Specificity of the DNA Mismatch Repair System (MMR) and Mutagenesis Bias in Bacteria

Abstract

The mutation rate of an organism is influenced by the interaction of evolutionary forces such as natural selection and genetic drift. However, the mutation spectrum (i.e., the frequency distribution of different types of mutations) can be heavily influenced by DNA repair. Using mutation-accumulation lines of the extremophile bacterium Deinococcus radiodurans ΔmutS1 and the model soil bacterium Pseudomonas fluorescens wild-type and MMR⁻ (Methyl-dependent Mismatch Repair-deficient) strains, we report the mutational features of these two important bacteria. We find that P. fluorescens has one of the highest MMR repair efficiencies among tested bacteria. We also discover that MMR of D. radiodurans preferentially repairs deletions, contrary to all other bacteria examined. We then, for the first time, quantify genome-wide efficiency and specificity of MMR in repairing different genomic regions and mutation types, by evaluating the P. fluorescens and D. radiodurans mutation data sets, along with previously reported ones of Bacillus subtilis subsp. subtilis, Escherichia coli, Vibrio cholerae, and V. fischeri. MMR in all six bacteria shares two general features: 1) repair efficiency is influenced by the neighboring base composition for both transitions and transversions, not limited to transversions as previously reported; and 2) MMR only recognizes indels <4 bp in length. This study demonstrates the power of mutation accumulation lines in quantifying DNA repair and mutagenesis patterns.

Introduction

Studying mutations not only provides insight into the evolution of organisms but also into pathogenesis caused by either germline or somatic mutations. Previous studies have shown that mutations are nonrandom with respect to genomic regions and neighboring nucleotide contexts, as are the premutations that are fixed by repair systems (Benzer 1961; Jones et al. 1987; Friedberg et al. 2006). However, these biases have not been rigorously quantified because of the numerous assumptions and the limited genomic regions covered by reporter-construct methods applied in most earlier mutational studies. Quantifying such biases in a statistically rigorous manner requires a large number of mutations acquired in a nonselective fashion.

Fortunately, the mutation accumulation (MA) technique conquers these difficulties by imposing repeated single-individual bottlenecks on large numbers of parallel lineages over hundreds to thousands of cell divisions (Bateman 1959; Mukai 1964). During this process, the extreme population bottlenecks maximize random genetic drift, which greatly reduces the efficiency of selection, so that essentially all mutations except a tiny fraction with lethal fitness effects can proceed to fixation. Deep whole-genome sequencing and advanced bioinformatic tools further promote this experimental procedure by directly and accurately detecting mutations in the MA lines (Lynch et al. 2008). Furthermore, genome-wide mutagenesis prior to MMR involvement can be revealed in vivo by mutations accumulated in MMR-deficient (MMR⁻) strains (Lee et al. 2012; Foster et al. 2013; Long, Kucukyildirim, et al. 2015; Long, Sung, et al. 2015). By comparing the molecular mutation spectra of wild-type and MMR⁻ strains of phylogenetically diverse bacteria, general or lineage-specific patterns in mutagenesis and DNA repair can be directly and systematically studied.

MMR, one of the major DNA repair pathways, fixes mismatched base pairs in DNA (reviewed in Kunkel and Erie 2005). In this study, we ran MA experiments on an MMR⁻ strain of Deinococcus radiodurans R1 (the most radiation-resistant organism known on the earth) and both wild-type and MMR⁻ strains of another model bacterium Pseudomonas fluorescens SBW25 (Gram-negative). Together with recently published MA data sets for wild-type Bacillus subtilis subsp. subtilis NCIB 3610 (Gram positive, Sung et al. 2015), D. radiodurans BAA-816 (Gram positive, Long, Kucukyildirim, et al. 2015), Escherichia coli MG1655 (Gram negative, Lee et al. 2012; Long et al. 2016), Vibrio cholerae 2740-80, and V. fischeri ES114 (Gram negative, Dillon et al. 2017), we quantified the efficiency/specificity of the MMR system and mutagenesis bias in bacteria.

Results and Discussion

General information on all bacterial MA data sets is listed in table 1. Mutation details of wild-type, ΔmutS, and ΔmutL P. fluorescens SBW25 and ΔmutS1 D. radiodurans R1, which are first reported in this study, as well as reanalysis of results based on published raw sequences, are in supplementary data set D1: supplementary tables S1−S10, Supplementary Material online.

Table 1.

MA Data Sets.

Species	GC%	µBPS	ts/tv	µindel	µindel_CDS	µindel_NCDS in/del		N	G	References
Bacillus subtilis	43.35	3.28	3.23	1.18	0.69	2.16	0.32	50	5,078	Sung et al. (2015)
B. subtilis ΔmutS	43.35	101×a	3×	44×b	64×b	48×b	1.56b	19	2,000	Sung et al. (2015)
Deinococcus radioduransc	66.61	4.99	1.71	0.22	0.21	0.26	1.13	43	5,961	Long, Kucukyildirim, et al. (2015)
D. radiodurans ΔmutLc	66.61	4×	12×	21×	18×	51×	0.84	24	993	Long, Kucukyildirim, et al. (2015)
D. radiodurans ΔmutS1c	66.61	4×	9×	15×	13×	31×	0.53	29	3,101	This study
Escherichia coli	50.79	2.02	1.28	0.37	0.28	1.84	0.60	59	4,246	Lee et al. (2012); Sung et al. (2016)
E. coli ΔmutL	50.79	136×	34×	141×	–	–	–	34	375	Lee et al. (2012)
E. coli ΔmutSd	50.79	115×	39×	119×	80×	91×	1.33	12	763	Long et al. (2016)
Pseudomonas fluorescens	60.50	0.93	1.72	0.12	0.11	0.19	0.39	80	5,141	This study
P. fluorescensΔmutL	60.50	279×	48×	176×	149×	332×	11.89	30	1,465	This study
P. fluorescens ΔmutS	60.50	309×	45×	206×	165×	389×	11.77	30	1,454	This study
Vibrio cholerae	47.57	1.07	1.56	1.71	0.15	0.31	0.29	49	6,453	Dillon et al. (2017)
V. cholerae ΔmutS	47.57	85×	29×	142×	116×	234×	1.07	22	1,254	Dillon et al. (2017)
V. fischeri	38.35	2.07	0.86	5.68	0.35	2.03	0.58	48	5,187	Dillon et al. (2017)
V. fischeri ΔmutS	38.35	317×	37×	102×	121×	83×	0.77	19	810	Dillon et al. (2017)

^aMutation rates in MMR⁻ strains are shown in fold-change relative to the wild-type rates. GC% is the G/C content of the genome; µ_BPS and µ_indel are genome-wide mutation rates of base substitutions and indels per site per cell division; ts/tv is the ratio of transition to transversion mutations; µ_{indel_CDS}, µ_{indel_NCDS} are conditional indel mutation rates per site per cell division in coding and noncoding sequences, respectively; in/del is the ratio of insertions to deletions; N is the number of MA lines; G is the average number of cell divisions that each MA line passed during the experiment. All mutation rates are in units of ×10⁻¹⁰ per nucleotide site per cell division.

^bReanalysis of published MA-line sequences using GATK.

^cAll indels and BPSs are chromosomal.

^dBased on MA lines SA1–SA12 (Long et al. 2016).

Mutation Rates of P. fluorescens and D. radiodurans Strains

For P. fluorescens, we detected 253 base-pair substitutions (BPSs) in 80 SBW25 wild-type MA lines, 8,425 in 30 ΔmutS MA lines, and 7,529 in 30 ΔmutL MA lines. These yield BPS mutation rates of 9.31 ± 0.66 × 10⁻¹¹ (SEM; one of the lowest mutation rates in bacteria), 2.87 ± 0.12 × 10⁻⁸ and 2.59 ± 0.09 × 10⁻⁸ per nucleotide site per cell division for the wild-type, ΔmutS, and ΔmutL strains, respectively. Knocking out mutS elevates the mutation rate ∼309-fold, to a level that is highly similar to that of a naturally mutS-deficient P. fluorescens strain, 2.34 × 10⁻⁸ (Long, Sung, et al. 2015). Knocking out mutL inflates the P. fluorescens mutation rate ∼278-fold. The extremely high mutation-rate elevation of MMR⁻ lines makes clear that this species is one of the most efficient among bacteria in repairing BPS premutations by MMR (table 2).

Table 2.

MMR Efficiency of Different Types of Mutations in Bacteria.

Mutations	Bacillus subtilis	Deinococcus radiodurans	Escherichia coli	Pseudomonas fluorescens	Vibrio cholerae	V. fischeri
G:C>A:T	0.9925 (0.0007)	0.7533 (0.0270)	0.9867 (0.0027)	0.9982 (0.0002)	0.9887 (0.0015)	0.9973 (0.0003)
A:T>G:C	0.9922 (0.0007)	0.8779 (0.0161)	0.9975 (0.0007)	0.9995 (0.0001)	0.9967 (0.0008)	0.9993 (0.0001)
A:T>T:A	0.9327 (0.0152)	–a	0.7407 (0.1399)	0.9935 (0.0046)	0.8691 (0.0845)	0.9935 (0.0034)
A:T>C:G	0.9051 (0.0230)	–a	0.7055 (0.1079)	0.8933 (0.0335)	0.5056 (0.3096)	0.7321 (0.1213)
G:C>T:A	0.7749 (0.0719)	–a	0.8622 (0.1022)	0.9049 (0.0204)	0.4547 (0.2937)	0.8582 (0.0291)
G:C>C:G	0.9579 (0.0168)	0.7119 (0.1399)	0.7659 (0.2034)	0.9658 (0.0072)	0.9128 (0.0504)	0.9485 (0.0221)
Total BPS	0.9901 (0.0005)	0.7294 (0.0196)	0.9906 (0.0011)	0.9968 (0.0002)	0.9882 (0.0011)	0.9968 (0.0002)
Insertions	0.9907 (0.0017)	0.8991 (0.0408)	0.9938 (0.0016)	0.9985 (0.0005)	0.9969 (0.0014)	0.9918 (0.0019)
Deletions	0.9710 (0.0051)	0.9524 (0.0192)	0.9862 (0.0029)	0.9558 (0.0110)	0.9888 (0.0029)	0.9893 (0.0019)
Total indels	0.9830 (0.0022)	0.9335 (0.0188)	0.9905 (0.0015)	0.9952 (0.0009)	0.9930 (0.0016)	0.9902 (0.0013)
Indels_SSR	0.9852 (0.0009)	0.9618 (0.0152)	0.9925 (0.0013)	0.9980 (0.0001)	0.9974 (0.0009)	0.9935 (0.0011)
Indels_non_SSR	0.7628 (0.1113)	0.8126 (0.0828)	0.4573 (0.5607)	0.7991 (0.0784)	0.3892 (0.4617)	0.7064 (0.1615)
Indels_coding	0.9844 (0.0020)	0.9233 (0.0234)	0.9876 (0.0028)	0.9940 (0.0012)	0.9914 (0.0022)	0.9917 (0.0016)
Indels_noncoding	0.9792 (0.0039)	0.9676 (0.0256)	0.9890 (0.0022)	0.9975 (0.0011)	0.9957 (0.0020)	0.9880 (0.0025)
Ne	6.119×107	–	1.796×108	–	4.783×108	–

Note.—SEs, calculated with equation A1.19b in Lynch and Walsh (1998) assuming no covariance between wild-type and ΔmutS MA lines, are in parentheses. Each measure of repair efficiency represents the proportion of mutations repaired by MMR (1.0 implying that all premutations are completely repaired) and is calculated by (µ_ΔmutS−µ₊)/µ_ΔmutS, where µ_ΔmutS is the mutation rate of the ΔmutS MA lines and µ₊ is the mutation rate of the wild-type MA lines.

^aNegative number; Indels_SSR and Indels_non_SSR represent indel repair efficiencies in SSR and non-SSR regions; Indels_coding and Indels_noncoding are indel repair efficiencies in coding and noncoding regions; Ne, effective population size. Data sources: for B. subtilis, BPS data from Sung et al. (2015), indel data from this study; D. radiodurans, Long, Kucukyildirim, et al. (2015), this study; E. coli, (wild-type indel data were from Lee et al. 2012; all BPS and ΔmutS data were from Long et al. 2016); P. fluorescens, this study; V. cholerae and V. fischeri, Dillon et al. (2017); Ne data from Supplementary Information S1 of Lynch et al. (2016).

BPSs of the GC-rich P. fluorescens SBW25 wild-type strain do not show significant bias, with µ_G/C→A/T (the mutation rate in the A: T direction) being 9.85 × 10⁻¹¹ (95% Poisson confidence intervals: 8.30 × 10⁻¹¹ to 11.59 × 10⁻¹¹), and µ_A/T→G/C (the mutation rate in the G: C direction) being 8.39 × 10⁻¹¹ (6.66 × 10⁻¹¹ to 11.45 × 10⁻¹¹). Details of the mutation spectra for all P. fluorescens strains are shown in supplementary data set D1: supplementary tables S1−S3, Supplementary Material online and figure 1.

Fig. 1.

Conditional mutation rates of Pseudomonas fluorescens SBW25 wild-type and MMR⁻ strains. Error bars are SEM.

We also detected 32 small indels in the P. fluorescens wild-type MA lines, yielding a small-indel mutation rate of 1.18 ± 0.21 × 10⁻¹¹ per nucleotide site per cell division. The insertion/deletion ratio is 0.39, showing that this strain has a deletion bias. Knocking out mutS or mutL increases the small-indel mutation rates to 2.47 × 10⁻⁹ and 2.11 × 10⁻⁹ per nucleotide site per generation, respectively, and the insertion/deletion ratio to 11.77 and 11.89 (table 1). The insertion-bias in MMR⁻ and deletion-bias in MMR⁺P. fluorescens shows that MMR preferentially repairs insertions over deletions.

For D. radiodurans, 418 BPSs accumulated in the chromosomes of 29 ΔmutS1 MA lines (a full list of BPSs including plasmid BPSs are in supplementary data set D1: supplementary table S7, Supplementary Material online). The chromosomal BPS mutation rate for this strain is 1.84 ± 0.08 × 10⁻⁹ per nucleotide site per cell division, which is ∼3.3× of the wild-type chromosomal mutation rate (Mennecier et al. 2004; Long, Kucukyildirim, et al. 2015). This low-level elevation of the mutation rate makes D. radiodurans MMR one of the least efficient among bacteria (table 2). This may be compensated by other highly efficient alternative DNA repair pathways, such as the oxidative damage repair pathways (Mennecier et al. 2004; Long, Kucukyildirim, et al. 2015), or by multiple copies of the genome enhancing gene conversion as a repair strategy. The µ_A/T→G/C mutation rate 2.92 × 10⁻⁹ (95% Poisson confidence intervals: 2.54 × 10⁻⁹ to 3.33 × 10⁻⁹) is significantly higher than µ_G/C→A/T 1.25 × 10⁻⁹ (1.08 × 10⁻⁹ to 1.44 × 10⁻⁹), yielding a GC-bias close to that in the wild-type strain (Long, Kucukyildirim, et al. 2015).

75 small indels accumulated in the D. radiodurans ΔmutS1 chromosomes, yielding a chromosomal small-indel mutation rate of 3.31 ± 0.50 × 10⁻¹⁰. Deletions are more abundant than insertions, with an insertion/deletion ratio of 0.53, which is highly unusual in other MMR⁻ bacteria (table 1). The deletion bias in MMR⁻D. radiodurans implies that DNA polymerases of D. radiodurans generally have more forward slippage errors than backward ones. This might be explained by insertions requiring melting and replication of a previously replicated DNA fragment, whereas only skipping of unreplicated fragments is needed for deletions (Petrov 2002), assuming that no other repair pathways are involved in this process in D. radiodurans MMR⁻ strains.

Mutation Spectrum

Mutations in MMR⁻ MA lines reveal the primary mutation spectrum of organisms, unmodified by the subsequent MMR pathway. Among MMR⁻ strains of all six bacteria—B. subtilis, D. radiodurans, E. coli, P. fluorescens, V. cholerae, and V. fischeri, transitions are ∼16- to 82-fold more abundant than transversions, in contrast to only 0.86- to 3.23-fold in MMR-functional strains (table 1). The higher relative abundance of transitions in MMR⁻ strains reflects the fact that major DNA polymerases are prone to cause transition mutations (Curti et al. 2009).

Most indels in MMR⁻ strains are found in simple sequence repeats (SSR), which are tandem repeats of short DNA nucleotides, such as homopolymer runs, and microsatellite repeats, and are usually generated by unequal crossover or polymerase slippage during replication of SSRs (Levinson and Gutman 1987). For example, 81–99% of small-indels are in SSR motifs in mutS-inactivated MA lines (supplementary data set D1: supplementary tables S5, S8, and S10). In order to explore patterns of the indel mutation rate at SSR motifs, we analyzed the indel mutation rate at 3–9 bp homopolymer runs for each category (≥10 bp homopolymers are extremely rare or missing in the genomes of the four studied bacteria). As shown in figure 2, for either A/T or G/C homopolymers, the indel mutation rate per homopolymer per generation increases with homopolymer length, presumably due to a decline in proofreading efficiency of DNA polymerases as homopolymer-run length increases (Lujan et al. 2015). The indel mutation rate at long homopolymer motifs can be orders of magnitude higher than that of shorter ones; for example, the indel rate of 5′GGGGGGGGG3′ and 5′AAA3′ motifs in E. coli differs by 200,861×, the same order of magnitude increase in the indel rate at homopolymer runs in yeast MMR⁻ MA lines (Lujan et al. 2015). The very high indel mutation rate of long homopolymers could also explain the rarity of longer homopolymers in organisms’ genomes, which may have been eroded by indels during long-term evolution.

Fig. 2.

Conditional indel mutation rates at different homopolymer-run lengths in MMR⁻ strains. Error bars are 95% Poisson confidence intervals.

The base composition of homopolymers also influences the indel mutation rate, with G or C homopolymers having a significantly higher indel mutation rate than A or T homopolymers (fig. 2). This may be explained by the stronger base-stacking interaction of G: C base pairs compared with A: T (Sagher et al. 1999), as well as by stronger base-pairing of G: C (Lujan et al. 2015). Thus, the relative abundance of G/C and A/T homopolymers in a genome influences the indel mutation rate.

MMR Repair Specificity

Using mutation rates of wild-type and MMR⁻ MA lines, we calculated the repair efficiency of different types of mutations (table 2). The repair efficiency and effective population size (Ne) are not correlated with each other, though there are only three available data points. This may be a simple consequence of selection operating on the overall mutation rate, leaving degrees of freedom for individual components to evolve in mutually compatible upward/downward fashions, e.g., a local decrease in MMR efficiency may be balanced by an increased in accuracy at the polymerization and/or proofreading steps (Lynch 2012).

Although transitions are preferentially repaired over transversions in general, we find that MMR-repair specificity is highly variable among bacterial species and different types of base-pair substitutions (table 2). Our results suggest that A: T→T: A, A: T→C: G, and G: C→T: A premutations are not repaired by MMR in D. radiodurans (tables 1 and 2). MMR specificity for insertions/deletions in D. radiodurans is also unique: contrary to all other observed bacteria, deletions are repaired more efficiently than insertions in D. radiodurans. This is consistent with a prior argument that the MMR system in D. radiodurans is extremely divergent from that of most other bacteria (Mennecier et al. 2004). Considering the insertion/deletion ratio of 1.13 in the wild-type strain versus the above ΔmutS1 ratio of 0.53 (table 1), MMR of D. radiodurans preferentially repairs deletions. The deletion-preference of MMR might be associated with recombination-bias caused by the polyploid nature of the D. radiodurans genome (four to ten ploidy). Alternatively, this might be an adaptive trait, which has evolved under the extreme radiation level that all other bacteria cannot tolerate, assuming D. radiodurans is geographically distributed in high-radiation habitats.

We also find some general features of MMR specificities in repairing indels across all six bacteria: only indel rates of 1–3 bp size are significantly lower in wild-type than in MMR⁻ strains, that is, these short indels are efficiently repaired by MMR, but larger ones are not (supplementary data set 1: supplementary tables S13 and S14, Supplementary Material online). This finding is highly consistent with previous in vitro and in vivo studies on E. coli MMR repair of indels (Learn and Grafstrom 1989; Parker and Marinus 1992; Carraway and Marinus 1993), which found that MMR only repairs insertion–deletion loops smaller than 5 bp and only 1–3 bp loops efficiently. Such indel-repair specificity of MMR appears to be explained by the biochemical property of MutS, which binds to heteroduplexes with 1–4 unpaired bases, but not to those with five or more unpaired bases (Learn and Grafstrom 1989; Parker and Marinus 1992; Carraway and Marinus 1993). Such specificity is not only influenced by indel size but also by the genomic regions within which indels occur, for example, indels occurring within SSR motifs are preferentially repaired relative to those at non-SSR genomic regions (table 2 and supplementary data set D1: supplementary table S11, Supplementary Material online), consistent with previous findings in eukaryotic organisms (Sia et al. 1997; Denver et al. 2005). The low MMR repair efficiency at non-SSR regions may be explained by the possibility that the key MMR protein MutS preferentially binds low-complexity DNA regions.

Flanking nucleotides such as guanines or cytosines can elevate the base-substitution mutation rate by orders of magnitude via their differential base-stacking and base-pairing power (Yakovchuk et al. 2006; Lee et al. 2012, 2014; Long, Sung, et al. 2015; Sung et al. 2015). Nucleotide-context was also reported to influence MMR efficiency for repairing transversions in E. coli (Jones et al. 1987). Here, we explore such context-dependent MMR efficiency (context is defined as the content of the two flanking nucleotides on the same DNA strand) in the three non-E. coli bacterial species in this study. We observed higher repair efficiency in contexts with at least one adjacent G: C pair for some species. This trend applies to both transition and transversion mutations, and is not limited to transversions as previously reported (Jones et al. 1987). Contexts without G: C pairs are usually among the ones with the lowest repair efficiency, such as 5′AGA3′ (fig. 3 and supplementary data set D1: supplementary table S12, Supplementary Material online). By contrast, we did not observe context-dependency of the indel repair-efficiency, which may be convoluted by that most indels occur in SSRs.

Fig. 3.

Heatmap of MMR repair efficiency in different bacteria. Trinucleotides on the left show the nucleotide contexts in the 5′ to 3′ direction (reverse-complement of one nucleotide-context is deemed as the same context, due to the uncertainty of the DNA strand of each mismatch causing the base-substitution), with the focal bases in the middle. Red bars with stars denote 100% repair efficiency, an outcome of inadequate mutation sample size in the wild-type lines.

Correlation between the Base-Substitution and Indel Mutation Rate

Because small indels and base substitution experience similar DNA replication and repair processes, their rates are likely not independent of each other. After pooling the Pseudomonas fluorescens data set in this study with another 23 published MA data sets of various MMR-functional bacteria (supplementary table S15, Supplementary Material online), we find a strong positive correlation between the base-substitution and small-indel mutation rates (Pearson’s correlation test, r = 0.97, P = 1.72 × 10⁻¹⁴), consistent with another recent study using both eukaryotic and prokaryotic organisms (Sung et al. 2016). This correlation still holds even when MMR is inactive (r = 0.84, P = 0.04; supplementary table S15, Supplementary Material online). The ratios of small-indel to base-substitution mutation rates are similar in different bacteria, regardless of whether they are MMR functional, 0.22 ± 0.04 (SEM), or dysfunctional, 0.16 ± 0.03.

The aforementioned shared DNA-replication and repair machinery could be one cause of the correlation between base-substitution and indel mutation rates. Another potential explanation is an elevation of the base-substitution mutation rate around indel hotspots (McDonald et al. 2011), owing to indels stalling high-fidelity DNA polymerases and recruiting error-prone translesion DNA polymerases that increase the downstream mutation rate.

Conclusions

We report the genomic mutation rate and spectrum of wild-type and MMR⁻P. fluorescens, and ΔmutS1 D. radiodurans. We find that P. fluorescens has one of the lowest mutation rates in bacteria and the highest MMR repair efficiency among the six studied bacteria. Using mutations in these original data sets and previously reported studies, for the first time, we quantify the efficiency and specificity of MMR over different types and genomic locations of mutations. We also discover that the MMR of D. radiodurans preferentially repairs deletions over insertions, thus causing a unique insertion-bias in this extremophile bacterium. Some features of mutagenesis and MMR are shared among all studied species: 1) the influence of the flanking nucleotide base-composition of a mismatch on MMR efficiency, with MMR preferentially repairing sites with at least one flanking G/C nucleotide; 2) G/C homopolymers having higher indel mutation rates than A/T homopolymers; and 3) MMR specifically repairing indels with 1 − 3 base pairs. This research demonstrates the effectiveness of mutation accumulation techniques in studying the specificity of the DNA repair systems.

Materials and Methods

MA Lines Information

The Pseudomonas fluorescens SBW25 wild-type strain was provided by the Vaughn Cooper lab, University of Pittsburgh, and we created 80 MA lines from a single-cell ancestor. We single-colony transferred MA lines every other day for 244 times per line. The P. fluorescens SBW25 ΔmutS (30 MA lines) and ΔmutL (30 MA lines) strains were provided by Angus Buckling, University of Exeter Cornwall Campus, UK (Pal et al. 2007; O’Brien et al. 2013), and MA lines from these mutator strains were transferred 71 times on average. The Deinococcus radiodurans R1 ΔmutS1 (we initiated 30 MA lines, line D33 was removed from the final analysis due to cross contamination detected after genome sequencing) was from the Suzanne Sommer lab, Université Paris-Sud 11, Orsay, France and MA lines from this strain were transferred ∼130 times. We cultured all P. fluorescens MA lines on nutrient agar (Becton, Dickinson and Company, Sparks, MD) at 25°C; we grew and single-colony transferred D. radiodurans ΔmutS1 MA lines at 30°C every other day on nutrient agar supplemented with 1% glucose. About every five weeks, we estimated cell divisions that the MA lines had passed by CFU (colony forming units) of diluted single colonies from ten randomly selected MA lines; the mean cell divisions from a single cell to a colony were estimated by log₂(CFU). The total cell division number of each MA line is the product of the grand mean (21.1 cell divisions for P. fluorescens wild-type, 20.5 for P. fluorescens ΔmutS, 20.7 for P. fluorescens ΔmutL; 23.8 for D. radiodurans ΔmutS1) of all cell division estimates and the total transfer number for each line.

Genome Sequencing

We extracted DNA of the final MA lines using the Wizard Genomic DNA Purification Kit (Promega, Madison, WI). DNA libraries were constructed using the Nextera DNA Library Preparation Kit (Illumina). We then size-selected the libraries for an insert size of 300 bp and sequenced them using HiSeq2500 2 × 150 rapid run at the Hubbard Center for Genome Studies, University of New Hampshire. Median depths of coverage were 70× (P. fluorescens wild-type), 62× (P. fluorescens ΔmutS), 78× (P. fluorescens ΔmutL) and 172× (D. radiodurans ΔmutS1). All raw sequence reads first reported in this study and those used but not published in previous studies (Long, Kucukyildirim, et al. 2015; Long, Sung, et al. 2015) were deposited in NCBI SRA (BioProject No. PRJNA397203).

Mutation Analyses

We trimmed off adaptors in raw reads with Trimmomatic 0.32 (Bolger et al. 2014), and then mapped reads to the reference genome using BWA-0.7.10 mem (GenBank genome accession numbers: NC_012660.1 for P. fluorescens SBW25; NC_000958.1—plasmid MP1, NC_000959.1—plasmid CP1, NC_001263.1—chromosome 1, NC_001264.1—chromosome 2 for D. radiodurans R1; GCA_000186085.1 for B. subtilis subsp. subtilis NCIB 3610) (Li and Durbin 2009). Duplicate reads were removed using picard-tools-1.141 and reads mapping around indels were realigned using GATK-3.5; SNP and indel discovery was performed with standard hard filtering parameters described by GATK Best Practices recommendations (except that Phred-scaled quality score QUAL >100 and RMS mapping quality MQ >59 for both variant and invariant sites) (McKenna et al. 2010; DePristo et al. 2011; Van der Auwera et al. 2013). Base-pair substitutions and small indels were called using the UnifiedGenotyper in GATK. We also required >99% of reads in a line to determine the line-specific consensus nucleotide at a candidate site—1% was set to allow for aberrant reads originated from sequencing errors, not absolutely pure indexes during library construction, or barcode degeneracy during sequence demultiplexing. Mutation rate µ was calculated by µ$=\frac{m}{\sum _1^{n}N\times T}$, where m is the total number of mutations pooled from all MA lines, n is the total number of lines, N is the analyzed sites in one line and T is the number of cell divisions the MA line passed.

Context-Dependency of MMR Repair

For each bacterial species, a 3-bp sliding window with a 1-bp step size was used to parse out the contexts in trinucleotides in the entire genome. Mutation rate at a certain context—µ_5′XXX3′ was calculated by:

$${\mu }_{5\mathrm{\text{'}}\mathrm{X}\underset{\_}{\mathrm{X}}\mathrm{X}3\mathrm{\text{'}}}=\frac{n}{N\times G\times L}$$

where n is the pooled number of mutations in both the 5′XXX3′ context and its reverse complement context across all MA lines of the species, N is the total number of both the trinucleotide 5′XXX3′ and its reverse complement in the same strand across the entire genome, G is the average number of generations passed in each MA line, and L is the total number of MA lines. We combined nucleotide triplets that are reverse complemented, for example, 5′GAT3′ is taken as the same context as 5′ATC3′, due to the uncertainty of the DNA strand of each mismatch causing the base-substitution.