Quantifying the contributions of base selectivity, proofreading and mismatch repair to nuclear DNA replication in Saccharomyces cerevisiae

Abstract

Mismatches generated during eukaryotic nuclear DNA replication are removed by two evolutionarily conserved error correction mechanisms acting in series, proofreading and mismatch repair (MMR). Defects in both processes are associated with increased susceptibility to cancer. To better understand these processes, we have quantified base selectivity, proofreading and MMR during nuclear DNA replication in Saccharomyces cerevisiae. In the absence of proofreading and MMR, the primary leading and lagging strand replicases, polymerase ε and polymerase δ respectively, synthesize DNA in vivo with somewhat different error rates and specificity, and with apparent base selectivity that is more than 100 times higher than measured in vitro. Moreover, leading and lagging strand replication fidelity rely on a different balance between proofreading and MMR. On average, proofreading contributes more to replication fidelity than does MMR, but their relative contributions vary from nearly all proofreading of some mismatches to mostly MMR of other mismatches. Thus accurate replication of the two DNA strands results from a non-uniform and variable balance between error prevention, proofreading and MMR.

1. Introduction

In an unperturbed eukaryotic cell cycle, high nuclear DNA replication fidelity is achieved through the sequential operation of three processes: the selectivity of three replicative DNA polymerases (replicases) for inserting correct nucleotides into properly aligned DNA substrates, exonucleolytic proofreading of mismatches made during replication, and DNA mismatch repair (MMR) of rare errors that escape proofreading. In Escherichia coli, where DNA polymerase III is the primary replicase for both DNA strands, the relative contribution of these processes to replication fidelity has been determined by comparing spontaneous mutation rates in a wild type strain to rates in strains deficient in either proofreading by Pol III, in MMR, or in both error correction mechanisms [1]. A similar approach has been applied to replication of the Saccharomyces cerevisiae nuclear genome [2,3], but in a less comprehensive manner and at a time when the identities of the major leading and lagging strand replicases were uncertain. Now however, compelling evidence using base substitution patterns (see [4] and references therein) and more recently using strand-specific ribonucleotide incorporation [5–8] as biomarkers of replicase actions indicates that the leading strand is primarily replicated by DNA polymerase ε (Pol ε), the product of the yeast POL2 gene. Synthesis of the nascent lagging strand involves limited synthesis by Pol α (POL1), which is naturally deficient in 3′-exonucleolytic activity and cannot proofread mismatches, which it generates at a rate of about 10⁻⁴ in vitro [9]. Synthesis by Pol α is followed by extensive synthesis by Pol δ (the product of the yeast POL3 gene). When this knowledge of strand specific replicase activity is combined with use of a mutational reporter gene placed close to a well-studied replication origin, it is now possible to deduce the identity of the mismatches that are being generated, proofread or corrected by MMR during nuclear DNA replication in a yeast cell.

Unlike Pol α, the catalytic subunits of Pol ε [10] and Pol δ [11] have 3′-exonuclease activity for proofreading their own errors. Moreover, there is evidence to suggest that the exonuclease activity of Pol δ, but not that of Pol ε, likely proofreads errors made by Pol α during lagging strand replication [12], and more recent evidence that Pol δ can proofread errors made by Pol ε [13]. Not only is proofreading more complicated in yeast as compared to E. coli, but so too is eukaryotic MMR more complicated than in E. coli. This is because two complexes of eukaryotic MutS Homologs are present for MMR. Single base–base mismatches that can result in base substitutions are primarily repaired by Msh2–Msh6 (MutSα), with Msh2–Msh3 (MutSβ) having but a much smaller role [14,15]. However, MutSα and MutSβ can both participate in repairing insertion-deletion (indel) mutations containing one or two unpaired bases, and MutSβ has primary responsibility for repairing mismatches containing multiple unpaired bases [16].

In this study, we have examined the contribution of all three replication fidelity processes to genome stability in budding yeast. The complexity described above necessitates the use of several yeast strains to quantify the contributions of nucleotide selectivity, proofreading and MMR to leading and lagging strand DNA replication fidelity. Here we use the URA3 reporter gene that scores all types of substitutions in many different sequence contexts. We compare mutation rates in a wild type strain to rates in strains defective in proofreading by Pol δ (pol3-5DV, subsequently referred to as pol3-exo^-) [17], or defective in proofreading by Pol ε (pol2-04, subsequently referred to as pol2-exo^-) [10], in both cases resulting from mutations in their exonuclease active sites. Rates in MMR-proficient strains are compared to rates in msh6Δ strains that cannot repair the vast majority of single base-base mismatches due to inactivation of MutSα. However, in these strains, most indel mismatches are repaired by MutSβ, thus reducing the possibility of error catastrophe due to lethal indels in strains lacking both proofreading and MMR. Here we have determined mutation rates and mutational spectra in all strains, and used these data to calculate rates for each type of base substitution. Pairwise comparisons of rates among strains then allowed us to estimate the contributions of the three fidelity processes to different single base-base mismatches in different sequence contexts, and to do so for leading and lagging strand replication. The results indicate that on average, proofreading contributes more to replication fidelity than MMR, but the relative importance of proofreading and MMR for correcting replication errors varies widely. In the absence of both MMR and proofreading, Pols δ and ε generate errors in vivo at much lower rates than have been measured in vitro, suggesting that additional fidelity mechanisms may operate at the replication fork in vivo.

2. Methods

2.1. Strain construction

The identities and sources of the strains used in this study are listed in Supplementary Table 1. For this study, we used pol2-04 (Pol ε) and pol3-5DV (Pol δ) mutants that are subsequently referred to as pol2-exo^- and pol3-exo^-, respectively. In the pol2-04 mutant, alanines were substituted for D290 and E292 in the 3′-exonuclease active site, inactivating 3′-exonuclease activity but leaving polymerase activity similar to that of wild type Pol ε [10]. The homologous allele for Pol δ is pol3-01 (D321A, E323A [18]), but this allele is lethal in combination with msh6Δ [19]. We therefore used the pol3-5DV mutant, in which a valine was substituted for D520 in the exonuclease active site to inactivate 3′-exonuclease activity but leave polymerase activity similar to that of wild type Pol δ [20].

2.2. Mutation rate measurements and mutational spectra

Spontaneous mutation rates at the URA3 locus were measured by fluctuation analysis as described [21,22]. URA3 mutation spectra were obtained by sequencing the URA3 gene in collections of independent 5-fluoroorotic acid-resistant (5-FOA^R) colonies. Each 5-FOA^R colony from the double mutant strains was obtained from an independent spore. Genomic DNA was isolated from 5-FOA^R colonies, the URA3 gene was PCR-amplified and the DNA product was sequenced. Rates for each type of mutation are calculated as the total number of each type of mutation divided by the total number of 5-FOA^R mutants sequenced and then multiplied by the total mutation rate. For individual types of base substitutions, the substitution rate per base pair per generation was calculated by dividing the mutation rate by the number of sites in the URA3 gene where that event is known to result in 5-FOA^R (Supplementary Fig. 1). The contributions of base selectivity, proofreading and mismatch repair to replication fidelity were calculated as described below.

3. Results and discussion

3.1. Approach

We measured mutation rates in yeast strains that were wild type for proofreading and MMR, deficient in proofreading (pol2-exo^- or pol3-exo^-) only, deficient in MMR of base-base mismatches (msh6Δ), or deficient in proofreading and MMR (pol2-exo^- msh6Δ or pol3-exo^- msh6Δ). Because the double mutant strains are highly mutable and rapidly accumulates mutations that could modulate mutation rates, we minimized the number of generations used to measure rates by sporulating diploid strains that were heterozygous for proofreading. Spore viability for the POL2/pol2-exo^-msh6Δ/MSH6 and POL3/pol3-exo^- msh6Δ/MSH6 strains were 94% and 96%, respectively. The pol2-exo^- msh6Δ mutants have normal colony size and the pol3-exo^- msh6Δ mutants are smaller than normal (Supplementary Fig. 2). 60–100 independent, double mutant spore colonies were used to measure the rate of resistance to 5-FOA, which primarily scores mutations in the URA3 gene, including substitutions resulting from all 12 possible single base mismatches at numerous locations in the coding sequence (Supplementary Fig. 1). In the strains used here, URA3 is located about 2000 base pairs from ARS306, an early-firing and efficient replication origin on chromosome 3. The next closest origin (ARS307) is about 35,000 base pairs away, such that the replication fork originating at ARS306 will generate the majority of errors in URA3, an interpretation strongly supported by recent ribonucleotide mapping studies [5–8]. In this situation, the coding sequence depicted in Supplementary Fig. 1 is the template for lagging strand synthesis, and the complementary non-coding sequence (not shown) is the template for leading strand replication.

3.2. Mutation rates and specificity

In the wild type strain, the base substitution rate for resistance to 5-FOA is low, and close to the unbiased substitution rate for the whole yeast genome [4]. This fact, and the observation that the vast majority of the 5-FOA-resistant mutants have mutations in URA3 (Table 1), underscores the utility and reliability of URA3 as a reporter for genome stability. Compared to the wild type strain, mutation rates in the pol2-exo^-, pol3-exo^- and msh6Δ single mutant strains are elevated by 19- to 45-fold (Fig. 1A). These increases are consistent with earlier studies in yeast demonstrating that defects in proofreading or MMR alone elevate mutation rates (see [23,24] and references therein). The mutation rates in the pol2-exo^- msh6Δ and pol3-exo^- msh6Δ double mutant strains are much higher than the sum of the single mutant strains (Fig. 1A, right). Again, the results are consistent with earlier studies [18,25] indicating that proofreading and MMR act in series to correct replication errors by Pols δ and ε.

Table 1.

Total and specific mutation rates in six yeast strains.

		WT			msh6Δ			pol2-exo-			pol2-exo- msh6Δ			pol3-exo-			pol3-exo- msh6Δ
Mutation rate (×10−7)		0.17			3.3			4.2			290			7.7			1300
95% Confidence intervals		(0.16–0.18)			(3.2–3.5)			(3.6–4.8)			(230–260)			(7.3–8.0)			(1000–1600)
# 5FOA Resistant clones sequenced		211			691			170			178			123			361
			Mutation rates (×10−9)			Mutation rates (×10−9)			Mutation rates (×10−9)			Mutation rates (×10−9)			Mutation rates (×10−9)			Mutation rates (×10−9)
	Target sizea	Total	Mutation typeb	Per basec	Total	Mutation typeb	Per basec	Total	Mutation typeb	Per basec	Total	Mutation typeb	Per basec	Total	Mutation typeb	Per basec	Total	Mutation typeb	Per basec
Base substitutions	285	122	9.8	0.034	292	140	0.49	139	340	1.2	137	22000	78	91	570	2	399	140000	500
Transitions	130	47	3.8	0.029	182	87	0.67	20	49	0.38	39	6400	49	64	400	3.1	363	130000	1000
A to G	18	4	0.32	0.018	2	0.96	0.053	1	2.5	0.14	2	330	18	0	6.3	0.35	2	720	40
T to C	27	7	0.56	0.021	20	9.6	0.35	3	7.4	0.27	2	330	12	37	230	8.6	154	55000	2100
C to T	36	11	0.89	0.025	6	2.9	0.08	4	9.9	0.27	7	1100	32	6	38	1	5	1800	50
G to A	49	25	2	0.041	154	74	1.5	12	30	0.61	28	4600	93	21	130	2.7	202	73000	1500
Transversions	225	75	6	0.027	110	53	0.23	119	290	1.3	98	16000	71	27	170	0.75	36	13000	58
A to T	39	9	0.73	0.019	3	1.4	0.037	91	220	5.8	8	1300	33	5	31	0.8	3	1100	28
T to A	51	7	0.56	0.011	3	1.4	0.028	8	20	0.39	5	810	16	3	19	0.37	3	1100	21
T to G	42	8	0.64	0.015	3	1.4	0.034	0	2.5d	0.059	0	160	3.9	11	69	1.6	11	4000	94
A to C	25	5	0.4	0.016	6	2.9	0.11	6	15	0.59	2	330	13	0	6.3	0.25	0	360	14
C to A	38	11	0.89	0.023	48	23	0.6	1	2.5	0.065	1	160	4.3	5	31	0.82	16	5800	150
G to T	58	19	1.5	0.026	40	19	0.33	11	27	0.47	82	13000	230	2	13	0.22	2	720	12
C to G	20	6	0.48	0.024	1	0.48	0.024	1	2.5	0.12	0	160	8.1	0	6.3	0.31	0	360	18
G to C	33	10	0.81	0.024	6	2.9	0.087	1	2.5	0.075	0	160	4.9	1	6.3	0.19	1	360	11

^aNumber of base pairs in URA3 where this substitution confers 5-FOA resistance, from Supplementary Fig. 1.

^bRates are the total mutation rate times the proportion of the observed mutations of that type (left-most column) among the total 5-FOA^R clones sequenced.

^cMutation rate per phenotypically detectable base pair.

^dItalicized values in bold are “less than or equal to” rates for events not seen, calculated based on the maximum mutation rate had one event been observed.

Fig. 1.

Mutation rates and fidelity factors for base selectivity, proofreading, and MMR on both the leading and lagging strands. (A) Mutation rates were measured at the agp1::URA3-OR1 locus in WT, pol2-exo^-, pol3-exo^-, msh6Δ, pol2-exo^- msh6Δ, and pol3-exo^- msh6Δ strains. Please note that the y-axis is on a logarithmic scale. (B) The apparent fidelity factors for base selectivity, proofreading, and MMR are depicted for the leading strand (dark gray) and the lagging strand (light gray). The y-axis values are number of correctly replicated or repaired bases per mutation and is on a logarithmic scale. The lines within the bars for base selectivity indicate the base selectivity of Pol ε[46] and Pol δ [45] measured in vitro.

Also, the mutation rate in the pol3-exo^- msh6Δ strain is higher than that in the pol2-exo^- msh6Δ strain. Given evidence that the exonuclease activity of Pol δ, but not that of Pol ε, proofreads errors made by Pol α[12], and evidence that Pol δcan also proofread errors made by Pol ε [13], the higher mutation rate in the pol3-exo^- msh6Δ strain could be due to loss Pol δ proofreading of errors made by any of the three replicases, whereas loss Pol ε may only proofread its own errors. The different mutation rates in the pol3-exo^- msh6Δ and pol2-exo^- msh6Δ strains could also be related to differences in activation of the S phase checkpoint in proofreading-deficient strains (see [26] and references therein).

To determine rates for individual substitutions, mutational spectra were obtained by sequencing the URA3 gene in collections of independent 5-FOA^R colonies. The spectra for wild type versus msh6Δ, for pol2-exo^- versus pol2-exo^- msh6Δ, and for pol3-exo^-versus pol3-exo^- msh6Δ strains are shown in Figs. 2, 3 and 4, respectively. Table 1 lists the number of occurrences and the rate per generation for each type of substitution, both before and after correcting the rate for the known number of detectable sites in URA3 (from Supplementary Fig. 1). From these rates, and knowing which strand acts as the template for the majority of leading or lagging strand replication, we calculated the contributions of base selectivity, proofreading and MMR for total substitutions (Fig. 1B) and for substitutions resulting from formation of each of the 12 possible single base mismatches generated during leading and lagging strand replication (Table 2). The contribution of MMR was calculated by dividing the total base substitution rate in the exo^- msh6Δ strains by the corresponding base substitution rates in the exo^-strains. The contribution of proofreading to replication fidelity was determined by dividing mutation rates in the exo^- msh6Δ strains by the corresponding rates in the msh6Δ strain. Apparent base selectivity was calculated as the inverse of the mutation rate for the two exo^- msh6Δ strains that lack both proofreading and MMR. We also examined the effects of DNA sequence context by calculating rates and fidelity factors for substitutions at specific base pairs in URA3 (Fig. 5). The latter examples focus on mismatches generated when dTTP, the dNTP present at the highest concentration in yeast [27], is misinserted opposite template G, C or T.

Fig. 2.

URA3 mutation spectra for the wild type and msh6Δ strains. Mutations observed in the wild type and msh6Δ strains are shown above and below the URA3 coding sequence, respectively. Red letters are transitions, blue letters are transversions, open black triangles are single base pair deletions and filled triangles with a letter are single base pair insertions.

Fig. 3.

URA3 mutation spectra for pol2-exo^- and pol2-exo^- msh6Δ strains. As in Fig. 2.

Fig. 4.

URA3 mutation spectra for pol3-exo^- and pol3-exo^- msh6Δ strains. As in Fig. 2.

Table 2.

Contributions of selectivity, proofreading and MMR to replication fidelity.

	Leading strand			Lagging strand
	Base selectivity	Proofreading	MMR	Base selectivity	Proofreading	MMR
	12000000	160	65	2000000	1000	250
Transitions	21000000	73	130	1000000	1500	330
T-dG	55000000	340	130	490000	5800	240
A-dC	83000000	34	44	25000000	750	>=120
G-dT	32000000	400	120	670000	990	550
C-dA	11000000	62	150	20000000	630	48
Transversions	14000000	300	54	17000000	250	77
T-dT	30000000	910	5.8	47000000	750	58
A-dA	63000000	570	41	36000000	750	35
A-dG	>=260000000	=<110	a	>=69000000	=<130	a
T-dC	77000000	110	22	11000000	2800	58
G-dA	230000000	7.1	66	81000000	38	58
C-dT	4300000	700	490	6600000	250	180
G-dG	>=120000000	=<340	=<66	92000000	130	58
C-dC	>=200000000	=<57	=<66	>=56000000	=<750	b

^aToo few events detected to quantify a contribution.

Fig. 5.

The contributions of proofreading and MMR to correcting base-base mismatches. Results on the left are for the pol2-exo^- strains, and results on the right are for the pol3-exo^- strains. The contributions of proofreading and MMR to correcting: (A) and (B) total base-base mismatches, (C) total C-dTTP mismatches (D) total G-dTTP mismatches, (E) individual C-dTTP mismatches at URA3 base pairs 345 and 679, and (F) individual G-dTTP mismatches at URA3 base pairs 368 and 764. The numbers superimposed within or above the pie slices indicate the error correction factors, either for proofreading (in blue) or MMR (in red).

3.3. Contribution of MMR to replication fidelity

Dividing the total base substitution rate in the pol2-exo^- msh6Δ strain (78 × 10⁻⁹, Table 1) by the corresponding base substitution rate in the pol2-exo^- strain (1.2 × 10⁻⁹) estimates the contribution of MMR to the fidelity of replication of the nascent leading strand. This comparison indicates that on average, MMR corrects 64 of every 65 single base-base mismatches generated by Pol ε in the absence of its intrinsic exonuclease activity (Fig. 1B, right-most dark gray bar, and Table 2). The corresponding comparison of the pol3-exo^- msh6Δ strain to the pol3-exo^- strain indicates even greater MMR efficiency for lagging strand errors, where 249 of every 250 mismatches are corrected. These mismatches are generated during lagging strand replication by Pol δ lacking its intrinsic exonuclease activity, as well as by Pol α, which lacks an intrinsic proofreading activity but whose mistakes may sometimes be edited by Pol δ [12]. These results confirm that MMR efficiently corrects replication errors in both nascent strands. They further imply that MMR is somewhat more efficient at correcting lagging strand mismatches as compared to leading strand mismatches. The latter interpretation is similar to that derived from previous studies examining mismatches containing modified bases [28,29], and from studies of mutator replicases rendered inaccurate by replacing an amino acid in the polymerase active site while leaving the proofreading exonuclease active site unperturbed (see [4,21,22,30–32] and references therein). An important difference between the present study and the earlier reports is that here we monitor mismatches involving unmodified bases in cells where a proofreading exonuclease has been inactivated but the polymerase active site and nucleotide selectivity remain wild type. In all these studies, MMR appears to be more efficient at correcting lagging strand errors as compared to leading strand errors, a fact that we previously suggested may reflect a higher density of signals for MMR in the lagging strand, possibly including PCNA [33] and the 5′-ends of Okazaki fragments in the discontinuously replicated lagging strand [32]. The idea that MMR is somewhat more efficient at correcting lagging strand replication errors due to a higher density of 5D́NA ends is further supported by studies showing that Exonuclease 1 (Exo1), a 5′-exonuclease, contributes more to MMR of lagging as compared to leading strand [34,35], and that Exo1 makes a larger contribution to MMR of errors made by Pol α as compared to errors made by Pol δ, which are always more distal to 5′-DNA ends of Okazaki fragments [35].

On average, transition mismatches are repaired more efficiently than are transversion mismatches (Table 2). This result is consistent with similar observations on MMR in E. coli [1]. Nonetheless, a general conclusion that transition mismatches are repaired more efficiently than transversion mismatches in yeast is unwarranted, because apparent MMR efficiencies vary widely among the 12 mismatches. For example, in the pol2-exo^- background, one of the four transversion mismatches (template C-incoming dTTP, 490-fold MMR correction factor, Table 2) is repaired more efficiently than any of the four purine–pyrimidine mismatches that result in transitions. Moreover, this same pyrimidine–pyrimidine mismatch (C-dT, 490-fold) is repaired much more efficiently than another pyrimidine–pyrimidine mismatch (T-dT, 5.8-fold). MMR efficiency also varies in the pol3-exo^- background, but less so (range of 16-fold, Table 2). In some cases, the efficiency of repairing a mismatch containing the same two bases varies depending on which base is in the template and which base is the incoming nucleotide. The largest such difference is in the pol2-exo^- background, where MMR repairs 489 of 490 template C–dT mismatches but only 21 of 22 T–dC mismatches (Table 2). Similar but smaller MMR biases are also observed in the pol3-exo^- background (Table 2). Theoretically, these variations may reflect the binding affinity of yeast MutSα for mismatches of different composition, which vary more than 10-fold [36], or they may reflect MMR events downstream of MutSα binding.

3.4. Contribution of proofreading to replication fidelity

The contribution of proofreading to replication fidelity was determined by dividing mutation rates in the pol2-exo^- msh6Δ or pol3-exo^- msh6Δ strain by the corresponding rates in the msh6Δ strain. When all substitutions are considered, the average contribution of proofreading to replication fidelity in vivo is 160-fold for Pol ε and 1000-fold for Pol δ (Fig. 2, middle, and Table 2). These results establish that, on average, (i) proofreading contributes several-fold more than MMR to both leading and lagging strand replication fidelity in vivo, and (ii) proofreading of lagging strand mismatches is more efficient than proofreading of leading strand mismatches (Fig. 1B). Just as for MMR, proofreading efficiency also varies widely (Table 2). In the pol2-exo^- background, transversion mismatches are proofread four times more efficiently than are transition mismatches (300-fold versus 73-fold), whereas the opposite is true in the pol3-exo^- background (250 versus 1500). In the pol2-exo^- background, proofreading efficiency among the mismatches varies by more than 100-fold, from 910-fold for T–dT to only 7-fold for G–dA in the pol2-exo^- strain, and from 5800-fold for T–dG to 28-fold for G–dA. Mismatch symmetry also matters, as exemplified by the 11-fold difference in proofreading efficiency between a C–dT and T–dC mismatch in the pol3-exo^- background (250 versus 2800).

3.5. Variations in the relative importance of proofreading and MMR

Replication errors in the URA3 open reading frame are non-uniformly distributed (Figs. 2–4). For example, among 178 independent 5-FOA^R colonies from the pol2-exo^- msh6Δ strain (Table 1), 43 colonies contained a G to T substitution at base pair 679 (Fig. 3). By comparison, a combined total of only 39 G to T substitutions was observed at all 56 other G-C base pairs in URA3 where this substitution results in resistance to 5-FOA (Supplementary Fig. 1). These substitutions reflect uncorrected misinsertion of dTTP opposite template C during leading strand replication by Pol ε. The existence of this and other such hotspots provides the opportunity to quantify the relative contributions of proofreading and MMR to correcting individual mismatches of the same base composition and symmetry. In Fig. 5, average proofreading and MMR correction factors for all substitutions (panels A and B) are compared to average correction factors for two mismatches (panels C and D), and to site-specific examples of these same mismatches (panels E and F). All of these mismatches involve the misinsertion of a dTTP, the dNTP present at the highest concentration in yeast [27], and whose misincorporation is responsible for the greatest proportion of substitutions. In the pol2-exo^- strain, misinsertion of dTTP opposite template C at base pair 345 is corrected more efficiently by proofreading than by MMR, whereas the opposite is true for the same mismatch but in a different sequence context at base pair 679 (Fig. 5E). Similarly in the pol3-exo^- strain, misinsertion of dTTP opposite template G at base pair 368 is corrected more efficiently by proofreading than by MMR, but the opposite is true for the same error but at base pair 764 (Fig. 5F). These examples clearly illustrate that even when mismatches of the same composition and symmetry are considered, the relative contributions of proofreading and MMR to genome stability strongly depend on the local sequence environment. Possible explanations for these observations come from studies showing that proofreading depends on local base-base stacking interactions and on the concentrations of correct dNTPs to be incorporated after misinsertion (reviewed in [24]). Both parameters affect partitioning between polymerization to embed the mismatch in the DNA duplex and thus protect it from removal, versus fraying of the primer terminus to permit movement of the error to the exonuclease active site for removal. Compared to proofreading, these parameters may be less relevant to MMR, whose efficiency may depend more on the affinity of MutSε to bind a mismatch, the flexibility of a mismatched DNA duplex, and/or steps further downstream in the MMR pathway. The observations showing that the relative contributions of proofreading and MMR to replication fidelity vary by mismatch composition, sequence environment, replicase and DNA strand (Table 2, Fig. 5) illustrate the difficulty in predicting (i) the consequences of defects in these two processes on specific genes or other functional non-coding sequences in the genome, and (ii) the relative importance of proofreading versus MMR in preventing the mutations that drive tissue-specific tumorigenesis [23,37–42].

3.6. Contribution of base selectivity to replication fidelity

Base selectivity was calculated here as the inverse of the mutation rate for the pol2-exo^- msh6Δ or pol3-exo^- msh6Δ strains that lack proofreading and MMR. Note that mismatches in the pol3-exo^-msh6Δ strain can be caused by either Pol δ or Pol α, such that the calculated selectivity values are estimates of “apparent” base selectivity, which could be different in vivo (discussed further below). The calculated selectivity values show that on average, Pol ε and Pol δ create only one mismatch that results in a base substitution for every 1.3 × 10⁷ or 2.0 × 10⁶ correct bases incorporated, respectively (Fig. 1B). Thus base selectivity appears to be by far the largest contributor to the fidelity of both leading and lagging strand replication. Interestingly, while Pol ε discriminates about equally well against transition and transversion mismatches (fidelity factors of 2.1 × 10⁷ and 1.4 × 10⁷ correctly replicated bases per mismatch, respectively, Table 2), Pol δ discriminates against transversions (blue events in Fig. 4) better than transitions (red events in Fig. 4) by 17-fold (Table 2, 17 × 10⁶ versus 1 × 10⁶). Among different mismatches generated in the pol2-exo^- background, selectivity against A–dG mismatches (≥2.6 × 10⁸) is 60-fold higher than against C–dT mismatches (4.3 × 10⁶), whereas in the pol3-exo^- background, selectivity against G–dG mismatches (9.2 × 10⁷) is 140-fold higher than against G–dT mismatches (6.7 × 10⁵). These results imply that, during nuclear DNA replication in vivo, (i) apparent base selectivity is higher for leading than lagging strand replication, and (ii) the two major nuclear replicases have substantially different misincorporation specificities regarding mismatch composition (Table 2) and sequence context (different hotspots in Figs. 3 and 4).

When the dNTPs responsible for the substitutions in Table 1 are totaled irrespective of the template base, dTTP is misincorporated more often than the other three dNTPs, and during replication of both strands. This fact is interesting because the intracellular concentration of dTTP in yeast is about 2 to 3-fold higher than the concentrations of the other three dNTPs [27]. This suggests that this natural dNTP pool imbalance influences replication error rate and error specificity by Pols ε and α, which has implications in mammalian cells where dNTP pools are also naturally imbalanced [43,44].

The rank order of selectivity against certain mismatches is somewhat unexpected. Previous studies in a variety of systems led to the idea that the most commonly formed base–base mismatch is G–T, whereas pyrimidine–pyrimidine mismatches, especially C–dC, are rarely formed. This generally appears to be the case for lagging strand replication, where selectivity against G–dT and T–dG is at least 10-fold lower than for the other 10 mismatches, and where selectivity against C–C and the other pyrimidine–pyrimidine mismatches is very high (Table 2). However, this does not appear to be the case for leading strand replication, where the lowest selectivity is for the C–dT mismatch, and selectivity against this and two of the other three pyrimidine–pyrimidine mismatches (T–dC and T–dT) is similar to selectivity against G–dT and T–dG mismatches. Differences in the error specificity of Pol ε and Pol δ are still apparent, albeit reduced, even after subtracting the events at the hottest of the substitution hot spots in the spectra (e.g., transversions at base pairs 279, 679 and 686 in Fig. 3). These data suggest that the active sites of the two major nuclear replicases impose different structural, kinetic and/or thermodynamic constraints on formation of mismatches of the same composition.

3.7. In vivo influences promoting increased replication fidelity

Proofreading-deficient Pol ε and Pol δ generate only one error for every 12,000,000 or 2,000,000 correct bases replicated in vivo, respectively (Fig. 1B). Both values are much higher than the base selectivity of proofreading-deficient yeast Pol ε and Pol δ measured during DNA synthesis in vitro (black lines in bars in Fig. 1B), where one error is observed for every 4.200 and 7700 correct bases incorporated, respectively [45,46]. These differences are large, 2900-fold for Pol ε and 280-fold for Pol α. There are several non-exclusive possibilities for these differences. One fact that may be relevant is that the DNA synthesis reaction mixtures in vitro contain all four dNTPs at a concentration of 100 μM, whereas dNTP concentrations in yeast are lower and slightly imbalanced [27,47]. Other reaction parameters in vitro may be less than optimal for high selectivity, including the pH and magnesium concentration, both of which are known to affect the fidelity of DNA synthesis in vitro [48–50]. Replication accessory proteins not yet examined in vitro could also improve fidelity in vivo, although studies to date have not revealed large increases in nucleotide selectivity by accessory proteins.

A perhaps more interesting possibility for the higher “apparent” base selectivity in vivo is extrinsic proofreading. Biochemical [51] and genetic evidence [2,12] suggest that errors made by naturally exonuclease-deficient Pol α, whose primary role is in lagging strand replication, can be proofread by Pol δ. Thus it is possible that in the absence of only one of the two proofreading activities intrinsic to Pol ε and Pol δ, the exonuclease that remains intact in one replicase can proofread mismatches generated by the other major, but proofreading-deficient replicase [2,12]. Indeed, a recent study [13] provides evidence that Pol δ proofreads errors made by Pol ε. Theoretically, extrinsic proofreading may also be catalyzed by other 3′-exonucleases (e.g., see [52]). To the extent that extrinsic proofreading may occur during replication in yeast, this implies that (i) the actual base selectivity of the replicases in vivo could be substantially lower than calculated here, and therefore more in line with the estimates from studies in vitro, and (ii) the base selectivity calculated here for lagging strand replication may be a mixture of the base selectivity of Pol α plus Pol ε, because both polymerases contribute to the mature lagging strand (see [21,53], and more recently [5,8]).

Also of interest is genetic evidence suggesting that Pol ε does not efficiently proofread errors made by Pol α [12] or errors made by proofreading-deficient Pol ε [13]. Given the participation of Pol δ in several other DNA transactions in cells, it may be that Pol δ is generally more efficient at extrinsic proofreading than is Pol ε. This could account for the 10-fold greater difference in apparent base selectivity in vivo versus in vitro mentioned above (2900-fold for the leading strand, 280-fold for the lagging strand). Possibly relevant here is evidence that Pol ε has a strong interaction with the CMG (Cdc45, Mcm2-7, and GINS) complex, while Pol δ has a very weak interaction with the same complex [54]. In contrast, Pol δ has a strong interaction with PCNA, which stimulates its processivity. It is possible that if mismatches created by Pol ε during leading strand replication cannot be proofread by its intrinsic exonuclease, Pol ε can transiently dissociate from the terminal mismatch to allow Pol δ proofreading, and then Pol ε may re-engage and continue leading strand replication [54,55]. Still other possibilities may lie in the identity of yet-to-be disclosed suppressors of mutability in MMR-deficient pol2-4 strains, leading to the suggestion that factors in addition to proofreading and MMR influence leading-strand DNA replication fidelity [56].

3.8. Complementarity among the three DNA replication fidelity processes

The observation that the two major nuclear replicases have substantially different base selectivity against different mismatches means that the two downstream error correction processes must meet different challenges in order to achieve high fidelity replication of both DNA strands. That proofreading and MMR have evolved to effectively meet these different challenges is illustrated by the following observations. (i) Average base selectivity is lower during lagging strand replication, but average proofreading and MMR efficiencies are higher, whereas the opposite holds for leading strand replication (Fig. 1B). (ii) On average, transitions are generated at a 20-fold higher rate in the pol3-exo^- msh6Δ strain (1300 × 10⁻⁷, Table 1) as compared to the pol2-exo^- msh6Δ strain (64 × 10⁻⁷), and this difference is counterbalanced by the fact that proofreading of transition mismatches is 20-fold more efficient on the lagging strand (Table 2, correction factor 1500-fold, as compared to the leading strand, 73-fold). (iii) As compared to the above example, average selectivity against transversion mismatches is similar during leading and lagging strand replication, and so too are proofreading and MMR efficiencies (Table 2). (iv) Selection against a C–dT mismatch is relatively low in the pol2-exo^- background, and highly efficient proofreading and MMR counterbalance this low selectivity. (v) Conversely, high selectivity against certain mismatches, such as C–dA, A–dC, G–dA, is associated with lesser contributions of proofreading and MMR. These observations, in combination with earlier studies of engineered variants of replicases harboring polymerases active site point mutations (see [4] and references therein), support the idea that the eukaryotic replicases, their attendant 3′-exonucleases, and MMR have coevolved to most efficiently correct replication errors made at the highest rates and in the most risky sequence contexts, in order to accurately replicate both DNA strands of the nuclear genome. This idea, and the fact that the contribution to replication fidelity of base selectivity, proofreading and MMR all strongly depend on mismatch composition and symmetry, the replicase, and the strand and sequence context in which the mismatch resides, likely contribute to defining the composition of nuclear genomes. The balancing act among the three replicases may partly underlie differences in life span and tissue-specific tumorigenesis in mice encoding proofreading deficient alleles of Pols ε and δ [57]. These differences may also be relevant to the nature and number of mutations that are thought to drive tissue-specific tumor development in humans harboring mutations in the exonuclease motifs of Pols ε and δ [23,37–42].

Appendix A. Supplementary data

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.dnarep.2015.04.006