Mutator Suppression and Escape from Replication Error–Induced Extinction in Yeast

Abstract

Cells rely on a network of conserved pathways to govern DNA replication fidelity. Loss of polymerase proofreading or mismatch repair elevates spontaneous mutation and facilitates cellular adaptation. However, double mutants are inviable, suggesting that extreme mutation rates exceed an error threshold. Here we combine alleles that affect DNA polymerase δ (Pol δ) proofreading and mismatch repair to define the maximal error rate in haploid yeast and to characterize genetic suppressors of mutator phenotypes. We show that populations tolerate mutation rates 1,000-fold above wild-type levels but collapse when the rate exceeds 10⁻³ inactivating mutations per gene per cell division. Variants that escape this error-induced extinction (eex) rapidly emerge from mutator clones. One-third of the escape mutants result from second-site changes in Pol δ that suppress the proofreading-deficient phenotype, while two-thirds are extragenic. The structural locations of the Pol δ changes suggest multiple antimutator mechanisms. Our studies reveal the transient nature of eukaryotic mutators and show that mutator phenotypes are readily suppressed by genetic adaptation. This has implications for the role of mutator phenotypes in cancer.

Author Summary

Organisms strike a balance between genetic continuity and change. Most cells are well adapted to their niches and therefore invest heavily in mechanisms that maintain accurate DNA replication. When cell populations are confronted with changing environmental conditions, “mutator” clones with high mutation rates emerge and readily adapt to the new conditions by rapidly acquiring beneficial mutations. However, deleterious mutations also accumulate, raising the question: what level of mutational burden can cell populations sustain before collapsing? Here we experimentally determine the maximal mutation rate in haploid yeast. We observe that yeast can withstand a 1,000-fold increase in mutation rate without losing colony forming capacity. Yet no strains survive a 10,000-fold increase in mutation rate. Escape mutants with an “anti-mutator” phenotype frequently emerge from cell populations undergoing this error-induced extinction. The diversity of antimutator changes suggests that strong mutator phenotypes in nature may be inherently transient, ensuring that rapid adaptation is followed by genetic attenuation which preserves the beneficial, adaptive mutations. These observations are relevant to microbial populations during infection as well as the somatic evolution of cancer cells.

Introduction

Accurate DNA replication ensures the faithful transmission of genetic information between mother and daughter cells. To accomplish this important task, organisms have evolved a network of conserved pathways that govern DNA replication fidelity (reviewed in [1]). Polymerase proofreading and postreplication mismatch repair (MMR) are key determinants of fidelity, functioning to correct errors introduced by DNA polymerases during cell division (reviewed in [2]–[4]). Defects in these and other DNA repair pathways produce mutator phenotypes, which are characterized by increased rates of spontaneous mutation.

Mutator phenotypes arise spontaneously in nature and have mixed biological consequences (reviewed in [5]–[10]). In Escherichia coli and other bacteria, changing environmental conditions favor high mutation rates, which increase the likelihood of genetic adaptation [11]–[16]. However, after adaptation, mutator bacteria progressively lose fitness as they accumulate deleterious mutations in other genes [14], [17], and clones with lower mutation rates can evolve from mutator populations [14], [16], [18]–[20]. Thus, mutation rates in E. coli rise and fall as populations cycle through periods of adaptive and non-adaptive growth.

Mutators also impact eukaryotes. In mammals, mutator phenotypes fuel oncogenesis by providing the genetic diversity necessary for emergence of malignant clones [21], [22]. Many sporadic human tumors show signs of an elevated mutation rate [23], and inherited defects in polymerase proofreading [24]–[26] or MMR (reviewed in [27], [28]) confer mutator phenotypes and increase cancer risk. In the budding yeast Saccharomyces cerevisiae, loss of proofreading or MMR also elevates spontaneous mutation [29]–[34], and defective MMR can facilitate adaptation to changing environments [35], [36].

Similar to bacteria, eukaryotic mutator alleles are detrimental in the long-term. Deleterious mutations accumulate faster in mutator compared to non-mutator yeast strains [37], [38], and mutators eventually become extinct in a mutational meltdown process after serial passage through population bottlenecks [39]. This decline is accelerated in yeast with extreme mutation rates. Diploids that are homozygous defective for both proofreading and MMR grow slowly and have mutations rates that are elevated 10,000-times above wild-type levels [40]–[42]. Double-mutant spores germinate but arrest at various cell-cycle stages after 6–7 mitotic divisions [40], suggesting that the accumulation of DNA replication errors drives the extinction of haploid mutator strains.

Here, we experimentally define the threshold of error-induced extinction in haploid S. cerevisiae and show that cells readily escape extinction via genetic suppression. These escape mutants emerge rapidly and carry second-site mutations that suppress the mutator phenotype. Our findings show that mutators are intrinsically unstable and that spontaneous suppressors moderate high mutation rates in yeast.

Results

Abrupt Loss in Viability with Increased Mutation Rate

To obtain a range of mutator strains suitable for defining the maximal mutation rate in yeast, we conducted a mutagenesis screen of the POL3 gene, which encodes the catalytic subunit of DNA polymerase δ (Pol δ). We used a plasmid shuffling strategy [29], [43] to introduce mutated pol3 alleles into strains proficient (MSH6) or deficient (msh6Δ) for MMR (Figure S1). The screen focused on conserved residues in the proofreading exonuclease domain of Pol δ [44], [45] that, when mutated in msh6Δ cells, are expected to preferentially increase base-substitution and ±1 frameshift mutations [40], [46]–[49].

Our analysis identified 21 amino acid substitutions in the Pol δ proofreading domain that individually conferred a range of increased spontaneous mutation frequencies (Figure S2). These alleles had no observable effect on colony formation in MSH6 cells. However, four alleles (pol3-01, pol3-F406A, pol3-D407A and pol3-Y516F) did not yield visible colonies when shuffled into msh6Δ cells. This result is consistent with previous reports of synthetic lethality between the proofreading-deficient pol3-01 allele and MMR-defective alleles pms1Δ, msh2Δ, or msh6Δ [40], [41], [50].

To determine whether the loss of growth capacity correlated with mutator strength, we quantified the spontaneous mutation rates of a subset of pol3 alleles in the presence or absence of MSH6 (Figure 1A). Alleles that imparted a 2- to 8-fold increase in the mutation rate of MSH6 cells (R459A, G400A, Y401A, D396A, Y410A, K491R and D463A) were compatible with survival when MSH6 was deleted. These pol3 msh6Δ double-mutants had mutation rates that were 15- to 150-times greater than the corresponding pol3 MSH6 strains, consistent with synergy between pol3 mutators and msh6Δ. In contrast, pol3 alleles that increased the mutation rate 25- to 50-fold in MSH6 cells (D407A, pol3-01, Y516F and F406A) conferred a loss of colony-forming capacity in msh6Δ cells (<1 colony/10⁵ cells plated; Figure 1B). Thus, the transition to no colony formation occurred over a narrow range of increasing mutation rates. This abrupt loss of growth capacity implies the existence of a threshold for error-induced extinction.

Figure 1. Evidence for a threshold of error-induced extinction.

A) Entry into error-induced extinction. Mutated pol3 alleles were introduced into haploid MSH6 and msh6Δ yeast by plasmid shuffling (Figure S1), and mutation rates were measured by fluctuation assays and calculated using the maximum likelihood method. Each bar represents the spontaneous mutation rate, expressed as canavanine-resistant (Can^r) mutants per cell division, conferred by a specific POL3 allele in MSH6 or msh6Δ cells. Mutation rate values (x 10⁻⁷) are indicated on each column. Error bars show 95% confidence intervals. WT, wild-type POL3; black, MSH6; gray, msh6Δ; X, no growth. B) Synthetic lethality of strong pol3 mutator alleles with msh6Δ. Serial dilutions of haploid yeast containing POL3–URA3 and pol3–LEU2 plasmids were plated on SC FOA medium to select for cells that spontaneously lost POL3–URA3. Similar numbers of cells (∼10⁵, 10⁴ and 10³) were plated for each set of alleles in the MSH6 and msh6Δ strains. Failed growth of msh6Δ cells carrying pol3-D407A or pol3-01 indicates synthetic lethality (right two panels). pol3-F406A and pol3-Y516F also failed to support colony formation in msh6Δ cells (not shown). Note the small size of pol3-D463A msh6Δ colonies.

Mutants That Escape Error-Induced Extinction

During our shuffling experiments, we observed occasional colonies that escaped pol3 msh6Δ synthetic lethality (Figure 2A). We speculated that second-site changes in Pol δ might rescue yeast from error-induced extinction by increasing DNA replication fidelity and thereby reducing the spontaneous mutation burden. To test this idea, we sequenced pol3 plasmids from error-induced extinction (eex) mutants that escaped synthetic lethality between msh6Δ and pol3-01, pol3-F406A or pol3-D407A. The plasmids retained the original pol3 mutator alleles, but also harbored additional second-site mutations in each pol3 sequence. Our initial experiment yielded three eex mutants encoding single amino-acid substitutions in Pol δ (E594G or W821C in pol3-01; T711P in pol3-D407A) and two mutants with multiple substitutions (K689E, S725L and I1076V in pol3-F406A; R470C and T655A in pol3-D407A). Another mutation (A894G) was found in a large-colony variant of pol3-D463A msh6Δ cells. When the second-site eex mutations were re-engineered into new plasmids together with their corresponding mutator alleles (pol3-01, pol3-F406A or pol3-D407A), they rescued colony-forming capacity in msh6Δ cells and decreased the mutation rate of MSH6 cells 10- to 33-fold (Figure 2B). The eex mutations appeared to be functionally interchangeable; T711P (the pol3-D407A suppressor) also rescued pol3-01 msh6Δ lethality, and either T711P or E594G (a pol3-01 suppressor) restored normal growth to pol3-D463A msh6Δ cells. Considered together, these initial findings suggested that eex mutations within POL3 confer escape from error-induced extinction by exerting an antimutator phenotype.

Figure 2. Escape from error-induced extinction.

A) Emergence of colonies that escape pol3-01 msh6Δ synthetic lethality. Each segment of an FOA-containing SC plate (eight segments per plate) was streaked with an individual colony of POL3–LEU2 POL3–URA3 msh6Δ (left) or pol3-01–LEU2 POL3–URA3 msh6Δ (right) cells to select for loss of the POL3–URA3 plasmid (see Figure S1). Resultant POL3 msh6Δ cells formed abundant visible colonies (left), whereas pol3-01 msh6Δ cells did not (right). Colonies that escape pol3-01 msh6Δ synthetic lethality (eex mutants) arose at low frequency near the outer margins of the plate (circled) where cell densities were highest. Similar results were seen when pol3-F406A or pol3-D407A were shuffled into msh6Δ cells (not shown). B) Antimutator effects of eex alleles encoding second-site changes in Pol δ. Rates of spontaneous mutation to canavanine-resistance (Can^r) conferred by pol3-01, pol3-F406A or pol-D407A alone (filled symbols) and combined with intragenic eex alleles (open symbols) were determined in MSH6 cells. Downward arrows illustrate the reduction in mutation rates (i.e., the antimutator effect) caused by the second-site, amino-acid substitutions indicated beneath each datum point. Error bars show 95% confidence intervals. C) Representative plate from large-scale screen for eex mutants. Approximately 10⁶ cells from multiple independent pol3-01–LEU POL3–URA msh6Δ parent colonies were plated separately in ∼1-cm spots on FOA-containing SC medium. LEU-only and POL3–LEU plasmids were also shuffled into msh6Δ cells as controls. FOA-resistant colonies arose at varied frequencies from each parent clone. Insert, magnified view showing colonies that are candidate eex mutants. plasmid, LEU-only plasmid with no POL3 gene. D) Mutation rates of eex mutants. Rates of spontaneous mutation to FOA-resistance (FOA^r) were measured in a MMR-proficient strain with a chromosomal URA3 reporter gene. Each datum point represents a different POL3 allele. Mutation rates were determined from multiple independent fluctuation analyses of each allele. Confidence intervals (95%) are shown for POL3 and pol3-01. Mutation rates and confidence intervals of individual eex alleles are in Table 1 and Table 2.

To obtain a broader view of escape mechanisms, we performed a large-scale screen for mutants that suppress the synthetic lethality between pol3-01 and msh6Δ (Figure 2C). Mutants emerged from nearly every pol3-01 msh6Δ parent clone, and there was wide fluctuation in the number and size of mutant colonies, suggesting that escape variants arise randomly prior to selection on FOA. We isolated 113 independent eex mutants (Table S1). Seventy-four of these eex mutants carried pol3-01 plasmids that still conferred lethality in a fresh msh6Δ strain. We infer that these mutants harbor mutations in chromosomal genes that influence DNA replication fidelity. The remaining 39 eex mutants carried pol3-01 plasmids that did not cause synthetic lethality when isolated and independently re-shuffled into msh6Δ cells. DNA sequencing of these plasmids showed that, in addition to the pol3-01 allele, each plasmid contained a different secondary mutation in pol3. These secondary mutations encoded single amino-acid changes in Pol δ (Figure 3 and Figure 4) and rescued colony-forming capacity when engineered de novo into pol3-01 plasmids and shuffled into naïve msh6Δ cells. Consistent with our initial experiment, all of these intragenic eex mutations suppressed the pol3-01 mutator phenotype, as measured at two different genetic loci (Figure 2D and Table 1). The weakest eex alleles suppressed mutation rates three-fold, while the strongest suppressors lowered mutation rates to wild-type levels (Figure 2D). Thus, cells escape pol3-01 msh6Δ lethality by acquiring any one of a variety of second-site mutations that suppress the mutator effect of Pol δ proofreading deficiency.

Figure 3. Amino-acid changes in Pol δ eex mutants.

Aligned sequences of five B-family DNA polymerases: Saccharomyces cerevisiae Pol δ (S.c. pol d), Mus musculus Pol δ (M.m. pol d), Thermococcus gorgonarius (T.g. pol B), bacteriophage T4 (T4 pol), and herpes simplex virus 1 (HSV1 pol). Secondary structural elements of yeast Pol δ [74] are indicated below the alignment and color coded to depict their domain locations (see Figure 4A): rectangles, α-helices; arrows, β-strands; solid lines, loops; dotted lines, structure not solved. Conserved polymerase (Pol) and exonuclease (Exo) motifs are framed [44], [45], [120]. Amino-acid substitutions of interest in yeast Pol δ are placed underneath the alignment at the relevant positions, highlighted according to the following scheme: no highlight, pol3-01,eex mutations; green, pol3-D407A,eex mutations (R470C and T655A in one mutant, T711P in another); orange, pol3-F406A,eex mutations (three substitutions in the same mutant); blue, A894G mutation that rescued slow growth of pol3-D463A msh6Δ cells; yellow, pol3-01 (D321A,E323A); gray, pol3-t (D643N) and G447S (previously identified antimutator alleles; [121], [122]). Residues that increase polymerase fidelity when mutated in T4 or HSV1 are indicated by aqua boxes in the alignment [4], [84]-[89], [123].

Figure 4. Locations of eex amino-acid substitutions in the S. cerevisiae Pol δ structure.

A) Overall distribution of eex amino-acid substitutions. The catalytic subunit of yeast Pol δ is shown as a ribbon diagram with color-coded structural domains: amino, gray; exonuclease (Exo), red; palm, purple; fingers, blue; thumb, green. Other important elements are indicated as follows: DNA template strand, brown sticks; DNA primer strand, yellow sticks; incoming dCTP, green CPK sticks; metal ions, small black spheres; active-site residues, gray CPK sticks extending out from the α-carbon backbone in the palm and exonuclease domains. Residues changed by eex mutations are shown as light blue spheres. The asterisks mark adjacent E642K and D643N eex substitutions located on the solvent-exposed surface of Pol δ. Structure from [74] (Protein Data Bank accession code 3IAY). B) Amino-acid substitutions near the polymerase active site. Palm domain eex residues are shown as space-filling spheres (light blue) and labeled to indicate the amino-acid substitutions. Important non-mutated residues proximal to the eex substitutions are also shown as space-filling spheres (purple). The fingers and thumb domains have been removed for clarity. C) Amino-acid substitutions in the DNA binding track. View looking down the DNA helical axis. The primer (yellow) and template (brown) are held by a series of interactions along the DNA minor groove. eex residues are light blue. Amino acids positioned by eex residues and minor-groove ‘sensing’ residues in the palm domain (K813, K814 and R815; [74]) are shown as space-filling spheres colored according to domain as in panel (A). The three unpaired 5′ nucleotides of the template have been removed for clarity.

Table 1. Mutation Rates (×10⁻⁷) of pol3-01,eex Mutant Strains.

Allele		MSH6 a				msh6Δ b		msh6Δ effect e
		FOAr		Canr		Canr
pol3-01 c		23	(19, 26)	135	(110, 157)	—
+	G207R	6.2	(4.6, 8.1)	—		—
+	H879Y	5.7	(4.5, 7.0)	53	(41, 65)	11000	(7700, 15000)	209
+	Y808C	5.5	(4.2, 7.0)	31	(25, 37)	7700	(6300, 9000)	198
+	S968R	2.9	(2.1, 3.8)	34	(26, 43)	5200	(4100, 6300)	153
+	A894V	2.4	(1.6, 3.5)	33	(25, 40)	3100	(2500, 3700)	96
+	G400S	2.2	(1.6, 3.0)	—		—
+	G204D	2.1	(1.4, 2.9)	29	(22, 36)	4500	(3400, 5600) d	157
+	L411R	1.8	(1.0, 2.9)	—		—
+	E642K	1.6	(1.1, 2.4)	23	(17, 28)	2600	(2000, 3000)	113
+	F486S	1.6	(1.0, 2.3)	—		—
+	T711A	1.4	(1.0, 2.0)	33	(25, 41)	5900	(4700, 7100)	179
+	R839H	1.4	(0.9, 2.1)	—		—
+	W821C	1.3	(0.8, 1.9	33	(24, 41)	2600	(2100, 3000)	80
+	C365Y	1.3	(0.8, 1.9)	—		—
+	A786V	1.3	(0.8, 1.9)	12	(8, 15)	1600	(1200, 1900) d	135
+	F793I	1.2	(0.8, 1.9)	6	(4, 8)	1600	(1300, 1800)	258
+	R475G	1.2	(0.8, 1.8)	—		—
+	N610D	1.1	(0.7, 1.6)	—		—
+	L531P	1.0	(0.7, 1.6)	—		—
+	E594G	1.0	(0.6, 1.6)	26	(17, 35) d	1700	(1200, 2100)	63
+	R475I	0.9	(0.5, 1.5)	14	(10, 18)	2700	(2300, 3200)	197
+	D831G	0.9	(0.5, 1.3)	11	(8, 15)	1700	(1400, 1900)	150
+	G921D	0.9	(0.4, 1.3)	—		—
+	D643N	0.8	(0.4, 1.5)	—		—
+	V838A	0.8	(0.5, 1.3)	—		—
+	R923H	0.8	(0.4, 1.3)	—		—
+	E800K	0.7	(0.4, 1.2)	13	(10, 17)	1700	(1400 1900)	126
+	K891T	0.7	(0.4, 1.2)	18	(14, 23)	1900	(1500 2300)	103
+	H620Y	0.7	(0.4, 1.1)	15	(12, 18)	3700	(3000, 4400)	248
+	G731S	0.5	(0.2, 1.0)	—		—
+	S615N	0.5	(0.3, 0.9)	—		—
+	V546M	0.5	(0.3, 0.8)	14	(10, 19)	2400	(1900, 2800)	171
+	G555S	0.4	(0.2, 0.8)	—		—
+	P614S	0.4	(0.2, 0.8)	11	(8, 15)	2100	(1700, 2400)	185
+	Q563R	0.3	(0.1, 0.5)	13	(9, 17)	2000	(1500, 2500)	158
+	I558L	0.3	(0.1, 0.5)	—		—

Rates of FOA or canavanine resistance (FOA^r or Can^r mutants per cell division) were determined by fluctuation analyses and maximum likelihood estimates using data from multiple independent experiments (except where noted). pol3-01,eex alleles are arranged from top to bottom according to the degree of mutator suppression as determined from rates of FOA resistance. Confidence intervals (95%) are in parentheses. A dash (–) indicates the mutation rate was not determined.

^aMutation rates were determined in two different strains: FOA^r, BP4001; Can^r, YP6.

^bMutation rates were determined in strain MP4.

^c pol3-01,eex alleles not analyzed: I616C, A677T, and P719T.

^dMutation rate from a single experiment.

^eThe fold-increase in mutation rate caused by loss of MSH6 (msh6Δ effect) was calculated by dividing each pol3-01,eex msh6Δ mutation rate by the corresponding pol3-01,eex MSH6 rate. On average, msh6Δ increased mutation rate 157-fold.

Contributions of MMR and Proofreading to DNA Replication Fidelity In Vivo

The eex mutants provided an opportunity to assess the proportion of Pol δ errors that are repaired by proofreading and MMR in vivo. Taking advantage of the viability of pol3-01,eex msh6Δ cells, we first compared the mutation rates of isogenic strains that lack Pol δ proofreading and differ only in their MMR activity. The average increase in mutation rate in pol3-01,eex strains after deletion of MSH6 was 157-fold (Table 1), consistent with Msh6-dependent repair of greater than 99% of the errors generated by proofreading-deficient Pol δ. As expected from the mutation biases conferred by pol3-01 or msh6Δ alone [40], [46]–[48], spontaneous mutations in pol3-01,eex msh6Δ strains were almost exclusively base substitutions (Figure S3 and Table S2). Thus, our estimate primarily reflects the efficiency of base-base mismatch repair. This estimate is an average across multiple scoreable sites in CAN1; MMR efficiencies at individual sites may vary widely [51].

To similarly estimate the efficiency of Pol δ proofreading in vivo, we initially determined the influence of eex alleles on mutation rates in the presence of proofreading. Most MMR-proficient pol3-eex strains had no discernable mutator phenotype (Figure 2D, Table 2). However, in the absence of MSH6 many of the pol3-eex alleles produced slightly higher mutation rates than the POL3 msh6Δ control (Table 2). These alleles were excluded from our analysis, because their weak mutator phenotypes may result from altered partitioning or other defects that reduce proofreading efficiency [4], [52]. Eight pol3-eex msh6Δ strains exhibited mutation rates within two-fold of POL3 msh6Δ: G204D, H620Y, T711A, E594G, Y808C, W821C, H879Y, and S968R. The mutation rates of these pol3-eex msh6Δ strains were compared to rates of the corresponding pol3-01,eex msh6Δ cells. This strategy allowed us to examine isogenic strains that differ only in their Pol δ proofreading activity and lack the masking effects of Msh6-mediated MMR. Proofreading deficiency increased mutation rates an average of 163-fold, indicating that the Pol δ exonuclease corrects greater than 99% of polymerase errors across the CAN1 reporter gene. Assuming Pol δ proofreading and Msh6-dependent MMR act in series [40], we estimate their combined contribution to DNA replication fidelity in yeast at greater than 10⁴. Proofreading and MMR contribute similarly to replication fidelity in bacteria [53].

Table 2. Mutation Rates (×10⁻⁷) of pol3-eex Mutant Strains.

Allele		MSH6 a				msh6Δ b		pol3-01 effect e
		FOAr		Canr		Canr
POL3 c		0.3	(0.1, 0.5)	4.9	(4.0, 5.8)	23	(20, 26)
+	G207R	0.4	(0.2, 0.7)	—		—
+	H879Y	0.4	(0.2, 0.6)	1.9	(1.4, 2.6)	56	(44, 67)	198*
+	Y808C	0.4	(0.2, 0.7)	2.1	(1.4, 2.9)	39	(30, 48)	198*
+	S968R	0.3	(0.1, 0.5)	2.2	(1.6, 2.8) d	23	(17, 30)	226*
+	A894V	0.6	(0.3, 0.9)	2.7	(2.0, 3.4)	110	(88, 130)	29
+	G400S	1.2	(0.8, 1.8)	—		—
+	G204D	0.7	(0.4, 1.1)	2.6	(1.9, 3.4)	54	(39, 70)	84*
+	E642K	0.5	(0.2, 1.0)	7.0	(5.6, 8.4)	150	(120, 180)	17
+	F486S	0.9	(0.6, 1.4)	—		—
+	T711A	0.3	(0.5, 0.6)	2.1	(1.7, 2.4)	22	(15, 39)	275*
+	R839H	0.5	(0.2, 0.9)	—		—
+	W821C	0.3	(0.1, 0.6)	3.0	(2.3, 3.8)	33	(27, 38)	78*
+	C365Y	0.3	(0.2, 0.5)	—		—
+	A786V	0.1	(0.0, 0.3)	6.1	(4.3, 7.9) d	80	(63, 98)	19
+	F793I	0.2	(0.1, 0.4)	2.8	(2.0, 3.6) d	89	(71, 110)	179
+	R475G	0.2	(0.1, 0.4)	—		—
+	N610D	0.5	(0.2, 0.9)	—		—
+	E594G	0.2	(0.1, 0.5)	8.3	(5.5, 11.2) d	15	(12, 18)	107*
+	R475I	0.4	(0.2, 0.8)	5.8	(3.6, 8.3) d	160	(130, 180)	18
+	D831G	0.9	(0.5, 1.4)	2.9	(2.1, 3.7)	64	(50, 78)	27
+	G921D	0.4	(0.1, 0.8)	—		—
+	V838A	0.5	(0.3, 0.9)	—		—
+	R923H	0.2	(0.0, 0.4)	—		—
+	E800K	0.7	(0.4, 1.2)	3.6	(2.7, 4.5)	120	(97, 160)	14
+	K891T	0.3	(0.1, 0.5)	6.6	(4.8, 8.3) d	220	(180, 260)	9
+	H620Y	0.3	(0.1, 0.6)	4.5	(2.8, 6.5) d	26	(19, 34)	141*
+	G731S	0.2	(0.1, 0.5)	—		—
+	S615N	0.3	(0.1, 0.7)	—		—
+	V546M	0.4	(0.2, 0.7)	4.3	(2.5, 6.5) d	71	(49, 93) d	33
+	G555S	0.6	(0.3, 1.0)	—		—
+	P614S	0.5	(0.3, 1.0)	4.4	(3.3, 5.6)	130	(100, 160)	16
+	Q563R	0.3	(0.1, 0.6)	4.3	(3.0, 5.7)	120	(98, 140)	17
+	I558L	0.5	(0.2, 0.8)	—		—

Rates of FOA or canavanine resistance (FOA^r or Can^r mutants per cell division) were determined by fluctuation analyses and maximum likelihood estimates using data from multiple independent experiments (except where noted). eex alleles are arranged from top to bottom in the same order as Table 1. Confidence intervals (95%) are in parentheses. A dash (–) indicates the mutation rate was not determined.

^aMutation rates were determined in two different strains: FOA^r, BP4001; Can^r, YP6.

^bMutation rates were determined in strain MP4.

^c pol3-eex alleles not analyzed: L411R, L531P, I616C, D643N, A677T, and P719T.

^dMutation rate from a single experiment.

^eThe fold-increase in mutation rate caused by loss of Pol δ proofreading (pol3-01 effect) was calculated by dividing each pol3-01,eex msh6Δ mutation rate (Table 1) by the corresponding pol3,eex msh6Δ rate (above). pol3,eex msh6Δ strains marked with an asterisk (*) have mutation rates within 2-fold of the POL3 msh6Δ strain, suggesting the eex alleles do not hamper primer partitioning or exonuclease function. In these strains, pol3-01 increased mutation rate an average of 163-fold.

Defining the Threshold of Error-Induced Extinction

The pol3-01,eex MMR-proficient strains formed colonies with similar size and efficiency as the POL3 control (Figure 5A, left). Thus, the corresponding eex mutant polymerases must suffice for the essential functions of Pol δ in replication [54]. However, in the absence of MSH6, pol3-01,eex alleles with the strongest mutator phenotypes impaired growth (Figure 5A, center). We used these mutants, together with synthetic-lethal alleles, to estimate the maximal mutation rate compatible with haploid yeast proliferation.

Figure 5. Defining the threshold of error-induced extinction.

A) Plasmid shuffling experiments to reveal synthetic interactions between pol3-01,eex alleles and MMR mutations. Ten-fold serial dilutions of yeast containing POL3–URA3 and pol3–LEU2 plasmids were plated on FOA-containing SC medium to select for cells that spontaneously lost POL3–URA3. Similar numbers of colony forming units were plated for each set of alleles in the MMR⁺, msh6Δ and msh2Δ strains. Failed growth indicates synthetic lethality. Small colonies reflect slow-growth phenotypes. Relative mutation rates are the Can^r mutation rates conferred by each pol3 allele relative to wild-type POL3 in MMR-proficient cells (see Table 1 and Table 2). Alleles are listed in decreasing order of mutator strengths. Some alleles with statistically similar mutation rates (as reflected by overlapping confidence intervals) have slightly different relative rates due to mathematical round-off. B) Relationship between growth capacity and CAN1 mutation rate for 62 haploid yeast strains. Colonies of pol3-01,E642K msh6Δ, pol3-01,G204D msh6Δ, pol3-01,H879Y msh6Δ, and pol3-01 msh6Δ cells are shown to illustrate wild-type (+++), moderately defective (++), severely defective (+), and failed (–) growth, respectively. The vertical dashed line indicates the estimated maximal mutation rate compatible with haploid yeast colony formation, which is our functional definition of the replication error threshold. Filled symbols, rates measured by fluctuation analyses. Open symbols, rates estimated as described in the text. Data in brackets with an asterisk (*) are pol3-01,T711A msh6Δ, pol3-01,S968R msh6Δ, pol3-01,G204D msh6Δ, and pol3-01,Y808C msh6Δ.

The upper and lower limits of the maximal rate were determined as follows. First, we calculated the predicted mutation rates of msh6Δ strains harboring synthetically lethal pol3 alleles (pol3-01, pol3-F406A, pol3-D407A or pol3-Y516F) as the mutation rate of each pol3 MSH6 strain (Figure 1A) times 157 (the average increase in rate observed upon deletion of MSH6; see preceding section). These predicted rates ranged from 2×10⁻³ Can^r mutants per cell division for pol3-01 msh6Δ and pol3-D407A msh6Δ to 4×10^-3 for pol3-F406A msh6Δ (Figure 5B). Second, we determined the growth capacities of all mutator and suppressor strains in our collection using a semi-quantitative scale based on colony size. Wild-type colony-forming capacity (+++) was consistently observed at rates as high as 5×10⁻⁵ Can^r mutants per cell division (Figure 5B). As the mutation rate exceeded 5×10^-5, several strains exhibited a slow-growth phenotype (++), and a single strain (pol3-01,H879Y msh6Δ) showed a severe growth deficit (+) at a mutation rate of 1×10⁻³. These results demonstrate that the maximal mutation rate is reached when there are ∼10⁻³ inactivating mutations in CAN1 per cell division (Figure 5B). Rates exceeding this maximum result in a failure to form visible colonies (i.e., error-induced extinction).

If the observed decline in viability is due to an error threshold, additional mutation stress should exacerbate the growth defect. We introduced pol3-01,eex alleles into msh2Δ cells, which are defective in both Msh6- and Msh3-mediated MMR and thus have Can^r mutation rates that are 2- to 3-fold higher than msh6Δ cells [3], [41], [47], [48]. Colonies were observed only in pol3-01 msh2Δ strains with the strongest mutator suppressor alleles (Figure 5A, right). Collectively, these data suggest that pol3-01 msh6Δ cells with weak mutator suppressors are on the edge of error-induced extinction and that eliminating MSH2 increases mutation rates beyond an extinction threshold. Although pol3-01 msh2Δ strains with strong mutator suppressors formed distinct colonies, these colonies were generally smaller and less uniform than the POL3 msh2Δ control. This variability in colony size suggests that viable pol3-01,eex msh2Δ cells quickly accumulate deleterious mutations that compromise replicative fitness. The observation that growth is impaired at mutation rates 10-times less than the 10⁻³ threshold (Figure 5B) suggests that accumulation of random mutations can impose a loss in fitness and shows that the growth capacity of haploid yeast declines even under conditions of non-lethal mutation burden.

Discussion

Mutators accelerate microbial adaptation and mammalian oncogenesis. However, the fitness cost of increased mutation imposes indirect selection pressure to reduce mutation rates. This counter-selection will occur after adaptation to a stable environment where conditions no longer favor the genetic potential of mutators. One possible mechanism to reduce mutation rates is the acquisition of compensatory alleles at modifier loci that suppress the mutator phenotype.

In this study, we took advantage of synergies between Pol δ proofreading and MMR to titrate yeast mutation rates up to lethal levels and study the fate of mutators under strong counter-selection. We found that msh6Δ cells carrying hypomorphic proofreading alleles abruptly lose viability over a narrow range of increasing mutation rates (Figure 1). Thus, cell survival requires both proofreading and MMR to limit potentially lethal mutations introduced by Pol δ. Mutant clones that escape this error-induced extinction arose spontaneously (Figure 2), frequently due to second-site changes in Pol δ (Figure 3 and Figure 4) that conferred antimutator phenotypes (Figure 2 and Table 1). Using our collection of mutator and antimutator strains, we found that the maximum mutation rate compatible with haploid yeast survival corresponds to ∼10^-3 inactivating mutations in CAN1 per cell division (Figure 5). These studies provide evidence for an error threshold in yeast and demonstrate that genetic suppressors of error-prone replication spontaneously arise in eukaryotic mutator cells.

Below we consider error thresholds in relation to genetic complexity and mutational robustness, and we discuss potential mechanisms of mutator suppression.

Error Thresholds, Genetic Complexity, and Mutational Robustness

We observed loss of growth capacity when the CAN1 mutation rate exceeds ∼10⁻³ inactivating mutations per cell division (Figure 5). The yeast genome is comprised of ∼6000 genes (http://www.yeastgenome.org). Thus, a mutation rate of 10⁻³ corresponds to the random inactivation of ∼6 genes per cell per replication cycle (assuming CAN1 is typical). On average, one of these six mutations will involve a gene required for haploid cell viability [55], [56]. Thus, there is a high probability that cells above the maximal mutation rate will acquire a lethal mutation after a few cell divisions. The restoration of cell growth via antimutator alleles (Figure 2) supports this hypothesis. Stalled DNA synthesis at nascent 3′ mispairs [57] and S-phase checkpoint signalling [58] could also contribute to growth arrest in strong mutators. However, it is not evident how MMR defects would exacerbate 3′ mispair extension by Pol δ, and simultaneous loss of proofreading and MMR does not halt growth specifically in S-phase [40]. Rather, proofreading/MMR double mutants arrest with varied cell morphologies [40] that resemble those observed in a systematic promoter-repression screen of essential genes [59] (Figure S4). Considered together, the evidence suggests that random mutations in essential genes are a primary cause of error-induced extinction. Synthetic cooperative interactions of non-lethal alleles will also contribute as cells accrue multiple mutations [60], [61]. In a similar manner, bacteria exhibit a replication error threshold that correlates with the number of indispensable genes [62], suggesting that maximal mutation rates can be used to estimate the genetic complexity of vital pathways in other organisms.

Error thresholds are also evident in diploids. Although diploid genomes generally buffer cells against the deleterious effects of mutation accumulation [63], haploinsufficient alleles still pose a significant threat to fitness. In a comprehensive library of diploid yeast heterozygotes, up to 20% of the hemizygous mutant strains exhibit reduced fitness during growth competition [64]. Observations of mutation meltdown in MMR-deficient cells [39] and lethality conferred by a hyper-mutator Pol δ variant [65] argue that diploid yeast are subject to an error threshold. The combined loss of Pol δ proofreading and MMR is also synthetically lethal in mice [25]. Similar to the situation in yeast [40], mouse cells defective for both proofreading and MMR are initially viable but arrest after a limited number of mitotic divisions [25]. This cessation of growth presumably results from an accumulation of mutations in genes required for cell propagation and embryo development.

Although cells eventually succumb to error-induced extinction, they tolerate substantial increases in mutation rate before losing viability. This mutational robustness is apparent in yeast, E. coli and mouse cells (Figure 6). We show that haploid yeast tolerate more than a 1,000-fold increase in mutation rate before exhibiting overt loss of colony-forming capacity (Figure 5B), and a comparable increase in mutation rate is required to cause catastrophic errors in E. coli [62], suggesting that prokaryotes and haploid eukaryotes share similar degrees of robustness toward DNA polymerase errors. In comparison, diploid yeast and mouse cells retain replication capacity at mutation rates 10,000-times higher than wild-type levels (Figure 6; [25], [40], [66]). Thus, diploidy extends the threshold of error-induced cell death by five- to tenfold. These data suggest that cells can survive and persist during periods of high mutational loads. The maximal mutation rate will likely vary, depending on environmental conditions [67], [68], genetic redundancy [63], [69], [70], the plasticity of genetic interactions [71], [72], and the ability of cells to buffer deleterious changes in essential proteins [73].

Figure 6. Mutational robustness of yeast, E. coli, and mice.

Comparison of spontaneous per-base-pair mutation rates of wild-type (WT) and strong mutator strains of yeast (haploid and diploid), E. coli and mouse cells. Gray boxes indicate the mutation rate intervals that coincide with the transition from wild-type growth (leftmost boundary) to failed growth (rightmost boundary). The data for haploid yeast mutators are from Figure 5B, with the left boundary corresponding to the mutation rate of pol3-01,E642K msh6Δ cells (6.5×10⁻⁷) and the right boundary corresponding to the lethal threshold (4.1×10⁻⁶). The mutation rates of pol3-01,G204D msh6Δ, pol3-01,H879Y msh6Δ, and pol3-01 msh6Δ haploid yeast are shown as examples of progressively stronger mutators with slow (++), very slow (+) and no-growth (–) phenotypes, respectively. The data for pol3-01/pol3-01 pms1/pms1 diploid yeast are from Morrison et al. [40]; pol3-01/pol3-01 pms1/pms1 cells divide very slowly with a growth phenotype presumably in the range of + to ++. The mouse Pold1^e/e Mlh1^Δ/Δ mutation rate is extrapolated from ouabain-resistance rates of cultured Pold1^+/e Mlh1^Δ/Δ fibroblasts as described in Materials and Methods; a growth phenotype between + and ++ is assumed [25]. E. coli mutation rates and growth phenotypes are from Fijalkowska et al. [62]; mutD5(pGW1842), mutD5 and dnaQ926 exhibit slow (++), very slow (+) and no-growth (–) phenotypes, respectively. The positions of the gray boxes for diploid yeast, mouse and E. coli are estimates based on the mutation rate and growth capacity relationships observed in haploid yeast (Figure 5B). The error thresholds (rightmost boundaries) for diploid yeast and mouse cells are not known. The yeast wild-type rate is the average of multiple independent determinations (data herein and [40], [108], [114], [124]). Wild-type mouse and E. coli mutation rates are from Drake et al. [9], [114].

Structural Implications of the eex Amino-Acid Substitutions

We observed that escape mutants readily emerge when moderate mutators are pushed above the error threshold (Figure 2). One-third of the escape mutants resulted from second-site changes in Pol δ that suppress the proofreading-deficient mutator phenotype. Recent structural studies of S. cerevisiae Pol δ [74] provide insight into potential mechanisms of mutator suppression by these intragenic eex alleles (Figure 4, Figures S5 and S6).

Many eex mutations alter amino acids around the polymerase active site that are predicted to influence dNTP binding or catalysis (Figure 4B). Effects may be mediated by direct interactions of mutated residues with the metal•dNTP substrate or via packing interactions that indirectly affect the substrate binding pocket. Other eex mutations map to a stretch of amino acids that bind the template near the active site and buttress the fingers domain, which contains residues that contour the template•dNTP base pair (Figure S5). Amino-acid changes affecting active-site geometry, positioning of the template nucleotide, or stability of the catalytic conformation may act as antimutators by increasing selectivity for correct dNTPs or by slowing the rate of catalysis so that mispaired template•primers have more time to dissociate from Pol δ. A model of dissociation and subsequent editing by an alternative enzyme [20], [75] may best explain eex mutations that change amino acids along the DNA binding track (Figure 4C). Similarly, eex mutations in the exonuclease domain may impart structural changes that promote Pol δ dissociation during failed proofreading attempts (Figure S6). Intriguingly, two eex amino-acid substitutions (E642K and D643N) are located on the solvent-exposed surface of Pol δ (asterisk in Figure 4A), suggesting that changes in protein-protein interactions influence mutagenesis. Proteins encoded by eex loci extragenic to POL3 (Table S1) are candidate interacting partners.

Several alternative enzymes may function to edit Pol δ errors in the eex mutants. One candidate is proofreading by Pol ε. Yeast with deficiencies in both Pol δ and Pol ε proofreading exhibit a synergistic increase in mutation rate, suggesting one or both polymerases may proofread for the other [46]. Other candidates include the 3′→5′ exonuclease activities of MRE11 [76] and Apn2 [77], or endonucleases such as Rad1/Rad10 or Mus81/Mms4 that cleave 3′ flap structures during replication fork restart [78]–[82]. An important consideration is that such alternative editing pathways may be redundant, with multiple activities masking the contributions of any one nuclease.

The locations of several eex substitutions in Pol δ resemble those of antimutators previously identified in bacteriophage T4 polymerase [4], [83]–[86] and in herpes simplex virus polymerase [87]–[89], two B-family DNA polymerases similar to Pol δ (Figure 3). Genetic screens have also identified E. coli DNA polymerase I and III antimutator variants, and similar to our findings, these E. coli antimutators result from diverse amino-acid substitutions throughout the polymerase structures [19], [20], [90]–[92]. Some amino-acid substitutions in T4 pol are thought to increase polymerase fidelity by promoting ‘hyper-editing’ of the primer terminus by the integral proofreading exonuclease (reviewed in [4]). However, the eex mutations we describe mediate their antimutator effects without the aid of an active exonuclease domain, similar to previously isolated E. coli antimutators [19], [20], [90].

Taken together, this structural analysis suggests two general antimutator mechanisms for Pol δ eex mutations: 1) increased dNTP discrimination, thereby making Pol δ more accurate, and 2) increased dissociation from mispaired primer-templates, thereby allowing other enzymes to proofread Pol δ errors. eex mutations could also decrease errors at Okazaki fragment junctions by suppressing the strand-displacement activity of proofreading-deficient Pol δ [52], [93]–[95].

Evidence for Other Pathways of Mutator Suppression

Our study took advantage of synthetically lethal interactions between Pol δ proofreading and MMR alleles to select for antimutators. Several lines of evidence indicate that mutator suppressors also arise under non-lethal conditions and are not restricted to the Pol δ proofreading – MMR pathway. Morrison and Sugino observed mutator suppression in a yeast clone defective for Pol ε proofreading and MMR [46], and an engineered second-site mutation in Pol ε suppresses the mutator effect of Pol ε proofreading deficiency [96]. In E. coli, suppressors of diverse mutator pathways (MMR, proofreading and DNA damage repair) emerge spontaneously in strains that are well below the error threshold [14], [18]–[20]. In our studies, large-colony variants of slow-growing mutators were frequently observed (see, for example, Figure 5A), and in the one variant we pursued, we found the A894G suppressor mutation. Collectively, these studies show that many defects in DNA replication fidelity can be genetically suppressed and suggest that both moderate and strong mutators are intrinsically unstable.

The facile emergence of mutator suppressors that we observed in yeast suggests that similar pathways of suppression exist in multicellular eukaryotes. This has implications for the role of mutator phenotypes in cancer [22], [97]. During neoplastic transformation, mutator alleles that promote the formation of tumor cells are likely to incur a fitness cost due to an increase in mutational load. To offset this cost, suppressor alleles that reduce the mutation rate may emerge during the later stages of oncogenesis after genetic barriers to immortalization and metastasis have been overcome. Although recent findings suggest that a mutator phenotype persists in at least some types of human tumors [23], our results raise the prospect that mutator phenotypes may be transient during tumor progression due to genetic suppression. An analysis of mutation rate dynamics in cancer is warranted.

Materials and Methods

Media and Growth Conditions

Yeast were grown at 30°C using YPD, synthetic complete (SC) media or SC drop-out media deficient in specific amino acids as needed to select for prototrophy [98]. Pre-formulated nutrient supplements for SC and SC lacking uracil and leucine were purchased from Bufferad. All other drop-out supplements were made as described [98]. URA3-deficient cells were selected on SC medium containing 1 mg/ml 5-fluroorotic-acid (FOA; Zymo Research) and an additional 50 mg/L uracil [43]. TRP1-deficient strains were selected on FAA selection media containing 0.5 mg/ml 5-fluroanthranillic acid (FAA) made as described [99]. Canavanine-resistant mutants were scored on SC plates lacking arginine that contained 60 µg/ml of canavanine. Reagents were obtained from Sigma-Aldrich or Fisher Scientific unless otherwise indicated.

Yeast Plasmids and Strains

Plasmids

pGL310 is the CEN4/ARS1/URA3 plasmid, YCp50 [100], modified to carry SUP11 and the wild-type POL3 gene under control of its native promoter [29], [101]. YCplac111POL3 and YCplac111pol3 derivatives are CEN4/ARS1/LEU2 plasmids derived from YCplac111 [102] and contain the entire wild-type POL3 or mutant pol3 genes (with native promoters) flanked by HindIII and EcoRI restriction sites. pRS414POL3 and pRS414pol3-01 are derivatives of the CEN6/ARS4/TRP1 plasmid, pRS414 [103], carrying the HindIII-EcoRI POL3 DNA fragments from YCplac111POL3 and YCplac111pol3-01, respectively. The construction of YCplac111 and pRS414 vectors and subsequent subcloning of eex mutants are described in detail in Text S1.

Strains

Yeast strains and their genotypes are listed in Table S3. Chromosomal gene disruptions were introduced using PCR products generated with primers, templates, and protocols detailed in Table S4 and Text S1. YGL27-3D (a kind gift from Michel Simon and Gerard Faye, Institut Curie) is a haploid strain that carries a lethal, partial deletion of chromosomal POL3 substituted by HIS3 and complemented by pGL310. Chromosomal MSH6 was replaced with TRP1 to create YGL27-3Dmsh6dis4. To limit gene conversion between the pol3 plasmids and residual POL3 sequences in the chromosome, the entire POL3 coding sequences in YGL27-3D and YGL27-3Dmsh6dis4 were replaced with kanMX [104], creating YP6 (previously called YGL27-pol3Δ [105]) and MP4, respectively.

The S288c derivative, BY4733 [103], was modified to create a set of isogenic strains in a standard genetic background. BY4733 was transformed with pGL310, and chromosomal POL3 was deleted and replaced with HIS3 to create P3H3a. MSH2 was deleted in P3H3a and replaced with TRP1 to create BP0109. MSH6 was deleted in P3H3a and replaced with kanMX [106] to create BP1506. P3H3a was modified as follows to allow mutation rate measurements at URA3. First, P3H3a was transformed with pRS414POL3 and plated on SC FOA to isolate a strain that lost pGL310. Then, to create BP4001, AGP1 on Chromosome III was replaced with URA3 oriented with the direction of transcription towards ARS306 [107].

Plasmid Shuffling

Plasmid shuffling with pGL310-containing strains was carried out essentially as described [29], [43] (Figure S1). Cells transformed with YCplac111pol3 plasmids, YCplac111POL3 (positive control), or YCplac111 (negative control) were plated on SC lacking uracil and leucine. Cells transformed with pRS414pol3-01, pRS414POL3 (positive control), or pRS414 (negative control) were plated on SC lacking uracil and tryptophan. After three days at 30°C, individual colonies were picked and resuspended in sterile H₂O, and serial dilutions containing approximately 10⁵, 10⁴, 10³, or 10² cells were plated onto SC or SC FOA to select for clones that spontaneously lost the URA3 plasmid pGL310. A similar approach was used for shuffling in strains carrying the TRP1 plasmid pRS414POL3; BP4001 transformants containing both pRS414POL3 and YCplac111-based plasmids were selected on SC lacking tryptophan and leucine and then plated onto SC FAA to select for clones that spontaneously lost the TRP1 plasmid.

eex Mutant Screen

For the systematic isolation of spontaneous pol3-01,eex mutant alleles (Figure 2C), a pol3-01 plasmid was transformed into pol3Δ msh6Δ + pGL310 yeast, and 1–5×10⁶ cells from independent transformants were plated separately on SC FOA. FOA-resistant clones were isolated in the MP4 strain carrying YCplac111pol3-01 or in BP1506 carrying YCplac111pol3-01 or pRS414pol3-01. Bona fide eex mutants were distinguished from FOA-resistant clones that result from ura3 mutation or pol3-01→POL3 gene conversion by using a genotyping assay described in Text S1. pol3-01 plasmids from individual eex mutants were recovered and reshuffled into naïve pol3Δ msh6Δ + pGL310 cells to identify suppressors intragenic to pol3-01. Plasmids that conferred consistent survival upon reshuffling were purified, and the pol3 genes were sequenced (primer sequences available on request). Intragenic eex alleles thus identified were individually re-engineered into fresh YCplac111POL3 and YCplac111pol3-01 vectors and re-transformed into MP4 or BP1506 stock strains as a final confirmation of the ability of each allele to confer the eex phenotype. The re-engineered mutants were used to assess the effects of eex alleles on mutation rates and plating efficiencies.

Mutation Frequencies and Rates

For the scanning mutagenesis screen (Figure S2), sequence-verified YCplac111pol3 plasmids were shuffled into YP6 or MP4 cells immediately prior to each experiment. Twelve to thirteen independent FOA-resistant colonies of each genotype were streaked onto SC plates in ∼1-cm patches, grown two days at 30°C, and then replica-plated to canavanine plates to qualitatively assess mutant frequencies based on the number of canavanine-resistant colonies [47] (Figure S2A).

To measure mutation rates at the CAN1 locus, freshly streaked YP6 or MP4 strains were transformed with YCplac111POL3 or YCplac111pol3 plasmids, and multiple independent transformants were shuffled on SC FOA plates to obtain well-isolated single colonies. For each genotype, seven to eleven independent colonies, 1–2 mm in diameter, were excised as an agar plug, resuspended in 1 ml of dH₂O, and sonicated briefly. To estimate the number of cell divisions (Nt) during colony formation, serial dilutions were plated on SC media, and the number of colony-forming units was counted after two days at 30°C. To determine the number of mutants for wild-type and weak mutator strains, all of the remaining cells were plated on canavanine plates; for stronger mutators, the cell suspension was diluted 1∶10 to 1∶200 in dH₂0 prior to plating. The numbers of canavanine-resistant colonies on each plate were scored after three to four days at 30°C.

To measure URA3 mutation rates, BP4001 was transformed with YCplac111POL3, YCplac111pol3-01, or their respective eex mutant derivatives. Four FAA-resistant colonies from independent transformants with each plasmid were inoculated into separate 100-µl SC overnight cultures. The following morning the cultures were diluted to 1000 cells/ml and, for each of the four isolates, 12 parallel 100-µl cultures (100 cells/culture) were set up in 96-well microtiter plates. The plates were sealed with adhesive PCR plate sealers (Abgene, AB-0558) to minimize evaporation [108]. After two days of growth at 30°C, the cells were re-suspended by vigorous vortexing, and nine of the replicate cultures were spot-plated in 200-µl volumes on SC FOA plates. To estimate the total number of cell divisions, the remaining three replica cultures from each isolate were combined, diluted, and plated on SC plates. Colony numbers were scored after 3–4 days. We confirmed that spot plating accurately determines the number of FOA-resistant colonies for the strongest mutator by dividing test cell suspensions in half and comparing colony counts in a 100-µl spot with 100 µl of the same suspension spread over an entire SC FOA plate.

Mutation rates were determined from the number of mutant colonies in each replica by calculating an estimate for m by maximum likelihood [109] using newtonLD in Salvador 2.1 [110] with Mathematica 6.0 (Wolfram Research) and dividing by the number of cell divisions inferred from colony forming units. Where values for Nt from independent experiments differed by less than 2-fold, the data sets were combined for the mutation rate calculations [109]. In some instances, Nt values from independent experiments differed by more than 2-fold. In most cases, the independently-derived mutation rates were similar and a single value was reported (noted in Table 1). Confidence intervals were calculated in Salvador 2.1 using LRIntervalLD, which relies on likelihood ratios [110].

From these mutation rates, the efficiency of Msh6-dependent MMR (e_m6), expressed as the percentage of errors corrected, was calculated using equation 1:

(1)

where Mr is the relative mutation rate of the strain indicated in the subscript. The efficiency of Pol δ proofreading (e_δexo), expressed as a percentage of errors corrected, was calculated similarly from equation 2:

(2)

CAN1 Mutation Spectra

For each strain, we isolated up to 48 canavanine-resistant mutants from 48 independent shuffling experiments. Cells were treated with Zymolyase (ICN Biomedicals; 50 u/ml in 10 mM Tris•HCl/0.1 mM EDTA, pH7.5 at 37°C for 30 min then 95°C for 10 min), and the can1 coding sequence was PCR-amplified in 50-µl reactions with Phusion polymerase (NEB) using primers can1F1N (5′-GGTTAAGATAAGTAGATAAGAGAATGATACG-3′) and can1S1 (5′-GCGTGGAAATGTGATCAAAGG-3′) with the following PCR conditions: 98°C, 1 min.; 35× (98°C, 10 sec.; 45°C, 30 sec.; 72°C, 90 sec.); 72°C, 1 min. The samples were treated with 5 units each of Antarctic phosphatase and Exo1 (New England Biolabs) to degrade excess primer and dNTPs, heated at 80°C for 20 min to inactivate the enzymes, and then sequenced with primers can1S1, can1S2 (5′-CCAAAGCGCCAAATGCAGCAG-3′), can1S3 (5′-TCCAATAACGGAATCCAACTG-3′) and can1S4 (5′-GGGCAATCATACCAATATGTC-3′). Mutation spectra were tabulated and compared using iMARS [111].

Per-Base-Pair Mutation Rates

Phenotypic mutation rates were converted to per-base-pair rates using the approach of Drake [112]–[114] according to equations 3 – 5:

(3)

(4)

(5)

C and C′ ´ (equations 3 and 4) are correction factors to adjust for undetected (phenotypically silent) base-substitution mutations in a reporter gene. B_CT  =  the number of chain-terminating base substitutions (3 possible codons), B  =  the number of all base substitutions (64 possible codons), and I  =  the number of insertions+deletions (indels) in representative mutation spectra from M mutants sequenced. The mutation rate per base pair (μ_b) is calculated using equation 5 from the experimentally determined phenotypic mutation rate (μ_T) multiplied by the correction factor C′ ´ and divided by the number of base pairs in the mutation-reporter target sequence (T). The effective target size (τ) is estimated by T / C ´′.

In our collection of 484 Can^r mutants from proofreading- and MMR-deficient yeast (Figure S3 and Table S2), there were 442 base substitutions (101 chain-terminating + 341 missense) and 42 indels (including infrequent complex mutations) in CAN1 (T = 1773). Thus, C = 4.87 and τ = 391 base pairs. These values from mutator yeast strains are similar to those previously determined by others scoring spontaneous mutation in wild-type yeast (C = 4.73, [113]; τ = 236, [108]). The per-base-pair rates for haploid yeast plotted in Figure 6 were calculated from our Can^r μ_T values (Table 1 and Table 2) with C = 4.87, C′ ´ = 4.53 and T = 1773. For diploid pol3-01/pol3-01 pms1/pms1 yeast, we used the FOA^r mutation rate (μ_T) of 3.5×10^-4 reported by Morrison et. al. [40]; C = 8.18 (determined from the data of Lang and Murray [108]), C′ ´ = 6.79 and T = 804 base pairs for the URA3 target gene. Thus, τ is 118 base pairs, and the per-base-pair mutation rate of pol3-01/pol3-01 pms1/pms1 diploids at the URA3 locus is [(3.5×10⁻⁴) ×6.79 / 804]  = 3.0×10⁻⁶. In Figure 6 we multiply this rate by 1.8 to adjust for the lower intrinsic mutation rate of URA3 compared to CAN1 (Table 1 and Table 2 and [108]).

For mouse cells, per-base-pair mutation rates were calculated from ouabain-resistance (Oua^r) rates determined in our laboratory using spontaneously immortalized mouse embryo fibroblasts ([25] and unpublished data). The effective target size (τ) is estimated as follows. Base substitution mutations in any one of sixteen codons in the Na,K-ATPase α1 gene (Atp1a1) are known to confer genetically dominant resistance to µM concentrations of ouabain in human cells [115]. Mouse cells, however, are naturally resistant to µM concentrations of ouabain due to differences at 2 of these 16 codons (Q111R and N122D; [116], [117]. Our fluctuation assays were conducted with 2 mM ouabain [25], conditions expected to only detect mutations that confer exceptionally high ouabain resistance. We estimate the target size to be ∼5 base pairs per allele, corresponding to two Atp1a1 codons (D121 and T797) known to effect >50-fold ouabain-resistance when mutated [115], [118]. Mouse fibroblast cell lines are typically tetraploid [119]. Therefore τ = 5 base pairs per allele×4 alleles = 20 base pairs. Pold1^+/e Mlh1^Δ/Δ cells, which are heterozygous defective for Pol δ proofreading and nullizygous for MMR, exhibited a mutation rate of 65×10⁻⁷ Oua^r mutants per cell division (95% confidence interval = 56–75×10⁻⁷). This phenotypic rate corresponds to a per-base-pair rate of 65×10⁻⁷ / 20 base pairs  = 3.3×10⁻⁷. Mouse cells that are homozygous deficient for both Pol δ proofreading and MMR (Pold1^e/e Mlh1^Δ/Δ) are viable but divide slowly up to embryonic day E9.5 [25]. Based on the relative mutation rates of MMR^Δ/Δ diploid yeast with +/- or −/− Pol δ proofreading alleles [40], we estimate the per-base-pair rate of Pold1^e/e Mlh1^Δ/Δ mouse cells to be 5×10⁻⁶.

Supporting Information

Figure S1

Plasmid shuffling strategy. Mutated pol3 alleles were introduced into MSH6 and msh6Δ yeast by plasmid shuffling. Haploid yeast with a chromosomal deletion of POL3 (pol3Δ) complemented by a wild-type POL3–URA3 plasmid (left) are transformed with mutant pol3–LEU2 plasmids. Individual colonies carrying both the pol3 and POL3 plasmids are isolated (center), and dispersed cells are then plated on 5-fluoroorotic acid (FOA) media to select mutant pol3–LEU2 clones that lost the wild-type POL3–URA3 vector (right).

(PDF)

Figure S2

Mutagenesis screen of the Pol δ exonuclease domain. Each allele was engineered into a wild-type POL3 vector (YCplac111POL3) by site-directed mutagenesis and then introduced into yeast by plasmid shuffling (see Figure S1). Independent FOA-resistant colonies were patched onto synthetic complete (SC) plates and then replica-plated to SC plates lacking arginine and containing canavanine (60 µg/ml) to assess mutator phenotypes [47]. A) Representative canavanine plates used to assess mutator phenotypes. Each plate has twelve or thirteen patches of cells derived from independent colonies of the indicated genotypes. Spontaneous canavanine-resistant (Can^r) mutants appear as small colonies in the ∼1-cm patches. Mutant frequencies were qualitatively judged from the lowest (wild type, WT) to highest (pol3-G400A msh6Δ) as indicated by –, +, ++ and +++ scores. B) Summary of POL3 alleles and corresponding mutator and growth phenotypes in MSH6 and msh6Δ cells.

(PDF)

Figure S3

Spontaneous CAN1 mutations from pol3-01,eex msh6Δ cells. The can1 coding sequences from 30–48 independent canavanine-resistant (Can^r) mutants of each strain were PCR-amplified and sequenced. Spontaneous mutations identified in different strains are color coded according to the key at the bottom. Each base letter above the wild-type CAN1 sequence indicates the site and nature of an independent base substitution or frameshift (+ or -) mutation. CAN1 sequences involved in complex mutations are indicated by horizontal colored lines. Multiple mutations identified in can1 from the same mutant clone are designated by the same superscript in the same color code. We observed mutation hotspots in CAN1 that arose in multiple independent Can^r clones. One hotspot, a C to T mutation at nt 899, occurred in eight independent pol3-01,K891T Can^r clones. Two of the eight mutants had a second mutation elsewhere in the CAN1 sequence, unambiguously identifying each clone as unique. This suggests that the abundance of mutations at this site is unlikely to be an artifact. One Can^r POL3 msh6Δ mutant (not shown) contained an insertion/deletion mutation that was evident by a larger PCR product; sequencing with nested CAN1-specific primers revealed an insertion containing the gene RRP45.

(PDF)

Figure S4

Similar cell morphologies in yeast after error extinction or repression of essential genes. Black bars: terminal cell morphologies of pol3-01 pms1Δ haploid cells that ceased growing due to error extinction [40]. White bars: cell-cycle arrest phenotypes of 563 haploid strains, each with a different repressed essential gene [59]. In the repression study, 82 additional essential genes showed a growth defect but with no defined cell-cycle arrest phenotype.

(PDF)

Figure S5

eex amino-acid substitutions near the template nucleotide in Pol δ. Schematic of the α-carbon backbone of Pol δ with residues of interest depicted as space-filling spheres. Structural elements are color-coded as in Figure 4 with the template•dNTP (T₀P₀) and polymerase active-site residues shown as CPK sticks. Amino acids changed by eex mutations are shown as light blue spheres and labeled to indicate the eex substitutions. Residues V546, G555, I558, and Q563 from the amino domain are in three closely associated α-helices that bind the template and buttress the fingers domain. The exo domain has been removed for clarity, and the penultimate T₁P₁ base-pair (brown and gold spheres) is included to delineate the binding pocket. Panel (A) is a view looking down on the DNA major groove. Panel (B) is the same image rotated 90° around the x-axis. The T₁P₁ base-pair was removed in Panel (B) to reveal positions of amino-acid substitutions around the template•dNTP. Structure from [74] (Protein Data Bank accession code 3IAY).

(PDF)

Figure S6

eex amino-acid substitutions in the exonuclease domain of Pol δ. The exo domain (red) is shown as a schematic of the α-carbon backbone, and exonuclease active-site residues are gray CPK sticks. Amino acids changed by eex mutations are shown as light blue spheres and labeled to indicate the eex substitutions. The red dotted line corresponds to a missing loop in the structure (amino acids 491–496). The β-hairpin in T4 and RB69 pols affects partitioning of the primer between polymerase and exonuclease active sites [4]. Structure from [74] (Protein Data Bank accession code 3IAY).

(PDF)

Table S1

Genotypes of candidate eex mutants.

(PDF)

Table S2

Types of spontaneous CAN1 mutations in pol3-01,eex msh6Δ strains.

(PDF)

Table S3

Yeast strains.

(PDF)

Table S4

Construction of chromosomal gene disruptions.

(PDF)

Text S1

Supplementary methods.

(PDF)