mutation accumulation with WGS [2]2.20 1.00 – Mutation accumulation with WGS [3]0.33 0.15 Rifampicin resistance Fluctuation analysis [3]

lidixi

aasfroomwqoIrm

aioaymre[htdaiw

2

hteuyTsf

2

fllstw

0.21 0.10 Na1.99 0.92 –

genetic approach that allowed rare recessive mutations thatccumulated over several generations to be detected. While heucceeded in experimentally calculating a spontaneous mutationrequency, even under apparently controlled laboratory conditions,epeated experimentation resulted in frequencies that varied by anrder of magnitude due to unknown causes. Nevertheless, in lightf the contemporary state of the art, Muller’s work represented aajor technical and scientific breakthrough that he rightly believedould (in his own words) “open up a new field of genetics—the

uantitative study of gene mutation, as occurring throughout oner more entire chromosomes under purposely varied conditions.”n the decades since the publication of this study the pursuit of accu-ate spontaneous mutation rates has blossomed into an important,ulti-faceted, and sometimes controversial, area of research.Many techniques have been devised to increase the accuracy

nd reproducibility of spontaneous mutation rate measurementsn several model organisms. Because of early advances in the devel-pment of Escherichia coli as a model system for molecular geneticnalysis, it became a favorite for studying mutagenesis and hasielded the richest collection of literature on the topic. The pri-ary aim of this review is to summarize several notable and related

ecent studies from three different laboratories that provide newstimates for the E. coli mutation rate using current techniques1–3]. Two of these studies, both from Patricia Foster’s laboratory,ave revealed especially compelling insight into the mutationalopology of the E. coli chromosome and are discussed in furtheretail. To put these new studies in context, the review begins with

brief update on the standard repertoire of techniques for measur-ng mutation rates in E. coli, especially focusing on the strengths and

eaknesses of each technique and their underlying assumptions.

. Techniques for mutation rate analysis

Three approaches for measuring spontaneous mutation ratesave taken particularly strong footholds in the literature: fluctua-ion analysis, mutation accumulation experiments, and, to a lesserxtent, comparative genomics. All of these approaches have beensed to measure spontaneous mutation rates in E. coli, but haveielded values that vary by more than an order of magnitude (seeable 1 and discussions in [1,4–6]). To begin to understand pos-ible reasons for this wide variation, the technical and theoreticalrameworks supporting each approach must be appreciated.

.1. Estimation of mutation rates by fluctuation analysis

Salvador Luria and Max Delbrück reported results from theiructuation experiments on the nature of mutagenesis in 1943 [7],

Please cite this article in press as: A.B. Williams, Spontaneous mutathttp://dx.doi.org/10.1016/j.dnarep.2014.09.009

aying not only an essential conceptual foundation for our under-tanding of mutagenesis in general, but also providing an effectiveechnical methodology for estimating mutation rates. When usedith mathematical approaches developed soon after by Lea and

c acid resistance Fluctuation analysis [3]Mutation accumulation with WGS [3]

Coulson [8], a revolutionary feature of the Luria–Delbrück fluctu-ation experiment was that it could be used to calculate mutationrates based on extremely rare mutational events, thus providing amajor leg up in overcoming this major challenge noted by Muller15 years earlier.

While fluctuation experiments can be very laborious in practice,the basic experimental design is quite simple [9]. In a typical exper-iment, a small number of cells are inoculated into many parallelliquid cultures that are then grown to saturation. Under opti-mal experimental conditions, Kendal and Frost [10] estimated thatabout 40 cultures are sufficient (see [11] for further discussion ofthis point). Next, the cultures are plated on appropriate selectivemedia to detect mutants that arose during the culturing and a fewcultures are plated on non-selective medium to determine finalnumber of cells. At the end of the experiment, the number of mutantcells in each culture depends on both the mutation frequency andthe timing of when the original mutational event occurred duringthe growth of the culture—cultures in which a mutation occurredearlier will contain many more mutant cells than those with later amutational event. The mutation rate can then be calculated by ana-lyzing the distribution of the final number of mutant cells presentin the different cultures.

The important feature of the Lea–Coulson method is that itcan distinguish the number of mutational events from the num-ber of mutants detected at the end of the experiment; however,the applicability of the underlying mathematical theories requiresthat several critical assumptions about the experimental conditionshold true:

i. The cells must grow exponentially.ii. The mutation rate must be constant during the lifetime of a

cell.iii. A mutational event cannot affect the mutation rate.iv. The growth rate of mutant cells must be the same as non-

mutant cells.v. Mutant cells must not die during the course of the experiment.

vi. Forward mutations must be the dominant mutational eventand the rate of reversion mutations must be so low that it isnegligible.

vii. Once the cells have been plated on a selective medium, no newmutations can arise. Although pathways leading to increasedmutation rates under non-lethal selection have been described(reviewed in [4–6,12]), lethal selection methods can be usedto prevent even weak proliferation on the selective media tominimize this potential problem.

viii. In experiments using batch cultures, the number of the cells inthe initial inoculum must be negligible compared to the num-

ion rates come into focus in Escherichia coli, DNA Repair (2014),

ber of cells at the end of culturing. The mathematical methodsused to calculate the mutation rate are valid only when the ini-tial number of cells is no more than 1/1000 of the final numberof cells [7,13].

ING ModelD

Repair

srdcycm(e

2

mussrgbesagp

i

i

tcpfafsEmtgomsgit00r

gortl

ARTICLENAREP-2004; No. of Pages 7

A.B. Williams / DNA

As the above list illustrates, fluctuation analysis is highlyensitive to uncontrolled variables and reproducible mutationates can only be calculated from experiments that are properlyesigned to support the best possible adherence to these theoreti-al assumptions. Even if fluctuation experiments do not necessarilyield extremely accurate estimates of mutation rates, carefullyontrolled experiments can give very precise, i.e. reproducible,utation rates. In this way, classical fluctuation analysis has been

and still remains) an extremely valuable tool for examining theffects of many variables on mutagenesis.

.2. Estimation of mutation rates by comparative genomics

The increased accessibility of DNA sequencing and the amass-ent of sequence data has made it possible to study mutation rates

sing comparative genomics—that is through comparison of DNAequence from organisms that diverged on an evolutionary timecale (typically millions of years). In cases where tractable labo-atory models are unavailable—for example in studies of humanenome evolution and cancer biology—comparative genomics haseen invaluable (for some specific examples, see [8,14–16]). How-ver, this approach has also been very informative when applied toimpler tractable organisms such as bacteria and yeast. Like fluctu-tion analysis, determination of mutation rates using comparativeenomics also depends on the validity of several assumptions, inarticular:

i. A value for the cellular generation time that is reasonably appli-cable on an evolutionary time scale must be estimated.

ii. The estimated divergence time of the organisms being com-pared is reasonably accurate.

ii. Synonymous mutations are selectively neutral over the courseof evolutionary time. How well this assumption holds true innature is questioned (see [6,9] and references therein), but syn-onymous mutations, which by definition do not alter amino acidsequences, should be more resistant to selection.

v. Codon usage and GC-content are not subject to selective pres-sure.

Given these underlying assumptions, the calculation of muta-ion rates using comparative genomics is, at least superficially,onceptually simple. A significant 1999 study by Howard Ochman,robably the main champion of comparative genomic techniquesor the study of E. coli evolution and mutagenesis, provides

constructive case study [10,17] (for other important workrom Ochman, see [18]). Using phylogenetic comparison of DNAequence from the closely related species Salmonella enterica and. coli, their divergence rate has been calculated as 0.90% perillion years [19,20]—this value represents how much variation

he genomes had to accumulate per unit time since their diver-ence from a common ancestor to yield the respective genomesf today. Based on this divergence rate, Ochman et al. [17] esti-ated that E. coli has a mutation rate of 0.0045 mutations per

ite per million years. Relying on data from earlier studies on therowth kinetics of modern commensal E. coli, they estimated thatn nature, E. coli undergoes 100–300 generations per year. Usinghese values, they then calculated a mutation rate of 0.1 × 10−3 to.2 × 10−3 mutations per genome per generation, or a rate of about.45 × 10−10 mutations per nucleotide per generation—the lowesteported mutation rate for E. coli to date (see Table 1).

Why is this mutation rate determined using comparativeenomics so low compared to the mutation rates calculated by

Please cite this article in press as: A.B. Williams, Spontaneous mutathttp://dx.doi.org/10.1016/j.dnarep.2014.09.009

ther methods? One very interesting possibility is that the variationeflects a true difference in the mutation rates between labora-ory strains and natural populations. It could be that a commensalifestyle, as in the digestive tract of animals, promotes genome

PRESS xxx (2014) xxx–xxx 3

stability. This hypothesis is supported by the recent report of a“mutation rate plasticity” that depends on the social environment:more crowded cells have lower mutation rates than isolated cells[21]. A less interesting explanation is that this mutation rate wasinfluenced by weak adherence to the underlying assumptions innature or during evolution, or by shortcomings in the suppor-ting computational methods, which continue to be improved andrefined (for further discussion of possible explanations, see [18]).Certainly the later possibility does not diminish the quality or valueof Ochman’s work. In fact, in light of these questions, perhapsit is remarkable that the mutation rate obtained by compara-tive genomics is so close to rates from more direct experimentalapproaches. Whether or not the true basis of these disparate valuesis ever dissected (which may be facilitated by new approaches dis-cussed below), what can be gathered from these data is that there isa robustness in the experimentally determined values for sponta-neous mutation rates, even based on extremely different methods,supporting the credibility of current estimates.

2.3. Estimation of mutation rates using mutation accumulationexperiments

Ongoing advances in whole-genome sequencing have providedanother tool for measuring spontaneous mutation rates and, tothis end, this technology has been used effectively with the well-established mutation accumulation (MA) approach [22,23]. Unlikefluctuation tests and comparative genomic approaches, which relyon computational methods to extrapolate whole-genome muta-tion rates from either locus-specific measurements or phylogeneticdata, the great strength of MA and whole-genome sequencingis that the entire genome is surveyed, to the base pair level, atthe same time. Because of this ‘direct detection’ approach, thesetypes of experiments are liberated from the sometimes unrealis-tic assumptions underlying fluctuation analysis and comparativegenomics. From the number of mutations recovered, and given agood estimate of the number of generations accrued in the exper-iment, the mutation rate of a lineage can be calculated using justa few simple operations. A truly accurate spontaneous mutationrate would incorporate the contributions of all types of unbiasedmutations (i.e. mutations not influenced by selection). This condi-tion cannot be met using single loci approaches, which typicallydetect only one type of mutation. For example, a commonly usedmarker in fluctuation analysis is forward mutation from nalidixicacid sensitivity to nalidixic acid resistance via a handful of inde-pendent base substitution mutations in the gyrA and gyrB genes[24–27]. Thus, whole-genome mutation rates extrapolated fromfluctuation data using this phenotypic marker are based solely onthe rate of a few specific non-synonymous base substitution muta-tions. While other markers can be used (for example mutations inthe rpoB gene that confer resistance to rifampicin [28,29]), theysuffer from this same weakness. This is not the only problem withsingle loci approaches. The position of the relevant gene and themethylation state of the local DNA can significantly affect muta-bility, thus skewing the resulting mutation rates (for example, see[3] discussed below and [30]). Furthermore, markers that dependon forward mutations to confer drug resistance may suffer fromphenotypic lag [31], a requirement that several generations passbefore the drug-senstitive molecules are replaced by the mutantdrug-resistant versions. Phenotypic lag would decrease the mea-sured mutation rate (see Table 1: mutation rates to rifampicinand nalixidic acid resistance are lower than experiments that usethe Lac and His metabolic markers). Whole-genome sequencing

ion rates come into focus in Escherichia coli, DNA Repair (2014),

of MA lines avoids these problems since all classes of mutationsare detected simultaneously over the entire genome, includingsingle nucleotide events (base substitutions, deletions, insertions)and larger scale events (long insertions and deletions, duplications,

ING ModelD

4 Repair

ibon

cc[sotdE

2e

auewrtsostbbrtsetiibtdtwnmmFmlv

vmttoetDutto

2b

f

ARTICLENAREP-2004; No. of Pages 7

A.B. Williams / DNA

nversions, among others). Thus, whole-genome mutation rates cane calculated based on any specific type of mutation via directbservation of the mutations—thus reducing or eliminating theeed for extrapolation.

In 2008 Michael Lynch and colleagues reported the first suc-essful application of whole-genome sequencing of MA lines foralculating spontaneous mutation rates in Saccharomyces cerevisiae22] and a new level of accuracy in spontaneous mutation rate mea-urements was attained. Methods for whole-genome sequencingf MA lines have been developed in a number of model systemso address a variety of experimental questions, but the followingiscussion focuses specifically on two prominent MA studies in. coli.

.3.1. Estimation of mutation rates from long-term evolutionxperiments

The complex interactions between mutagenesis, competition,nd evolution have been addressed experimentally for some timesing long-term evolution experiments (for just a few recentxamples, see [32–34]). When this approach is reinforced withhole-genome sequencing, an ancillary benefit is that mutation

ates can be calculated from the resulting sequence data. During aypical long-term evolution experiment, MA lines are produced byerially culturing bacteria over many generations via the transferf small samples from one culture to the next. At the conclu-ion of the experiment, the genomes of the lines are sequencedo identify the accumulated mutations. Mutation rates can thene calculated based on the number of generations and the num-er of mutations. A recent study by Wielgoss and colleagues [2]eported a successful application of this approach. Since the muta-ion rates of their passaged lines were similar to that of the foundertrain, any accumulated mutations should reflect spontaneousvents unaffected by any acquired genome instability (unless aransient mutator state occurred during the culturing). Interest-ngly, but not particularly surprising, analysis of the mutationsn their lines revealed that, in every case, the ratio of the num-er of non-synonymous substitutions per non-synonymous site tohe number of synonymous substitutions per synonymous site (theN/dS ratio) was greater than one, suggesting the presence of selec-ion during the long-term culturing. Because effects of this selectionould skew a mutation rate calculated using both synonymous andon-synonymous mutations, Wielgoss et al. used only synonymousutations. After 300,000 cumulative generations, 52 synonymousutations were recovered from 35 independent mutational events.

rom these data, Wielgoss et al. calculated a rate of 0.89 × 10−10

utations per base pair per generation, a value between the ear-ier reported rates of Drake and Ochman, but much lower than thealue reported by Lee et al. [3] (see Table 1 and discussion below).

Despite this inconsistency, Wielgoss et al. claimed that this newalue for the spontaneous base substitution rate in E. coli was theost accurate at that time. Considering the meticulous experimen-

al design and the statistical power afforded by their approach,his may have been a reasonable assertion; however, this claimnly holds true when, over the course of their long-term evolutionxperiment, synonymous mutations really were selectively neu-ral. The validity of this assumption came under scrutiny by Johnrake based on previous literature that showed strong bias in codonsage in E. coli, implying that it is actually subject to strong selec-ion ([6] and references therein). For this reason, Drake argues thathe mutation rate calculated by Weilgoss et al. is depressed becausef underappreciated selection during the long-term culturing.

Please cite this article in press as: A.B. Williams, Spontaneous mutathttp://dx.doi.org/10.1016/j.dnarep.2014.09.009

.3.2. Estimation of mutation rates using MA lines withottlenecks

If the mutation rate calculated by Wielgoss et al. was, inact, underestimated due to selection, could selection in the

i

PRESS xxx (2014) xxx–xxx

experimental system be reduced to a negligible level? One com-monly used and well-characterized approach is to repeatedlyreduce the population size carried through a serial passage in a MAexperiment to a single individual [35–38]. Such ‘bottlenecks’ allowdeleterious mutations to accumulate (as long as they are not lethal)by rescuing the individual at the bottleneck from competition withcells with potentially more favorable genetic backgrounds. This tac-tic increases the probability that deleterious mutations are fixed bygenetic drift as efficiently as neutral mutations, thus reducing oreliminating the effects of selection.

The first MA experiment integrating bottlenecks to measurethe spontaneous mutation rate in E. coli was performed in Patri-cia Foster’s laboratory and reported in the previously cited studyby Lee et al. [3]. The MA experiment was performed as follows:each day, rather than directly transferring a sample of the liquidculture, cells from an overnight culture were streaked for singlecolonies. For the subsequent passage, a line was drawn throughthe middle of plate and the colony closest to the line was cho-sen and inoculated into fresh medium—simultaneously eliminatingboth human bias in selecting the colony and introducing a geneticbottleneck. Because a bacterial colony is clonal, the effective popu-lation size at each passage was one. The number of generations thatgave rise to each chosen colony was estimated based on empir-ical data, thus increasing the reliability of the value for the totalnumber of accumulated generations in the MA lines. At the endof the experiment, the genomes of two sets of MA lines weresequenced—one with 117,040 total generations and the other with133,476 total generations. Based on the total number of mutations,including base pair substitutions and small insertions and deletions(less than four nucleotide), two mutation rates were calculated:1.88 × 10−10 and 2.45 × 10−10 mutations per nucleotide per gener-ation. Even though these values differ by 30%, since they are within95% confidence limits Lee et al. claim that the midpoint of thesevalues, 2.2 × 10−10, represents the most accurate estimate of themutation rate. When the mutation rate was calculated using onlysynonymous base pair substitutions, the value decreased slightlyto 1.99 × 10−10 base pair substitutions per nucleotide per genera-tion. As discussed above, Wielgoss et al. reported a mutation rate of8.9 × 10−11 synonymous base pair substitutions per nucleotide pergeneration, nearly three times lower than that determined by Leeet al.

When confronted with such a discrepancy between the muta-tion rates calculated using two similar methods, one must ask: isit possible to objectively evaluate the different methods to decidewhether one value is likely to be more accurate than the other? It isreasonable to suppose that the experiment of Lee et al., which inte-grated genetic bottlenecks, would yield the most accurate mutationrate since it was subject to the least influence of selection. But didLee et al. provide convincing evidence that the bottlenecks actuallydid eliminate selection? At least four pieces of evidence from theiranalysis argued in their favor:

i. Stronger selection on non-synonymous mutations than onsynonymous mutations would cause a difference betweenthe experimentally obtained ratio of synonymous to non-synonymous mutations and the ratio expected due only tostochastic events. Lee et al. detected no significant differencein these ratios.

ii. Differences in selective pressure between synonymous muta-tions and non-synonymous mutations should result in dif-ferences between the mutation rate calculated using onlysynonymous base pair substitutions and one calculated using

ion rates come into focus in Escherichia coli, DNA Repair (2014),

all base pair substitutions. While these rates were different, thedifference was small.

ii. As Drake pointed out, under competitive conditions, E. colimutants are subject to selection based on codon usage. Such

IN PRESSG ModelD

Repair xxx (2014) xxx–xxx 5

i

Maaf

3c

mltmMimDdir(cpcpeale

1iWaarttgDttrac

Fig. 1. The mutational topology of the E. coli chromosome. As the two replicationforks formed from the opening of the replication bubble at the oriC locus movethrough the two replichores (delineated by the diagonal dashed line) toward thetermination region, they encounter chromosomal features that cause a symmetrical

ARTICLENAREP-2004; No. of Pages 7

A.B. Williams / DNA

selection would be registered by a skew toward mutations thatlead to more commonly used codons compared to mutationsthat result in less commonly used codons. In fact, Lee et al. sawthat most synonymous base pair substitutions in their MA linesactually resulted in less commonly used codons, arguing againstthis possibility.

v. Nonsense mutations in coding sequences, which often com-pletely inactivate genes, could result in mutant cells withdecreased fitness that could be selected against during cultur-ing. Such selection would result in fewer nonsense mutationsin the coding sequences of the MA lines compared to the statis-tically predicted number in the total absence of selection. Overthe entire E. coli genome, 3% of all base pair substitutions areexpected to be nonsense mutations. This value can be used topredict the expected fraction of the mutations in the MA linesthat would result in chain-terminating codons. Lee et al. foundthat the predicted number and the actual number differed byonly one.

Taken together, these arguments strongly suggest that in theA experiment performed by Lee et al., selection was likely not

confounding factor. Therefore, it is reasonable to agree with theuthors that this newly calculated spontaneous mutation rate is, inact, the most accurate to date.

. New insight into the mutational topology of the E. colihromosome

While the highly accurate spontaneous mutation rate in E. colieasured by Lee et al. is a welcome addition to the mutagenesis

iterature, perhaps their most novel, and certainly most provoca-ive, findings came from their subsequent analysis of spontaneous

utagenesis in a mismatch repair (MMR) defective E. coli strain.MR suppresses the accumulation of spontaneous mutations as

t surveys newly synthesized DNA to identify and repair mis-atched bases incorporated during DNA replication that escapeNA polymerase proofreading (reviewed in [39]). Previous workemonstrated that loss of MMR causes a 100- to 200-fold increase

n spontaneous mutagenesis [39] and Lee et al. report a compa-able 138-fold increase in an MMR-defective mutL mutant strain2.75 × 10−8 base pair substitutions per nucleotide per generationompared to 2.2 × 10−10 base pair substitutions per nucleotideer generation in the wild type strain). The sequence data alsolarified fine details of spontaneous mutation in the absence ofost-replicative repair, the details of which should be very inter-sting to the more serious students of spontaneous mutagenesisnd MMR; however, perhaps the most generally interesting reve-ation came when the positional distribution of the mutations wasxamined.

In a follow-up study to Lee et al., Foster et al. [40] plotted the625 base pair substitutions identified by whole genome sequenc-

ng of 34 parallel MMR-defective MA lines onto the chromosome.hen the mutations were grouped into bins, each consisting of

bout 100 kb of the 4.6 million base pair chromosome, a remark-ble pattern emerged (Fig. 1): not only were the mutations notandomly distributed around the chromosome, the spatial muta-ion densities were symmetrical in the two replichores (defined inhe figure by the dashed line passing through the replication ori-in, oriC, and the termination region containing the TerA, B, C, and

loci). The regions immediately adjacent to the origin, slightly lesshan half way through each replichore, and on either side of the

Please cite this article in press as: A.B. Williams, Spontaneous mutathttp://dx.doi.org/10.1016/j.dnarep.2014.09.009

ermination regions, were most prone to mutagenesis (shown ined in Fig. 1). Foster et al. subjected their data to rigorous statisticalnd computational analyses and demonstrated that stochastic pro-esses could not explain the symmetrical mutation distributions;

spatial variation in the mutability of the DNA, shown here by changes in color fromgreen (low mutability) to red (high mutability).

thus this striking mutational topology likely resulted from specialbiological properties of the chromosome.

The symmetry in the mutational density suggests that thereare large-scale, but spatially separated, similarities between cor-responding regions of the two replichores, despite their uniqueDNA sequences. The simplest interpretation is that as the two repli-cation forks move in opposite directions from the origin to thetermination region they experience symmetrical features of thechromosome that profoundly affect the local mutability of the DNA.In an attempt to produce a predictive model for the mutationaltopology that integrates known characteristics of the chromosome,Foster et al. considered many features that could, by virtue of theirown spatial variation around the chromosome, contribute to thisphenomenon. These features included (among others): the degreeof gene expression; AT content; codon usage bias; genes regulatedby nucleoid-associated proteins (HU, FIS, H-NS) or DNA gyrase; andgenes whose expression is sensitive to DNA topology. Numerousstudies have shown that highly expressed genes are more proneto mutagenesis [41–45]; thus, a reasonable hypothesis was thatthe mutational topology could reflect differences in the densityof highly expressed genes on the chromosome. Although nearlycontemporaneous work in Nicholas Luscombe’s lab, reported inMartincorena et al. [46], demonstrated a correlation between muta-bility and gene expression, Foster et al. observed no such correlation(although it should be noted that this conclusion of Martincorenaet al. has been recently challenged [47]). Of the remaining fea-tures, a positive correlation was observed between regions ofincreased mutability and regions enriched for genes activated byHU and repressed by FIS, two highly abundant nucleoid-associatedproteins. In fact, statistical models predicted that HU- and FIS-associated features could explain about a third of the variation inthe mutational distribution. From this result, Foster et al. proposedthat highly complex regions of the chromosome might be moresusceptible to mutation, a reasonable hypothesis when consider-

ion rates come into focus in Escherichia coli, DNA Repair (2014),

ing the challenges a replication fork may encounter as it passesthrough DNA highly decorated with DNA binding proteins such asHU or FIS. Furthermore, HU has been previously shown to have

ING ModelD

6 Repair

aaicttspdmmim

tgosdttwbtdiobmtg

4

tetTTwiatpidueatnmsBsp

etErtso

[

[

[

[

[

[

[

[

[

[

ARTICLENAREP-2004; No. of Pages 7

A.B. Williams / DNA

pro-mutagenic role [REF]. When three more features from thebove list were added to the model (now incorporating five factors),ts predictive power increased to only about 44%; thus it becamelear that additional, yet unknown, complexity influences the sys-em. Foster et al. discuss some possibilities, including variation inhe activities of different DNA repair pathways around the chromo-ome, unequal accessibility of the chromosome to error-prone DNAolymerases, and possible decreases in DNA replication fidelityue to unknown processes around the termination sites. Could thisutational symmetry also reflect a not yet fully understood chro-osomal superstructure (Kuzminov; Gahlmen)? Additional work

s certainly needed, and likely underway, to understand the detailedechanism of this new observation.In the meantime, an important question can be considered: has

his mutational topology influenced the evolution of the E. colienome? If so, then it should have left footprints on the genomesf divergent strains. At least two studies already cited above offerome support for this prediction. Earlier work by Sharp et al. [30]emonstrated that the base substitution rate in genes proximal tohe replication origin was about half that of genes closer to theermination region. More recent analysis of the sequence diversityithin 2659 genes from 34 evolutionarily divergent E. coli strains

y Martincorena et al. revealed increased sequence divergence nearhe terminus. Whether or not the spatial variation in sequenceiversity described by Martincorena et al. or the positional effect

dentified by Sharp et al. are direct evolutionary consequencesf the mutational topology uncovered by Foster et al. may nevere known; nevertheless, the consistent observations of increasedutability around the terminus region support the possibility that

he mutational topology has helped to shape the modern E. colienome.

. Conclusions

The mutational topology of DNA has been of interest for morehan 50 years, notably spotlighted by Seymour Benzer’s detailedxploration of the topic in his 1961 publication “On the topology ofhe genetic fine structure” [48]. Benzer was studying mutations in4 phage that limited its ability to grow on its obligate host, E. coli.hrough laborious crosses and genetic analysis, Benzer produced,hat he called, a ‘topographic map’ of spontaneous mutagenesis

n a segment of the phage genome. As noted by Foster et al., theccessibility of whole-genome sequence has allowed researcherso expand Benzer’s work over entire genomes, on a scale that wouldrobably have been difficult to fathom in the 1960s; however, this

s not to say that these new methods have eliminated or invali-ated the need for older techniques. For example, in an experimentsing whole-genome sequencing of MA lines, each base position isffectively analyzed one time. For this reason, and because the MApproach is too laborious to allow extensive experimental replica-ion, it is not useful for analyzing mutational hotspots on the singleucleotide level (as in Benzer’s earlier work on the fine structure ofutagenesis)—instead it allows us a unique opportunity to under-

tand mutagenesis on a larger scale on the entire chromosome.ecause of these limitations, it is clear that a complete picture ofpontaneous mutagenesis can only be obtained via the combinedower of multiple techniques—old and new.

As with all effective science, the new data discussed here havexposed many additional questions. It would be very interestingo apply the experimental pipeline developed by Lee et al. to other. coli strains. For example, comprehensive analysis of other DNA

Please cite this article in press as: A.B. Williams, Spontaneous mutathttp://dx.doi.org/10.1016/j.dnarep.2014.09.009

epair mutants could provide new information on the relative con-ributions of different DNA repair pathways in maintaining genometability through evolution. Analysis of the mutational topologyf commensal strains, or of cells grown in different media, could

[

[

PRESS xxx (2014) xxx–xxx

help reconcile the differences in the mutation rates determined bydifferent methods (especially comparative genomics) and providemore insight into the mutation rate plasticity described by Krasovecet al. noted above. It is clear that much remains to be learned aboutmutagenesis and how it shapes evolution—interesting and surpris-ing results continue to be reported, and arguments are certain topersist. The early genetic methods established by Muller openedthe door to a highly productive period of innovative mutagenesisresearch. Now, nearly nine decades later, application of whole-genome sequencing in E. coli and higher organisms seems to bespawning a modern renewal in the field that is certain to providenew clarity to the complexities of this extremely important biolog-ical process.

Conflict of interest statement

None declared.

Acknowledgement

I am grateful for critical reading and comments on themanuscript by Dr. Björn Schumacher.

References

[1] H.J. Muller, The measurement of gene mutation rate in Drosophila, its highvariability, and its dependence upon temperature, Genetics 13 (1928) 279–357.

[2] S. Wielgoss, J.E. Barrick, O. Tenaillon, S. Cruveiller, B. Chane-Woon-Ming, C.Médigue, et al., Mutation rate inferred from synonymous substitutions in along-term evolution experiment with Escherichia coli, G3 (Bethesda) 1 (2011)183–186, http://dx.doi.org/10.1534/g3.111.000406.

[3] H. Lee, E. Popodi, H. Tang, P.L. Foster, Rate and molecular spectrum ofspontaneous mutations in the bacterium Escherichia coli as determinedby whole-genome sequencing, Proc. Natl. Acad. Sci. U. S. A. 109 (2012)E2774–E2783, http://dx.doi.org/10.1073/pnas.1210309109.

[4] J.W. Drake, A constant rate of spontaneous mutation in DNA-based microbes,Proc. Natl. Acad. Sci. U. S. A. 88 (1991) 7160–7164.

[5] J.W. Drake, B. Charlesworth, D. Charlesworth, J.F. Crow, Rates of spontaneousmutation, Genetics 148 (1998) 1667–1686.

[6] J.W. Drake, Contrasting mutation rates from specific-locus and long-term mutation-accumulation procedures, G3 (Bethesda) 2 (2012) 483–485,http://dx.doi.org/10.1534/g3.111.001842.

[7] S.E. Luria, M. Delbrück, Mutations of bacteria from virus sensitivity to virusresistance, Genetics 28 (1943) 491–511.

[8] D.E. Lea, C.A. Coulson, The distribution of the numbers of mutants in bacterialpopulations, J. Genet. 49 (1949) 264–285.

[9] P.L. Foster, Methods for determining spontaneous mutation rates, Meth-ods Enzymol. 409 (2006) 195–213, http://dx.doi.org/10.1016/S0076-6879(05)09012-9.

10] W.S. Kendal, P. Frost, Pitfalls and practice of Luria–Delbrück fluctuation analy-sis: a review, Cancer Res. 48 (1988) 1060–1065.

11] M.E. Jones, S.M. Thomas, A. Rogers, Luria–Delbrück fluctuation experiments:design and analysis, Genetics 136 (1994) 1209–1216.

12] A.B. Williams, P.L. Foster, Stress-induced mutagenesis, EcoSal (2012),http://dx.doi.org/10.1128/ecosalplus.7.2.3.

13] S. Sarkar, W.T. Ma, G.H. Sandri, On fluctuation analysis: a new, simple and effi-cient method for computing the expected number of mutants, Genetica 85(1992) 173–179.

14] M.W. Nachman, S.L. Crowell, Estimate of the mutation rate per nucleotide inhumans, Genetics 156 (2000) 297–304.

15] A. Hodgkinson, A. Eyre-Walker, Variation in the mutation rate across mam-malian genomes, Nat. Rev. Genet. 12 (2011) 756–766, http://dx.doi.org/10.1038/nrg3098.

16] I.R. Watson, K. Takahashi, P.A. Futreal, L. Chin, Emerging patterns of somaticmutations in cancer, Nat. Rev. Genet. 14 (2013) 703–718, http://dx.doi.org/10.1038/nrg3539.

17] H. Ochman, S. Elwyn, N.A. Moran, Calibrating bacterial evolution, Proc. Natl.Acad. Sci. U. S. A. 96 (1999) 12638–12643.

18] H. Ochman, Neutral mutations and neutral substitutions in bacterialgenomes, Mol. Biol. Evol. 20 (2003) 2091–2096, http://dx.doi.org/10.1093/molbev/msg229.

19] P.M. Sharp, Determinants of DNA sequence divergence between Escherichia coliand Salmonella typhimurium: codon usage, map position, and concerted evolu-

ion rates come into focus in Escherichia coli, DNA Repair (2014),

tion, J. Mol. Evol. 33 (1991) 23–33.20] J.G. Lawrence, H. Ochman, Amelioration of bacterial genomes: rates of change

and exchange, J. Mol. Evol. 44 (1997) 383–397.21] R. Krasovec, R.V. Belavkin, J.A.D. Aston, A. Channon, E. Aston, B.M.

Rash, et al., Mutation rate plasticity in rifampicin resistance depends

ING ModelD

Repair

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[48] S. Benzer, On the topology of the genetic fine structure, Proc. Natl. Acad. Sci. U.

ARTICLENAREP-2004; No. of Pages 7

A.B. Williams / DNA

on Escherichia coli cell–cell interactions, Nat. Commun. 5 (2014) 3742,http://dx.doi.org/10.1038/ncomms4742.

22] M. Lynch, W. Sung, K. Morris, N. Coffey, C.R. Landry, E.B. Dopman, et al.,A genome-wide view of the spectrum of spontaneous mutations in yeast,Proc. Natl. Acad. Sci. U. S. A. 105 (2008) 9272–9277, http://dx.doi.org/10.1073/pnas.0803466105.

23] D.L. Halligan, P.D. Keightley, Spontaneous mutation accumulation studiesin evolutionary genetics, Annu. Rev. Ecol. Evol. Syst. 40 (2009) 151–172,http://dx.doi.org/10.1146/annurev.ecolsys.39.110707.173437.

24] J. Ruiz, D. Castro, P. Goni, J.A. Santamaria, J.J. Borrego, J. Vila, Analysis of themechanism of quinolone resistance in nalidixic acid-resistant clinical isolatesof Salmonella serotype Typhimurium, J. Med. Microbiol. 46 (1997) 623–628.

25] M.P. Reche, J.E. García de los Ríos, P.A. Jiménez, A.M. Rojas, R. Rotger, gyrAMutations associated with nalidixic acid-resistant salmonellae from wild birds,Antimicrob. Agents Chemother. 46 (2002) 3108–3109.

26] Y. Sáenz, M. Zarazaga, L. Brinas, F. Ruiz-Larrea, C. Torres, Mutations in gyrA andparC genes in nalidixic acid-resistant Escherichia coli strains from food prod-ucts, humans and animals, J. Antimicrob. Chemother. 51 (2003) 1001–1005,http://dx.doi.org/10.1093/jac/dkg168.

27] E. Becket, L. Tse, M. Yung, A. Cosico, J.H. Miller, Polynucleotide phos-phorylase plays an important role in the generation of spontaneousmutations in Escherichia coli, J. Bacteriol. 194 (2012) 5613–5620, http://dx.doi.org/10.1128/JB. 00962-12.

28] D.J. Jin, C.A. Gross, Mapping and sequencing of mutations in the Escherichia colirpoB gene that lead to rifampicin resistance, J. Mol. Biol. 202 (1988) 45–58.

29] L. Garibyan, T. Huang, M. Kim, E. Wolff, A. Nguyen, T. Nguyen, et al., Use of therpoB gene to determine the specificity of base substitution mutations on theEscherichia coli chromosome, DNA Repair. 2 (2003) 593–608.

30] P.M. Sharp, D.C. Shields, K.H. Wolfe, W.H. Li, Chromosomal location and evolu-tionary rate variation in enterobacterial genes, Science 246 (1989) 808–810.

31] P. Armitage, The statistical theory of bacterial populations subject to mutation,J. R. Stat. Soc. 14 (1952) 1–40.

32] N. Leiby, W.R. Harcombe, C.J. Marx, Multiple long-term, experimentally-evolved populations of Escherichia coli acquire dependence upon citrate as aniron chelator for optimal growth on glucose, BMC Evol. Biol. 12 (2012) 151,http://dx.doi.org/10.1186/1471-2148-12-151.

33] M.T. Stanek, T.F. Cooper, R.E. Lenski, Identification and dynamics of a beneficial

Please cite this article in press as: A.B. Williams, Spontaneous mutathttp://dx.doi.org/10.1016/j.dnarep.2014.09.009

mutation in a long-term evolution experiment with Escherichia coli, BMC Evol.Biol. 9 (2009) 302, http://dx.doi.org/10.1186/1471-2148-9-302.

34] V.S. Cooper, The origins of specialization: insights from bacteria held 25years in captivity, PLoS Biol. 12 (2014) e1001790, http://dx.doi.org/10.1371/journal.pbio.1001790.

[

PRESS xxx (2014) xxx–xxx 7

35] D.R. Denver, K. Morris, M. Lynch, L.L. Vassilieva, W.K. Thomas, High directestimate of the mutation rate in the mitochondrial genome of Caenorhabditiselegans, Science 289 (2000) 2342–2344.

36] D.R. Denver, K. Morris, M. Lynch, W.K. Thomas, High mutation rate and pre-dominance of insertions in the Caenorhabditis elegans nuclear genome, Nature430 (2004) 679–682, http://dx.doi.org/10.1038/nature02697.

37] D.R. Denver, K. Morris, A. Kewalramani, K.E. Harris, A. Chow, S. Estes, et al.,Abundance, distribution, and mutation rates of homopolymeric nucleotideruns in the genome of Caenorhabditis elegans, J. Mol. Evol. 58 (2004) 584–595,http://dx.doi.org/10.1007/s00239-004-2580-4.

38] C. Haag-Liautard, M. Dorris, X. Maside, S. Macaskill, D.L. Halligan, D. Houle, et al.,Direct estimation of per nucleotide and genomic deleterious mutation rates inDrosophila, Nature 445 (2007) 82–85, http://dx.doi.org/10.1038/nature05388.

39] M.G. Marinus, DNA methylation and mutator genes in Escherichia coli K-12,Mutat. Res. 705 (2010) 71–76, http://dx.doi.org/10.1016/j.mrrev.2010.05.001.

40] P.L. Foster, A.J. Hanson, H. Lee, E.M. Popodi, H. Tang, On the muta-tional topology of the bacterial genome, G3 (Bethesda) 3 (2013) 399–407,http://dx.doi.org/10.1534/g3.112.005355.

41] N. Kim, S. Jinks-Robertson, Transcription as a source of genome instability, Nat.Rev. Genet. 13 (2012) 204–214, http://dx.doi.org/10.1038/nrg3152.

42] A. Aguilera, The connection between transcription and genomic instability,EMBO J. 21 (2002) 195–201, http://dx.doi.org/10.1093/emboj/21.3.195.

43] J.Q. Svejstrup, Mechanisms of transcription-coupled DNA repair, Nat. Rev. Mol.Cell Biol. 3 (2002) 21–29, http://dx.doi.org/10.1038/nrm703.

44] P.C. Hanawalt, G. Spivak, Transcription-coupled DNA repair: two decadesof progress and surprises, Nat. Rev. Mol. Cell Biol. 9 (2008) 958–970,http://dx.doi.org/10.1038/nrm2549.

45] C. Park, W. Qian, J. Zhang, Genomic evidence for elevated mutation ratesin highly expressed genes, EMBO Rep. 13 (2012) 1123–1129, http://dx.doi.org/10.1038/embor.2012.165.

46] I. Martincorena, A.S.N. Seshasayee, N.M. Luscombe, Evidence of non-randommutation rates suggests an evolutionary risk management strategy, Nature 485(2012) 95–98, http://dx.doi.org/10.1038/nature10995.

47] X. Chen, J. Zhang, No gene-specific optimization of mutation rate inEscherichia coli, Mol. Biol. Evol. 30 (2013) 1559–1562, http://dx.doi.org/10.1093/molbev/mst060.

ion rates come into focus in Escherichia coli, DNA Repair (2014),

S. A. 45 (1959) 1607–1620.49] J.W. Drake, Avoiding dangerous missense: thermophiles display especially low

mutation rates, PLoS Genet. 5 (2009) e1000520, http://dx.doi.org/10.1371/journal.pgen.1000520.