Sources of spontaneous mutagenesis in bacteria

Abstract

Mutations in an organism’s genome can arise spontaneously, i.e., in the absence of exogenous stress and prior to selection. Mutations are often neutral or deleterious to individual fitness, but can also provide genetic diversity driving evolution. Mutagenesis in bacteria contributes to the already serious and growing problem of antibiotic resistance. However, the negative impacts of spontaneous mutagenesis on human health are not limited to bacterial antibiotic resistance. Spontaneous mutations also underlie tumorigenesis and evolution of drug resistance. To better understand the causes of genetic change and how they may be manipulated in order to curb antibiotic resistance or the development of cancer, we must acquire a mechanistic understanding of the major sources of mutagenesis. Bacterial systems are particularly well-suited to studying mutageneisis because of their fast growth rate and the panoply of available experimental tools but, efforts to understand mutagenic mechanisms can be complicated by the experimental system employed. Here we review our current understanding of mutagenic mechanisms in bacteria and describe the methods used to study mutagenesis in bacterial systems.

Introduction

Mechanisms of adaptive genetic change such as CRISPR/Cas systems have received much attention lately as mechanisms by which bacteria modify their genome sequence to evolve resistance in response to selective pressures such as phage infection [reviewed in (Barrangou and Marraffini, 2014)]. However, more than 70 years ago Salvador Luria and Max Delbruck recognized in their classical paper that Escherichia coli mutants resistant to phage could arise “spontaneously”, i.e., prior to phage exposure (Luria and Delbruck, 1943). By demonstrating the importance of spontaneous mutations in generating genetic diversity prior to selective pressure, Luria and Delbruck ushered in the era of bacterial genetics (Luria and Delbruck, 1943, Brock, 1990). In addition to phage resistance, spontaneous mutagenesis can also result in genetic change important for antibiotic resistance, host evasion, and adaptation to new environments, thus driving bacterial evolution. Although much attention is devoted to the effects of environmental and behavioral factors on mutagenesis in bacteria and the development of cancer in humans, approximately 66% of cancer driver mutations in humans are attributable to random spontaneous mutagenesis that occurs during normal DNA replication (Tomasetti et al., 2017). This underscores the importance of understanding the fundamental mechanisms of spontaneous mutagenesis in biology.

In this review, we describe experimental designs for examining spontaneous mutagenesis and mutagenic mechanisms. We also discuss recent progress that has provided a deeper understanding of the cellular processes generating spontaneous mutations. Spontaneous mutations are distinct from mutations induced by environmental stresses such as UV irradiation and genotoxic agents (Ciccia and Elledge, 2010) that are covered by previous Critical Reviews articles (Foster, 2007, Galhardo et al., 2007, Robleto et al., 2007). For readers interested in details of the chemistry behind the myriad types of DNA damage that can cause mutagenesis, we recommend (Friedberg et al., 2006).

Detection of spontaneous mutations

Mutations are heritable changes to an organism’s genotype, which is the term used to describe the entirety of the genetic information in an organism’s genome (Table 1). The word “mutation” is often used in reference to local changes within a gene, including base substitutions and small insertions and deletions. Mutations also include intergenic changes, large genome rearrangenements, and movement of mobile genetic elements. For a period of time, studies of mutagenesis were primarily focused on point mutations, often referring to base pair substitutions, and sometimes one base pair insertions and deletions (Coulondre and Miller, 1977, Cupples and Miller, 1989, Cupples et al., 1990). Base pair substitutions include transitions and transversions. Transitions refer to substitution of a pyrimidine for another pyrimidine (Table 1). Transversions represent substitution of a pyrimidine for a purine or vice versa (Table 1). Functionally, base pair substitutions within the open reading frame of a gene can be synonymous, in which the amino acid sequence of the gene product is unchanged, or they can be nonsynonymous, in which the amino acid encoded at the affected position of the open reading frame is altered. Nonsynonymous substitutions can also result in premature translation termination in the case when a codon for an amino acid is substituted with a stop codon. One base-pair insertions and deletions shift the frame of the gene in which they occur and are almost certain to alter gene function (Table 1). Point mutations within genes are not the only important mutations. In addition to the open reading frame of genes, mutations in cis-regulatory elements are also important because they alter gene expression. In addition to one base-pair insertions and deletions, “small” and large insertions and deletions also occur, with important consequences for gene function. There is no objective standard threshold for which insertions and deletions are considered to be small. For instance, recent work has used insertion and deletion lengths of less than 5, 10, and 28 as a cutoff for considering insertions and deletions to be small (Lee et al., 2012, Long et al., 2015a, Long et al., 2015b).

Table 1.

Key definitions (Friedberg et al., 2006)

Term	Definition
Mutation	A heritable change in an organisms's genome sequence
Spontaneous mutation	A mutation that occurs in the absence of exogenously applied stress or selective pressure
Mutant	An organism with one or more mutations in its genome
Genotype	The genetic information that an organism encodes in its genome
Phenotype	The ensemble of observable characteristics of an organism
Mutagenesis	The process by which mutations are produced
Mutation reporter	A gene or set of genes in which mutations may yield mutantswith a selectable phenotype
Transition	A type of base pair substitution: Purine mutating to purine (G → A, A → G) or pyrimidine mutating to pyrimidine (C → T, T → C)
Transversion	A type of base pair substitution: Purine mutating to pyrimidine (G → T, G → C, A → T, A → C) or pyrimidine mutating to purine (T → A, T → G, C → A, C → G)
Insertion/deletion	Addition or removal of nucleotides from a genome
Frameshift	Insertion/deletion of (3n ± 1) nucleotides within the open reading frame of a gene, where n is an integer.
Target size	The number of unique mutations that are observable in a mutation assay. A simplified example: If one type of mutation at three distinct sites in a reporter gene can generate selectable mutants, the target size is three. If three types of mutations at three distinct sites in the gene can generate selectable mutants, the target size is nine.

Spontaneous mutations occur in the absence of strong selective pressure. Although the mutation rate may in some cases be affected by growth conditions and environmental factors, spontaneous point mutations are consistently found to occur at a rate of 10⁻¹⁰ to 10⁻⁹ per nucleotide per generation for many bacteria and growth conditions (Table 2). As a result, mutants are always the extreme minority within a population, making mutants and their corresponding mutations difficult to detect using bulk approaches. Mutations are also made difficult to detect biochemically due to the fact that replacement of one nucleotide with another does not give rise to a new chemical signature in the mutated DNA that can be conveniently detected in vitro. Therefore, genetic experiments are designed to detect rare mutants. In this section we will discuss several mutation detection methods: the widely used fluctuation test, the increasingly popular mutation accumulation experiment, and the newly emerging maximum-depth sequencing approach. Considerations for use of each method are discussed below.

Table 2.

Mutation rate estimates from various bacteria, methods, and environments

Bacterium	Methoda	Conditionb	Mutation type	Mutation rate per nucleotide per generation (×10−10)c	Reference
Bacillus subtilis	MA/WGS	LBA	BPS	3.35	(Sung et al., 2015, Schroeder et al., 2016)
Bacillus subtilis	MA/WGS	LBA	Indel	1.20	(Sung et al., 2015, Schroeder et al., 2016)
Bacillus subtilis	RifR (rpoB)	LB	BPS	0.34	(Lynch, 2010, Schroeder et al., 2016)
Bacillus subtilis	TmpR (thyP3)	Co-directional replication-transcription	Promoter BPS	85.4 × 10−8 per locus per generation	(Sankar et al., 2016)
Bacillus subtilis	TmpR (thyP3)	Head-on replication-transcription	Promoter BPS	2.68 × 10−8 per locus per generation	(Sankar et al., 2016)
Bacillus subtilis	TmpR (thyP3)	Co-directional replication-transcription	Coding sequence BPS	0.26d	(Sankar et al., 2016)
Bacillus subtilis	TmpR (thyP3)	Head-on replication-transcirption	Coding sequence BPS	0.45d	(Sankar et al., 2016)
Burkholderia cenocepacia	MA/WGS	TSA	BPS	1.33	(Dillon et al., 2015)
Burkholderia cenocepacia	MA/WGS	TSA	Indel	0.17	(Dillon et al., 2015)
Deinococcus radiodurans	MA/WGS	NAG	BPS	4.99	(Long et al., 2015a)
Deinococcus radiodurans	MA/WGS	NAG	Indel	0.22	(Long et al., 2015a)
Escherichia coli	MA/WGS	LBA	BPS	2.00	(Lee et al., 2012)
Escherichia coli	MA/WGS	LBA	Indel	0.37	(Lee et al., 2012)
Escherichia coli	MA/WGS	Minimal medium agar	BPS	1.94	(Foster et al., 2015)
Escherichia coli	Long-term evolution	DMM	BPS	0.89	(Wielgoss et al., 2011)
Escherichia coli	RifR (rpoB)	LB	BPS	0.33	(Lee et al., 2012)
Escherichia coli	NalR (gyrA)	LB	BPS	0.21	(Lee et al., 2012)
Escherichia coli	Histidine auxotrophy (hisGDCBHAFE)	Minimal medium	BPS	5.06	(Lieb, 1951, Drake, 1991)
Escherichia coli	Lac reversion	NR	BPS	5.5	Mean of two estimates in (Drake, 1991)
Helicobacter pylori	RifR (rpoB), CipR (gyrA), ClaR (23S rRNA)	BHI-YE	BPS	27.2	(Lynch, 2010)
Mesoplasma florum	MA/WGS	Mycoplasma medium agar	BPS	97.8	(Sung et al., 2012)
Mesoplasma florum	MA/WGS	Mycoplasma medium agar	Indel	23.1	(Sung et al., 2012)
Mycobacterium smegmatis	MA/WGS	7H10 agar with glycerol and OADC	BPS	5.27	(Kucukyildirim et al., 2016)
Mycobacterium smegmatis	MA/WGS	7H10 agar with glycerol and OADC	Indel	1.24	(Kucukyildirim et al., 2016)
Mycobacterium tuberculosis	MA/WGS	7H10 agar	BPS	2.54	(Rock et al., 2015)
Mycobacterium tuberculosis	MA/WGS	7H10 agar	Indel	1.98	(Rock et al., 2015)
Mycobacterium tuberculosis	RifR (rpoB)	7H9 with OADC	BPS	2.1	(Ford et al., 2011)
Pseudomonas aeruginosa	MA/WGS	LBA	BPS	0.79	(Dettman et al., 2016)
Pseudomonas aeruginosa	MA/WGS	LBA	Indel	0.14	(Dettman et al., 2016)
Salmonella enterica	RifR (rpoB)	LB	BPS	1.30	(Lynch, 2010)
Salmonella enterica	Lac reversion	LB	BPS	2.10	(Lynch, 2010)
Salmonella typhimurium	RifR (rpoB)	LBA, anaerobic	BPS	0.34	(Hughes and Andersson, 1997, Lynch, 2010)
Salmonella typhimurium	RifR (rpoB)	M9 agar with glucose, anaerobic	BPS	1.0	(Hughes and Andersson, 1997)
Salmonella typhimurium	RifR (rpoB)	M9 agar with galactose, anaerobic	BPS	0.38	(Hughes and Andersson, 1997)
Salmonella typhimurium	RifR (rpoB)	LBA, aerobic	BPS	1.0	(Hughes and Andersson, 1997)
Salmonella typhimurium	RifR (rpoB)	M9 agar with glucose, aerobic	BPS	1.1	(Hughes and Andersson, 1997)
Salmonella typhimurium	RifR (rpoB)	M9 agar with galactose, aerobic	BPS	0.57	(Hughes and Andersson, 1997)
Salmonella typhimurium	RifR (rpoB)	M9 agar with galactose, anaerobic	BPS	0.38	(Hughes and Andersson, 1997)
Salmonella typhimurium	RifR (rpoB)	LBA, aerobic	BPS	1.0	(Hughes and Andersson, 1997)
Salmonella typhimurium	RifR (rpoB)	M9 agar with glucose, aerobic	BPS	1.1	(Hughes and Andersson, 1997)
Salmonella typhimurium	RifR (rpoB)	M9 agar with galactose, aerobic	BPS	0.57	(Hughes and Andersson, 1997)

^aMA/WGS = mutation accumulation and whole genome resequencing, RifR = rifampin resistance, TmpR = trimethoprim resistance, NalR = nalidixic acid resistance, CycR = cycloserine resistance, CipR = ciprofloxacin resistance, ClaR = clarithromycin resistance.

^bLBA = LB agar, TSA = tryptic soy agar, NAG = nutrient agar with glucose, DMM = Davis minimal medium, BHI-YE = brain heart infusion with yeast extract, NR = not reported

^cBPS rate per site per generation estimates from mutation reporters were calculated as described in (Lynch, 2010) unless otherwise noted.

^dThe BPS mutation rate per site per generation, μ_BS, was calculated using μ_BS = μ_L × f_T × f_BS/(L × f_L × f_D) from (Lynch, 2010), where μ_L is the per locus mutation rate, f_T, f_BS, and f_L are all 1, and L = 840. 220 unique BPSs at 173 sites were detected in the thyP3 coding sequence by (Sankar et al., 2016). Therefore, f_D = 220/(3 × 173) = 0.475. Dividing by three in calculation of f_D accounts for the three possible nucleotides that could have substituted for the native nucleotide (Lynch, 2010).

Fluctuation tests and mutation reporter genes

The fluctuation test is the workhorse experiment for estimating spontaneous mutation rates in bacteria (Luria and Delbruck, 1943). The fluctuation test is useful in bacterial systems in that selection is applied to large bacterial populations to enable estimation of low mutation rates. In a fluctuation test, parallel cultures from small innocula are first grown in the absence of selection so that mutations may spontaneously arise (Figure 1A). Mutants that arise in culture proliferate exponentially along with the rest of the cells during the growth phase, often constituting a tiny fraction of each culture. After the growth phase, selection is applied under which only selectable mutants grow to form colonies on solid media (Figure 1A). Mutations allowing survival and growth during selection arise in mutation reporter genes, for which mutations affecting function of the gene product give rise to a selectable phenotype (Table 1, Figure 1B). Measuring the distribution of the number of mutant colonies arising from each culture in a fluctuation test enables estimation of mutation rate, which is the probability of mutagenesis per genome duplication.

Figure 1.

Fluctuation tests, mutation reporters, mutation accumulation experiments, and maximum-depth sequencing. (A) An illustration of the fluctuation test. Mutants (black circles) arise during exponential growth of independent cultures. If selection is lethal to non-mutants, only mutants that arose during culture growth will form colonies (black colonies). However, if selection is for growth, slow growth during selection may give rise to additional mutants that form colonies during selection (red colonies). (B) The rationale behind reverse and forward mutation reporters is illustrated. For reverse mutation reporters, a reversion assay may be performed (top), in which a functional product allows survival and growth under selection. In the other class of reverse mutation reporter (middle) an active gene product is produced before and after mutagenesis, but the mutant gene product remains functional during selection, in contrast to non-mutant gene products, which do not function during selection. Forward mutation reporters (bottom) have a larger target size, in that any mutation eliminating expression or gene product activity results in growth or survival under selection. (C) A table of data from Luria and Delbruck’s 1943 paper (Luria and Delbruck, 1943) with examples of jackpot cultures indicated by rows shaded in gray. (D) An example mutation accumulation line. Mutations accumulate in the line’s genome (shown below the example mutation accumulation line) as it is passaged. After the desired number of generations, genomic DNA from each line is purified and sequenced to identify mutations that arose during the experiment. (E) The schematic workflow of maximum depth sequencing. A unique molecular identifier (UMI) is appended to the 3′ end of the original molecule of the region of interest (ROI), followed by linear amplification of the ROI molecules with their associated UMIs. The consensus sequence for each ROI/UMI pool is determined and the proportion of ROI/UMI pools with mutations at a given position is determined in order to calculate mutation rate at the position. A color version of this figure is available online.

Distribution of mutants per culture in fluctuation tests

Commonly applied methods for calculating mutation rate from fluctuation tests include the p₀ method, which derives the mutation rate from the proportion of cultures that do not generate any selectable mutants (Luria and Delbruck, 1943), the Lea-Coulson method, which calculates mutation rate from the median number of mutants per culture (Lea and Coulson, 1949), and the MSS method, which uses a maximum-likelihood approach to determine the most likely number of mutants per culture given the observed distribution of mutants per culture (Sarkar et al., 1992). We encourage readers interested in mehods of calculating mutation rate to consult these reviews (Rosche and Foster, 2000, Foster, 2006) and the following resource (Hall et al., 2009).

In order to accurately infer mutation rate using mutation reporter genes, the assumptions of the fluctuation test must be valid (Luria and Delbruck, 1943, Lea and Coulson, 1949, Rosche and Foster, 2000, Foster, 2006). 1) Mutants should be selectively neutral during growth prior to selection. 2) Mutants must not be allowed to arise during selection. Because mutations will be inherited by the offspring of mutants, a mutant will proliferate to form a clone of mutants within its culture (Figure 1A). Therefore, cultures in which mutants arose early have vastly more mutants than the mean number of mutants per culture (Figure 1A and 1C). Such cultures are termed “jackpots”, and they cause the distribution of mutants per culture in a fluctuation test to follow the Luria-Delbruck distribution when the assumptions of the test are met. When the assumptions are not met, however, the distribution of mutants per culture will be shifted toward a Poisson distribution.

When the distribution of the number of mutants per culture is shifted toward Poisson, this may indicate fitness defects of mutants in bulk culture, or post-selection mutagenesis in addition to spontaneous mutagenesis during clonal growth (Cairns et al., 1988). Mutants that arise during selection typically form colonies after several days on selectable medium, but may not be distinguishable from slow-growing mutants that arose in culture prior to selection. Post-selection mutagenesis is particularly prone to occur when selecting for growth of mutants rather than for their survival (Figure 1A). Under conditions allowing post-selection mutagenesis, the total number of mutants observed in each culture is the sum of the number of mutants that arose spontaneously during growth in culture and the number that arose after selection was imposed (Figure 1A) (Cairns et al., 1988). It is because of this that post-selection mutagenesis causes estimates of spontaneous mutation rate to be erroneously high.

Estimates of spontaneous mutation rate can be erroneously low due to phenotypic lag (Demerec, 1946, Newcombe, 1948, Newcombe and Scott, 1949, Kissling et al., 2013). Phenotypic lag is the term for the delay in time between a mutant genotype arising and the display of the resulting phenotype. Phenotypic lag could emerge for a number of reasons when using mutation reporters. For instance, if the mutant allele is recessive to the wild type allele, sufficient time will be required to dilute or degrade the wild type protein after mutation of the reporter gene so that the mutant phenotype presents itself (Demerec, 1946). Similarly, in organisms with multiple copies of the genome or multi-fork replication, sufficient generations must transpire before a recessive allele will be expressed as a phenotype (Witkin, 1951, Kissling et al., 2013). Even if the mutant allele is dominant, sufficient time will be required to express the gene product and yield an observable phenotype, resulting in phenotypic lag.

You get what you select for – types of mutation assays

There are many mutation reporters available for use in a fluctuation test. They differ in the type of selection that is applied, and their target size, which is the number of mutations that are observable by a mutation assay (Table 1). Depending on the mutation reporter used, selection for mutants can be for either their growth or survival. These aspects of mutation reporter genes may affect the validity of the two major assumptions of fluctuation tests listed above. With that in mind, we discuss several mutation reporter genes in this section.

Mutation assays can be classified into two groups: reverse and forward (Lieb, 1951, Skopek et al., 1978). Reverse mutation assays rely on mutations in the reporter gene (reverse mutation reporter) that yield a functional product during selection (Figure 1B). In contrast, forward mutation assays generate a selectable phenotype upon acquiring loss-of-function mutations in the reporter gene (forward mutation reporter) (Figure 1B).

Reverse mutation assays

Reverse mutation assays detect mutations that alter a non-functional variant of a gene so that it becomes functional and encodes an active gene product after mutagenesis (Figure 1B). A fairly small target size and low diversity of mutation types, usually base pair substitutions, are observable in reverse mutation assays. A classic reverse mutation assay is the reversion of a mutant lacZ gene encoding β-galactosidase, which allows lactose utilization. Mutations causing reversion to an active lacZ gene product can be selected for growth on lactose as the only carbon source (Cupples and Miller, 1989, Cupples et al., 1990). In 1989, Cupples and Miller developed a series of lacZ variants, each designed specifically to detect only one base pair substitution at a defined location (Cupples and Miller, 1989). Thus, with a target size of one, the Cupples and Miller reversion system is highly specific and the identity of the mutations yielding selectable mutants are known without the need for sequencing the mutation reporter gene. The lacZ locus was later modified to enable detection of insertions and deletions (Cupples et al., 1990). These represent important improvements in reversion assays for detection of specific mutations.

Mutations in certain wild type genes can lead to antibiotic resistance phenotypes. Although these mutations are not reversion events per se, these assays are considered reverse mutation assays because the mutations yielding selectable mutants result in a gene product that is active under selection (Figure 1B). For example, fluoroquinolone antibiotics target DNA gyrase. Specific nonsynonymous substitutions in gyrA, the gene encoding DNA gyrase, result in gyrase variants that are functional in the presence of fluoroquinolone antibiotics (Yoshida et al., 1990, Becket et al., 2012). This enables the use of gyrA as a mutation reporter. The antibiotic rifampicin targets RNA polymerase. Certain non-synonymous substitutions in rpoB, which encodes the β subunit of RNA polymerase, render RNA polymerase uninhibited by rifampicin (Campbell et al., 2001, Garibyan et al., 2003). This allows estimation of mutation rate by selecting for resistance to rifampicin. Use of rpoB and gyrA as mutation reporters is simple and can be performed in many bacteria without the necessity of genome manipulations to set up a specific reporter system (Richardson and Stojiljkovic, 2001, Wang et al., 2001, Koskiniemi et al., 2010, Rock et al., 2015, Deiham et al., 2017).

Forward mutation assays

Forward mutation assays often have a much larger target size and diversity of detectable mutations (Figure 1B). This is because forward mutation assays detect mutations resulting in loss of function of the reporter gene rather than detecting mutations resulting in gain of function of the gene, as is the case for reverse mutation assays. Luria and Delbruck’s fluctuation experiment was a forward mutation assay, as loss-of-function mutations in phage receptor genes render E. coli resistant to phage T1 (Hantke and Braun, 1978, Kadner et al., 1980). A classical forward mutation assay with a more precisely defined target for mutagenesis is lacI. Loss-of-function mutations in lacI, which encodes the repressor of the lac operon, allow constitutive expression of the lac operon. This phenotype can be selected by requiring cells to grow on plates supplemented with phenyl-β-D-galactoside, which is not an inducer of lac operon expression but can be used as a carbon source if the lac operon is expressed (Coulondre and Miller, 1977, Schaaper et al., 1986, Schaaper and Dunn, 1991). Loss-of-function mutations in lacI can occur throughout the gene’s open reading frame and cis-regulatory elements, and selectable mutants can also arise due to mutations in lacO, which is a cis-regulatory element for the lac operon and a binding site for LacI. For this reason, the target size when selecting for Lac+ mutants in the lacI forward mutation assay is much larger than that of the lacZ reverse mutation assay. However, the problem of selection for growth and mutagenesis during selection still exists.

More recently, a forward mutation assay was designed to select for survival of mutants and to ensure that mutations were generated prior to selection in E. coli (Yoshiyama et al., 2001). The forward mutation reporter used was the rpsL gene, which encodes the small ribosomal subunit protein S12. A multicopy plasmid encoding wild type rpsL was allowed to replicate and accumulate mutations in E. coli. Owing to the multicopy nature of the plasmid, mutations in plasmid-encoded rpsL were recessive and did not cause fitness defects in culture. The plasmid population was then purified and used to transform E. coli cells in single copy (Yoshiyama et al., 2001). The cells into which the plasmid library was transformed harboured a chromosomally-encoded recessive allele of rpsL yielding streptomycin resistance. Therefore, cells transformed with a plasmid encoding wild type S12 expressed the dominant, wild type rpsL phenotype and were sensitive to streptomycin. In contrast, cells receiving a plasmid encoding inactive S12 were streptomycin resistant. Thus, transformants that were streptomycin resistant were inferred to have been transformed with a plasmid in which loss-of-function mutations occurred in rpsL (Fujii et al., 1999, Yoshiyama et al., 2001).

A fluctuation test to detect spontaneous mutations on the chromosome of Bacillus subtilis was recently developed to use the thyP3 gene, which encodes thymidylate synthetase, as a forward mutation reporter. A native, temperature sensitive thymidylate synthase, ThyB(ts) (Neuhard et al., 1978) was used as a phenotypic buffer during growth at its permissive temperature. This ensured that loss-of-function mutations in thyP3 did not cause fitness defects in culture prior to selection (Sankar et al., 2016). Therefore, mutations in thyP3 arose during growth at the permissive temperature for ThyB(ts) and were selected for survival of trimethoprim treatment at the nonpermissive temperature for ThyB(ts) (Sankar et al., 2016). This allowed thymidylate synthase to be used as a forward mutation reporter that accurately measured mutation rate in a fluctuation test without confounding fitness defects in culture and by selecting for survival of mutants rather than for their growth.

Mutation accumulation experiments

Although mutation reporters are easily used in bacterial systems, they are constrained by the mutations that can yield selectable mutants, and thus have a limited target size. Reductions in the cost of whole genome resequencing have led to a burst of mutation accumulation experiments in which mutations are detected by resequencing entire genomes. In comparison to mutation reporters, mutation accumulation experiments have a much larger target size: the entire mappable genome.

Mutation accumulation experiments allow detection of any mutation that is not extremely deleterious, despite possible fitness defects (Barrick and Lenski, 2013). This is achieved by subjecting a population to single-cell bottlenecks at regular intervals (Figure 1D). Serially bottlenecking populations in mutation accumulation experiments allows a lineage to accumulate mutations in a manner that is essentially neutral with respect to evolution. Accordingly, mutations which would have been eliminated from the population due to a fitness disadvantage instead become fixed in the population after bottlenecking (Lind and Andersson, 2008). This experimental design is particularly useful for studying bacterial mutagenesis, because introducing the bottleneck is as simple as performing colony purification by passaging single bacterial colonies from one plate to another (Figure 1D). This, in combination with the dramatic reduction in whole genome resequencing cost over the past decade, has enabled several studies to make use of mutation accumulation experiments followed by whole genome resequencing to investigate mechanisms of spontaneous mutagenesis in bacteria (Lind and Andersson, 2008, Lee et al., 2012, Foster et al., 2015, Long et al., 2015b, Rock et al., 2015, Sung et al., 2015, Bhagwat et al., 2016, Lee et al., 2016, Schroeder et al., 2016, Dillon et al., 2017). A large spectrum of distinct mutation types is observable using the mutation accumulation approach, from nonsynonymous single-nucleotide variants and small insertions/deletions within genes, which are also detectable by reporters, to silent mutations and intergenic mutations, which often are not detectable by reporters. In addition, large structural rearrangements of the chromosome and movement of mobile genetic elements such as insertion sequences are detectable in the mutation accumulation approach (Lee et al., 2012, Lee et al., 2016).

Maximum-depth sequencing

Maximum-depth sequencing is a recently developed approach to infer spontaneous mutation rates of specific regions of interest in a genome (Jee et al., 2016). High-throughput sequencing libraries are prepared using bulk genomic DNA from a culture of cells and a primer specific to the region of interest to target sequencing to that position within the genome (Figure 1E). The primer contains an adapter sequence with a type of barcode termed a “unique molecular identifier”, which can be used to determine which reads represent the sequence of a single, shared original molecule (Jee et al., 2016). Genomic regions of interest are directly tagged with a unique molecular identifier and linearly amplified. After high-throughput sequencing, a consensus sequence is determined for each pool of reads corresponding to a single region of interest and unique molecular identifier (Figure 1E). The error rate for consensus sequence calling is inversely proportional to the sequencing depth for each region of interest/unique molecular identifier pool. Therefore, by increasing sequencing depth, error rate for consensus calling can be reduced indefinitely. Even lethal mutations are detectable by this method, highlighting its use for detecting deleterious mutations. While maximum-depth sequencing cannot be used to study mutagenesis genome-wide, it can be performed much more quickly than the mutation accumulation approach.

Mechanisms of spontaneous mutagenesis

In this section, we discuss insights into spontaneous mutagenic mechanisms gained through use of the methods described above. While mechanisms that prevent or correct spontaneous mutagenesis are well understood (Lenhart et al., 2012), less is known about the mechanisms by which normal cellular processes promote spontaneous mutagenesis. These processes include DNA replication, transcription, recombination, and metabolic processes that generate reactive species that damage DNA, i.e., reactive oxygen species. These processes may interact. For example, mutagenesis can be promoted by DNA replication and transcription as well as by conflicts between the two processes. Here we review our current understanding of the major paths toward spontaneous mutagenesis in bacteria (Figure 2).

Figure 2.

Common sources of spontaneous mutagenesis in bacteria. Cellular processes that promote spontaneous mutagenesis are listed at the top of the schematic with arrows indicating the types of mutations with which they are associated.

DNA replication errors

DNA polymerases that replicate the genome are extremely accurate in terms of their ability to correctly pair nucleotides, but they do make errors. These errors include mispairing of noncognate dNTPs in violation of Watson-Crick rules and incorporation of noncanonical nucleotide substrates such as ribonucleotides or oxidized dGTP (8-oxo-dGTP, or 8-oxo-G). In addition, DNA replication can also be mutagenic by primer-template slippage or template switching, resulting in insertions, duplications, and deletions to the genome. Each mutagenic mechanism is discussed below.

Base pairing errors

DNA polymerases that replicate bacterial genomes form base pairing errors that cause mutations if they are not corrected before the next round of DNA replication by the proofreading activity of the replicative DNA polymerase or the DNA mismatch repair system (MMR). Base pairing error generation, evasion of proofreading activity, and evasion of MMR each contribute to mutagenesis, but the mechanisms and factors affecting each are somewhat poorly understood.

Leading- and lagging strand replication

In S. cerevisiae base pair substitutions are more common during lagging strand replication due to extension of RNA primers by the error-prone DNA polymerase, Pol α (Lujan et al., 2014, Reijns et al., 2015). Errors made by Pol α are proofread by Pol δ (Pavlov et al., 2006), but DNA-binding proteins such as nucleosomes and transcription factors may discourage such proofreading, providing a mechanism for errors made by Pol α during lagging strand replication to remain in the genome and become mutations in the following round of DNA replication (Reijns et al., 2015).

Considerably less is understood about mechanisms promoting differences in base pairing error rates between leading- and lagging strand replication in bacteria. Available literature suggests that E. coli Pol III may be more accurate in base pairing of nucleotides in the lagging strand, but the mechanism for this is unknown (Fijalkowska et al., 1998). Mutation accumulation studies have shown a bias in distribution of transitions between the leading- and lagging strands (Lee et al., 2012, Sung et al., 2015, Schroeder et al., 2016), but it remains unclear whether the errors responsible for the mutations were produced during leading- or lagging strand replication. DNA damage in the lagging-strand template of B. subtilis may enlist an error-prone translesion DNA polymerase, PolY1, to enable mutagenic synthesis of the lagging strand past the damaged site (Million-Weaver et al., 2015). However, loss of PolY1 in B. subtilis does not alter the spontaneous mutation rate in a mutation reporter (Duigou et al., 2004). Overexpression of PolY1, however, does increase mutagenesis, but strand bias of this effect has not been demonstrated (Duigou et al., 2004).

It is therefore unclear the extent to, and mechanism by which PolY1 affects spontaneous mutagenesis in B. subtilis.

Rare tautomer hypothesis

Rare tautomeric forms of bases in DNA are a mechanism for transitions, as proposed by the rare tautomer hypothesis (Watson and Crick, 1953, Topal and Fresco, 1976). Under normal circumstances, a base pairing error will cause movement of the mismatched 3′ terminus of the nascent strand from the polymerase active site of DNA polymerase to its exonuclease active site so that proofreading may occur (Fernandez-Leiro et al., 2017). However, if a base is present in its rare tautomeric form, the geometry of base-pairing will be very close to that of correctly paired DNA (Figure 3A), allowing the 3′ terminus to remain in the polymerase active site (Kool, 2002, Harris et al., 2003, Kim et al., 2005). It is difficult to test the rare tautomer hypothesis due to the low tautomeric constant of the bases found in DNA. However, by using the nucleoside analog dP (6-(2-deoxy-β-D-ribofuranosyl)-3,4-dihydro-6H,8H-pyrimido[4,5-c][1,2]oxazin-2-one), which readily tautomerizes to pair with either adenine or guanine, it was found that the tautomeric constant of the P base is highly correlated with its ability to be included as a mispair in nascent DNA (Harris et al., 2003). This work was carried out using proofreading-deficient Klenow fragment, so the effect of tautomerization on proofreading activity could not be discerned. The first direct evidence in support of the rare tautomer hypothesis came in 2011, when the structure of a DNA polymerase with a C∙A mispair in the active site was solved with the C in its tautomeric form (Wang et al., 2011). This study represents strong evidence that base pairing errors with similar geometry to cognate Watson-Crick pairing can be generated when a base is present as a rare tautomer. It is poorly understood whether tautomerization of nucleobases can be affected by factors such as local sequence context in vivo. It is also not known to what extent rare tautomeric bases remain in DNA after their incorporation compared to the frequency with which they switch back to their canonical tautomeric form. In addition, further study will be necessary to determine whether rare tautomers primarily affect incorporation of base pairing errors into DNA by DNA polymerases versus evasion of proofreading and MMR.

Figure 3.

Nucleobase modifications involved in mutagenesis. (A) Base pairing interactions between rare tautomeric forms of the nucleobases in DNA are shown for comparison to canonical base pairs. (B) Interactions between deoxyadenosine and syn-8-oxo-dG. (C) Cytosine deamination leads to uracil, causing C∙G→T∙A transitions, and adenine deamination leads to hypoxanthine, yielding A∙T→G∙C transitions (see section: Replication-transcription conflict). Structures were illustrated using ChemDraw software.

Sequence context

Local sequence context affects the rate at which base pairing errors are produced, proofread, and corrected by MMR. One mechanism by which sequence context can affect the accuracy of DNA polymerases is by base stacking interactions between neighboring nucleotides in DNA. Specifically, purines 5′ to a given position stack more stably than pyrimidines [for reviews, see (Goodman et al., 1993, Hunter, 1993, Kool, 2001)]. In agreement with this, a mutation accumulation study using MMR-deficient S. cerevisiae demonstrated that proofreading-deficient DNA polymerases were more likely to misinsert thymidine across from cytidine or guanosine when a purine was at the 3′ end of the primer strand (Lujan et al., 2014). Although this has not been tested in bacterial systems, base stacking interactions are a intrinsic property of all DNA. It is therefore reasonable to extrapolate the observations made in S. cerevisiae to bacterial systems.

Proofreading of errors made by DNA polymerases is also affected by local sequence context (Sinha, 1987). High GC content 5′ or 3′ to the mismatched nucleotide in nascent DNA makes proofreading less likely (Petruska and Goodman, 1985, Carver et al., 1994). The mechanism for this is thought to be the stability of base pairing of GC-rich DNA relative to AT-rich DNA. More stable base pairing would decrease the ability of the mismatched 3′ terminus to enter the exonuclease (proofreading) active site of DNA polymerase (Carver et al., 1994). Effects of sequence context on proofreading activity in vivo can be estimated by observing the relative mutation rate in differing contexts in cells without MMR, because most mutations in the absence of MMR are due to replicative DNA polymerase errors (Schaaper and Dunn, 1987, Schaaper, 1993). Mutation accumulation studies conducted using bacteria lacking MMR have demonstrated that neighboring G∙C base pairs strongly increase the base pair substitution rate at a given genome position in vivo (Lee et al., 2012, Sung et al., 2015, Schroeder et al., 2016).

The effect of local sequence context on MMR efficiency has been more difficult to estimate, because it requires comparison of substitution rates in various contexts from MMR-deficient and MMR-proficient lines, which accumulate few mutations. However, one study showed that MMR is particulary efficient at correcting mismatches neighbored by G∙C base pairs (Lee et al., 2012). Now that ample mutation accumulation experiments have been performed in MMR-proficient E. coli (Lee et al., 2012, Foster et al., 2015, Bhagwat et al., 2016), the question of how local sequence context affects MMR efficiency genome-wide can be more rigorously examined.

Use of 8-oxo-dGTPas a substrate for DNA replication

The nucleotide pool is affected by endogenous reactive oxygen species generated by respiration. Of the bases in DNA, guanine is the most frequent target for oxidation, forming 8-oxo-7,8-dihydroguanine (8-oxo-G) (Cadet et al., 2008, Kanvah et al., 2010). Genetic and biochemical evidence indicates that DNA polymerase III incorporates 8-oxo-dGTP into DNA and is inefficient at proofreading 8-oxo-dGMP after its incorporation (Yamada et al., 2012). 8-oxo-dGMP can also be generated when dGMP already in genomic DNA reacts with reactive oxygen species. The bases in nucleotides can adopt anti- or syn conformations. When 8-oxo-G is in the anti conformation it will pair with cytidine. However, it more often adopts the syn conformation and will readily pair with adenosine (Figure 3B) (Kasai et al., 1991, Kouchakdjian et al., 1991, McAuley-Hecht et al., 1994). Structural evidence has suggested that when the newly synthesized DNA is rigid within the active site of a DNA polyemrase, 8-oxo-G may be more mutagenic than when the newly synthesized DNA is flexible (Krahn et al., 2003). This is because 8-oxo-G is in the syn conformation that is required for its pairing with adenosine when the newly replicated double-stranded DNA is rigid (Krahn et al., 2003).

Mutations caused by 8-oxo-G are either A∙T → C∙G transversions or G∙C → T∙A transversions, depending on how 8-oxo-G enters genomic DNA. When 8-oxo-dGTP is incorporated across from adenosine during DNA replication, the following round of DNA replication may incorporate cytidine across from 8-oxo-G if the base has switched to the anti conformation, causing A∙T → C∙G transversions. Conversion of guanine to 8-oxo-G at a G∙C base pair within DNA can lead to insertion of adenosine instead of cytidine when syn-8-oxo-G is used as a template for the next round of replication, causing G∙C → T∙A transversions (Miller, 1996). The oxidized guanine (GO) system limits mutagenesis by sanitizing nucleotide pools of 8-oxo-dGTP, removing oxidized 8-oxo-G bases in double-stranded DNA, and removing adenine paired with 8-oxo-G in double-stranded DNA (Michaels and Miller, 1992, Lenhart et al., 2012). Work performed in E. coli used mutation accumulation in genetic backgrounds eliminated for several DNA repair pathways to determine the extent to which different mutagenic mechanisms affect spontaneous mutation in unstressed conditions. Strikingly, this work showed that, other than the mismatch repair pathway, loss of the GO system was the only pathway out of many tested that showed an increase in base pair substitution rates (Foster et al., 2015), with transversion rates increasing in the absence of the GO system.

Ribonucleotide misincorporation

Ribonucleotides enter genomic DNA via their misincorporation by DNA polymerases during DNA replication (Wang et al., 2012, Yao et al., 2013). Ribonucleotide misincorporation is by far the most frequent error made by DNA polymerases, with the error rate largely driven by the inequity of the cellular nucleotide pools, as NTPs vastly outnumber dNTPs (Traut, 1994, Buckstein et al., 2008, Ferraro et al., 2010, Nick McElhinny et al., 2010a, Nick McElhinny et al., 2010b, Reijns et al., 2012, Yao et al., 2013). For instance, the E. coli DNA polymerase III core with the clamp loader complex associated incorporates ribonucleotides into nascent DNA in vitro at a rate that would result in ≈ 2000 misinsertions per genome replication event in vivo (Yao et al., 2013). This is in stark contrast to the rate of base pairing errors produced by E. coli DNA replication, during which about 7.9 genome replication events occur between single base pairing errors escaping proofreading (Lee et al., 2012). Despite their high frequency of misincorporation into genomic DNA and their susceptibility to hydrolysis of the sugar-phosphate backbone, ribonucleotides covalently embedded in genomic DNA have a rather mild effect on bacterial mutation rate, yielding a small increase in base pair substitutions in B. subtilis (Yao et al., 2013).

Ribonucleotide excision repair (RER) is the biochemical pathway responsible for replacing ribonucleotides in DNA. Loss of RER in Saccharomyces cerevisiae leads to an increased rate of 2–5 nucleotide deletions (Nick McElhinny et al., 2010a, Lujan et al., 2014). These deletions are dependent on S. cerevisiae topoisomerase 1, encoded by the TOP1 gene (Kim et al., 2011), and exhibit a strand-biased distribution (Williams et al., 2013, Cho et al., 2015). Bacteria lack a clear homolog of yeast topoisomerase 1, and although loss of RER causes a mild increase in mutation rate, the nature of the changes are unknown (Yao et al., 2013). One explanation for the low mutability of ribonucleotides misincorporated by DNA polymerase during bacterial replication is that nucleotide excision repair (NER) is able to remove ribonucleotides in the absence of RER (Vaisman et al., 2013, Vaisman et al., 2014). Removal of ribonucleotides by NER in E. coli is error-free in contrast to their removal by topoisomerase 1 in yeast (Yao et al., 2013). Despite a growing volume of research on bacterial RER, the overall mechanism of RER and the mechanisms of increased mutagenesis due to unrepaired ribonucleotides in bacterial DNA are poorly understood.

Strand misalignment

The mutagenic mechanisms discussed above involve errors made during DNA replication. In contrast to mutations arising from misinsertion of nucleotides during DNA replication, a class of mutations termed “encoded errors” arises due to accurate insertion of nucleotides using a misaligned primer-template pair [reviewed in (Lovett, 2004)]. During normal DNA replication the DNA polymerase can move from using one template to another. This type of strand misalignment is termed template switching, and it can lead to a variety of mutations, including base pair substitutions, insertions and deletions, and large structural rearrangements (Yoshiyama et al., 2001, Dutra and Lovett, 2006, Seier et al., 2011, Anand et al., 2014, Tsaponina and Haber, 2014). Quasipalindromic sequences promote template switching by forming a hairpin structure during DNA replication such that one arm of the quasipalindrome serves as the template for the other arm (Ripley, 1982). Quasipalindromes are mutation hotspots in bacteria due to the increased likelihood of template switching events at hairpins (Viswanathan et al., 2000, Dutra and Lovett, 2006, Seier et al., 2011, Seier et al., 2012). In addition, replication fork stalling, due to either complex DNA structures in eukaryotes (Lee et al., 2007) or replication-transcrption conflict in bacteria (Sankar et al., 2016) can promote template switching.

Long homopolymer runs are hotspots for small insertions and deletions in bacteria (Streisinger et al., 1966, Fujii et al., 1999, Lee et al., 2012, Schroeder et al., 2016) due to a type of primer-template misalignment termed “strand slippage” during DNA replication (Streisinger et al., 1966, Canceill et al., 1999, Fujii et al., 1999). Small insertions and deletions are enriched intergenically in B. subtilis due to intergenic enrichment of homopolymer runs (Schroeder et al., 2016). The relative paucity of homopolymer runs in coding DNA likely reflects purifying selection against genes containing frameshift-promoting homopolymers.

DNA deamination

A major source of base substitutions is from DNA damage by deamination of the bases in nucleotides. For example, cytosine deamination leads to uracil. If uridine in DNA goes unrepaired, adenosine will be inserted into the nascent DNA strand when replication uses uridine as a template, causing a G∙C → A∙T transition. Nucleobases in single-stranded DNA are more susceptible to deamination than those in double-stranded DNA. For instance, cytosine deamination occurs with a rate constant of approximately 1 × 10⁻¹⁰ per cytosine site per second in single-stranded DNA (Figure 3C) (Frederico et al., 1990). This is in stark contrast to its rate in double-stranded DNA, which is 7 × 10⁻¹³ per cytosine site per second (Frederico et al., 1990). This represents about a 140-fold lower rate of cytosine deamination in double-stranded DNA compared to single-stranded DNA. Therefore, by transiently generating regions of single-stranded DNA, transcription and DNA replication strongly increase the likelihood that nucleobases will be deaminated. It is unclear what the relative contributions of transcription and DNA replication are to overall mutagenesis by their generation of single-stranded DNA. Our current understanding of the contribution of replication and transcription to this mutagenic mechanism is discussed below.

Single-stranded DNA generated during DNA replication

DNA replication is semidiscontinuous and generates transient patches of single-stranded DNA in the lagging-strand template throughout the genome. These single-stranded DNA patches are a major source of spontaneous cytosine deamination in bacteria (Bhagwat et al., 2016). Given the approximate rate of spontaneous cytosine deamination in single-stranded DNA such as that found in the lagging-strand template, replication-associated deamination of cytosine to uracil could be a significant contributor to G∙C → A∙T transitions in bacteria. The time from cytosine deamination in the lagging-strand template to replication over the resulting uridine is extremely short, and the uridine-containing template will be quickly used as a template for insertion of adenosine into the nascent lagging strand. Uracil-DNA glycosylase (Ung) removes uracil from DNA, leaving an abasic site. If an abasic site is produced in the lagging-strand template by Ung, the replicative DNA polymerase would stall at the site. A translesion DNA polymerase would then be required to synthesize DNA over the abasic site. In that event, the translesion DNA polymerase is likely to insert adenosine across from the abasic site, resulting in a G∙C → A∙T transition (Lawrence et al., 1990, Reuven et al., 1999, Tang et al., 1999).

Mutation accumulation studies of E. coli, B. subtilis and Mesoplasma florem have shown that G∙C → A∙T transitions are much more likely when cytidine is in the lagging-strand template than when guanosine is in the lagging-strand template (Sung et al., 2015, Schroeder et al., 2016). Cytosine deamination in the lagging-strand template may play a role in this strand bias. Deamination of cytosine in the lagging-strand template has also been proposed to contribute to the nearly universal enrichment of guanosine compared to cytosine in the lagging-strand template (GC skew) of bacteria (Lobry, 1996, Bhagwat et al., 2016).

Single-stranded DNA generated during transcription

RNA polymerase uses one strand of DNA as its template strand to pair with the newly synthesized RNA, leaving the non-template strand unpaired (Gnatt et al., 2001). Since the non-template strand becomes locally exposed to solvent as single-stranded DNA during transcription, it is prone to damage by various mechanisms, including cytosine deamination to uracil. As discussed, uridine pairs with adenosine when DNA is replicated, causing G∙C → A∙T transitions (Beletskii and Bhagwat, 1996). Mutation accumulation experiments in Salmonella typhimurium showed that G∙C → A∙T transitions were made more likely in coding sequences of Ung-deficient cells if the cytosine were in the non-template strand for transcription and if the gene were highly expressed, thereby exposing it during transcription and making it more likely to undergo deamination (Lind and Andersson, 2008). Most bacterial genes are transcribed co-directionally with DNA replication (Rocha, 2004), so the lagging-strand template for replication and the non-template strand for transcription are often the same physical strand of DNA. We find it likely, therefore, that strand-biased occurrence of G∙C → A∙T transitions (Sung et al., 2015, Schroeder et al., 2016) arises due to a combination of cytosine deamination in both the lagging-strand template at the replication fork (Bhagwat et al., 2016) and single-stranded DNA present in the non-template strand during transcription (Beletskii and Bhagwat, 1996, Lind and Andersson, 2008). These strand-biased mutagenic mechanisms are in addition to differences in leading- and lagging strand replication fidelity (Fijalkowska et al., 1998).

In addition to promoting cytosine deamination in coding sequences, transcription encourages premutagenic DNA damage in promoter regions. During transcription initiation, the promoter sequences are melted but are not base paired with RNA because the promoter region is upstream of the transcription start site. Thus, both the template strand and the non-template strand sequences may be susceptible to deamination during initiation before promoter escape by RNA polymerase. Forward mutation assays have shown that the conserved T₋₇ position in the −10 element of promoters for head-on genes is a hotspot for A∙T → G∙C transitions in both B. subtilis (Sankar et al., 2016) and E. coli (Yoshiyama et al., 2001). During transcription initiation, T₋₇ of the non-template strand is sequestered inside the housekeeping sigma factor during transcription initiation, leaving the complementary adenosine (A₋₇) unpaired as single-stranded DNA (Feklistov and Darst, 2011). Adenine in single-stranded DNA can be spontaneously deaminated to hypoxanthine (Figure 3C), which readily pairs with cytidine in the next round of DNA replication, causing A∙T → G∙C transitions. This mechanism may explain other base substitutions observed in cis-regulatory elements of genes in forward mutation assays (Schaaper et al., 1986, Yoshiyama et al., 2001, Sankar et al., 2016). The fact that promoter substitutions have not been more commonly observed in mutation assays is likely due to the prevalent use of reversion assays that only detect coding substitutions. In addition, promoter mutations are likely to cause major gene expression defects, and are therefore strongly selected against in essential genes and may be difficult to observe in mutation accumulation studies. However, the importance of such mutations should not be underestimated due to their fundamental impact in altering gene expression. Collectively, these studies show that the non-transcribed strand is prone to chemical modifications that have the potential to increase spontaneous mutagenesis of both coding sequences and regulatory regions of highly transcribed genes.

Replication-transcription conflict

DNA replication and transcription must both use the same DNA template. DNA replication forks move much faster than transcription in bacteria, leading to unavoidable conflicts between these two fundamental processes. There are two types of replication-transcription conflicts, depending upon gene orientation relative to DNA replication (Rocha, 2004). Co-directional conflict arises in genes for which the coding strand for transcription corresponds to the leading-strand template for replication, whereas head-on conflicts occur when the coding strand of a gene is the lagging-strand template for replication. Both types of conflict can lead to replication fork stalling, homologous recombination, and genome instability (Mirkin et al., 2006, Srivatsan et al., 2010, Dutta et al., 2011, Merrikh et al., 2011). We will focus on the mechanisms by which replication-transcription conflicts generate spontaneous mutations within the promoter and the coding region of a gene.

Head-on conflict between replication and transcription impedes replication fork progression and has been proposed to be more mutagenic than co-directional conflict (Vilette et al., 1995, Vilette et al., 1996, Mirkin and Mirkin, 2005, Srivatsan et al., 2010). In B. subtilis, increased nonsynonymous substitution rates in the coding sequence of head-on genes have been observed and proposed to be generated by replication-transcription conflict through the recruitment of the error-prone translesion polymerase PolY1 (Paul et al., 2013, Million-Weaver et al., 2015). However, a role for PolY1 in spontaneous mutagenesis in B. subtilis remains unclear (Duigou et al., 2004). Other studies using either a forward mutation assay or a mutation accumulation approach did not find strong evidence for replication-transcription conflict in generating base pair substitutions in coding sequences (Sankar et al., 2016, Schroeder et al., 2016). Therefore, the contribution of replication-transcription conflict on spontaneous base pair substitution rates in coding sequences has remained unclear.

Replication-transcription conflicts contribute strongly to insertions and deletions within a gene’s coding sequence. Insertions and deletions were found to coincide with the locations where replication and RNA polymerase first encounter each other, implicating replication fork stalling due to conflicts with transcription as a trigger for template switching by DNA replication or double-strand breaks that result in insertions and deletions (Dutta et al., 2011, Sankar et al., 2016). In addition, collision of the replication fork with either the transcription initiation complex or transcriptional repressors can occur (Mirkin et al., 2006), generating insertions and deletions in promoter regions (Vilette et al., 1992, Vilette et al., 1996, Sankar et al., 2016).

In addition, T → C transitions at the promoter increase when transcription is driven head-on with replication (Yoshiyama et al., 2001, Sankar et al., 2016). One explanation for this is that head-on conflicts interfere with trainscription initiation at a step after promoter open complex formation, either by promoting abortive initiation or prolonging single-stranded DNA exposure. If promoter substitutions are a major source of mutagenesis due to head-on replication-transcription conflict, it may explain why essential genes are highly enriched to be transcribed co-directionally with replication, as promoter mutations in head-on genes will result in strong purifying selection against a head-on orientation for essential genes.

Spontaneous induction of the SOS response

Although the SOS response is often thought of as being induced by externally imposed stress such as exposure to ultraviolet light and DNA-damaging agents, approximately 1% of exponentially growing E. coli are undergoing the SOS response in the absence of exogenous stress (Pennington and Rosenberg, 2007). Three distinct DNA polymerases, Pol IV, Pol V, and Pol II are upregulated during the E. coli SOS response. Pols IV and V are Y-family DNA polymerases (Jarosz et al., 2007). Pol IV is encoded by the gene dinB. Pol IV is accurate when replicating over adducts on the N² position of guanine nucleotides (Jarosz et al., 2006, Yuan et al., 2008), but Pols IV and V are both error-prone on undamaged DNA (Goodman and Woodgate, 2013). Increased expression of Pols IV and V, such as during SOS in E. coli, yields an elevated mutation rate (MacLean et al., 2013). Pol IV makes single-nucleotide deletions due to its ability to extend misaligned primer-template pairs (Wagner et al., 1999). To a lesser extent, Pol IV also produces base pair substitutions (Wagner et al., 1999, Tang et al., 2000). Pol V is encoded by the genes umuC and umuD. Pol V has been proposed to be primarily responsible for the increased mutation rate during the SOS response; it efficiently incorporates nucleotides across from abasic sites and UV lesions such as thymidine dimers in DNA and produces transversions (Steinborn, 1978, Tang et al., 1999, Tang et al., 2000, Curti et al., 2009, Robinson et al., 2015). Pol II is encoded by the polB gene. Pol II has exonucleolytic proofreading activity, and is therefore quite accurate and does not contribute greatly to SOS-induced mutagenesis (Banach-Orlowska et al., 2005). Rather, Pol II may correct errors made by Pols IV and V or remove mismatches at the nascent strand 3′ end prior to recruitment of Pols IV and V for extension, thereby reducing mutagenesis during SOS (Banach-Orlowska et al., 2005). Pols IV, V, and II can displace Pol III at the replication fork during SOS (Sutton and Duzen, 2006, Furukohri et al., 2008, Curti et al., 2009, Kath et al., 2014, Kath et al., 2016), so it is reasonable to expect that spontaneous induction of the SOS response may contribute to mutagenesis during unstressed conditions in E. coli. However, despite SOS induction in about 1% of exponentially growing E. coli in the absence of exogenous stress (Pennington and Rosenberg, 2007), E. coli lacking Pols IV and V, or all three SOS-induced DNA polymerases have no detectable change in base pair substitution or insertion/deletion rate in mutation accumulation experiments, suggesting they may not provide a substantial source of mutagenesis during normal growth (Foster et al., 2015).

The requirements for SOS induction differ substantially between B. subtilis and E. coli (Simmons et al., 2009). In B. subtilis and many other bacteria, two replicative DNA polymerase subunits, PolC and DnaE, are involved in Pol III function (Dervyn et al., 2001, Foster et al., 2003, Titok et al., 2006). In B. subtilis, DnaE is required to extend RNA primers during DNA replication prior to switching with the more processive and proofreading capable PolC that synthesizes the rest of the lagging strand (Sanders et al., 2010). DnaE is error-prone due to its lack of proofreading activity, and expression of DnaE is induced during SOS (Le Chatelier et al., 2004). PolY1 (YqjH) and PolY2 (YqjW) in B. subtilis are homologous to E. coli Pol IV and UmuC, respectively (Sung et al., 2003). B. subtilis does not have a clear Pol II homolog. PolY1 is constitutively expressed in B. subtilis (Duigou et al., 2004). PolY2 is SOS-induced and is required for UV-induced mutagenesis (Duigou et al., 2004). Although overexpression of either PolY1 or Y2 increased spontaneous mutagenesis, deletion of yqjH, yqjW, or both genes, had no detectable effect on spontaneous mutagenesis (Duigou et al., 2004). It is currently unclear whether spontaneous SOS induction during normal growth plays a role in spontaneous mutagenesis in B. subtilis.

Homologous recombination

Homologous recombination can cause genome rearrangements via crossover between non-allelic ribosomal RNA operons on sister chromosomes (Reams et al., 2014). Additionally, homologous recombination between similar genes within a genome causes base pair substitutions in a process termed “gene conversion” (Abdulkarim and Hughes, 1996, Paulsson et al., 2017). However, mutagenic mechanisms of homologous recombination remain poorly characterized. This is largely attributable to the methods commonly used to detect mutations. Mutation reporters often yield information preferentially on base pair substitutions and small insertions and deletions and will not detect large structural changes in a genome that are often associated with homologous recombination. Extraction of information on large genome rearrangements in mutation accumulation studies is not frequently performed, but it could reveal more details on the rates and mechanisms of mutagenesis due to homologous recombination. Recently, mutation accumulation methodology has been extended to examine the movement of a specific class of mobile genetic element, insertion sequences, in E. coli.

Insertion sequences

Insertion sequences (ISs) are a class of mobile genetic element present in diverse bacteria. ISs usually bear only the open reading frame encoding the transposase that encourages their mobilization [reviewed in (Siguier et al., 2006, Darmon and Leach, 2014)]. IS elements represent a previously underappreciated source of spontaneous mutagenesis in bacteria because their mobilization within a genome has been difficult to detect using classical genetic methods. A recent analysis of IS dynamics used whole genome resequencing data from 520 E. coli mutation accumulation lines from various studies (Lee et al., 2016). Strikingly, the rate of insertion of IS elements is within an order of magnitude of the rates of base pair substitution and insertion/deletion per genome per generation in Deinococcus radiodurans and E. coli (Lee et al., 2014, Long et al., 2015a, Lee et al., 2016, Shewaramani et al., 2017). ISs insert at a rate of 3.5 × 10⁻⁴ per genome per generation, and homologous-recombination-mediated deletions between ISs occur at a rate of 4.5 × 10⁻⁵ per genome per generation in E. coli. By comparison, the base pair substitution rate in E. coli is ≈ 1 × 10⁻³ per genome per generation (Foster et al., 2015).

It was determined that nine of the fourteen IS families in E. coli MG1655 were active, as five of the families showed no evidence of movement in any mutation accumulation line. IS1A and IS5A families were the most active, accounting for ≈ 80% of all insertions and deletions due to ISs. Another mutation accumulation study compared the mutation rates and spectra of E. coli mutation accumulation lines passaged under either aerobic or anaerobic conditions. The major finding of this study was that the insertion rate of a specific IS element, IS150, increased under anaerobic conditions (Shewaramani et al., 2017). The mechanism by which anaerobic growth promotes IS150 mobilization is currently unclear. Mutation accumulation studies have thus provided evidence for ISs being a common source of spontaneous mutagenesis in the E. coli genome.

Concluding remarks

Spontaneous mutagenesis is a major source of genetic diversity leading to evolution of species and antibiotic resistance in bacteria. Recent progress in spontaneous mutagenesis has demonstrated that spontaneous mutations arise from endogenous cellular processes such as DNA replication, transcription and replication-transcription conflict. Local sequence context heavily influences base pair substitution rate due to a combination of effects on base pair misinsertion and proofreading by DNA polymerases, and DNA mismatch repair efficiency. In addition, spontaneous mutagenesis extends far beyond point mutations in coding sequences of genes, with many mutations occurring in regulatory elements and intergenic regions. Finally, movement of and homologous recombination between insertion sequences provides a substantial source of spontaneous mutagenesis in bacteria. Further work on spontaneous mutagenesis in bacteria will continue to provide the foundation for a deeper understanding of mutagenic mechanisms and their contributions to evolution, antibiotic resistance and cancer.

Biographies

Jeremy Schroeder is a postdoctoral fellow at the University of Wisconsin – Madison working to understand mechanisms promoting diversity within bacterial populations. Dr. Schroeder received his Ph.D. in Molecular, Cellular, and Developmental Biology at the University of Michigan.

Ponlkrit Yeesin is a Royal Thai Scholar and a graduate student in the Microbiology Doctoral Training Program at the University of Wisconsin – Madison. His research focuses on mechanisms of DNA replication and recombination in Gram-positive bacteria.

Lyle Simmons is an Associate Professor in the Department of Molecular, Cellular, and Developmental Biology at the University of Michigan. Dr. Simmons received his Ph.D. in Biochemistry and Molecular Biology from Michigan State University and was a postdoctoral fellow in the Department of Biology at MIT.

Jue Wang is a Professor in the Department of Bacteriology at the University of Wisconsin – Madison. Dr. Wang received her Ph.D. in Biochemistry from University of California, San Francisco and was a postdoctoral fellow in the Department of Biology at MIT.