|Home | About | Journals | Submit | Contact Us | Français|
The mechanisms by which imbalanced dNTPs induce mutations have been well characterized within a test tube, but not in vivo. We have examined mechanisms by which dNTP imbalances induce genome instability in strains of Saccharomyces cerevisiae with different amino acid substitutions in Rnr1, the large subunit of ribonucleotide reductase. These strains have different dNTP imbalances that correlate with elevated CAN1 mutation rates, with both substitution and insertion–deletion rates increasing by 10- to 300-fold. The locations of the mutations in a strain with elevated dTTP and dCTP are completely different from those in a strain with elevated dATP and dGTP. Thus, imbalanced dNTPs reduce genome stability in a manner that is highly dependent on the nature and degree of the imbalance. Mutagenesis is enhanced despite the availability of proofreading and mismatch repair. The mutations can be explained by imbalanced dNTP-induced increases in misinsertion, strand misalignment and mismatch extension at the expense of proofreading. This implies that the relative dNTP concentrations measured in extracts are truly available to a replication fork in vivo. An interesting mutational strand bias is observed in one rnr1 strain, suggesting that the S-phase checkpoint selectively prevents replication errors during leading strand replication.
The four deoxyribonucleoside triphosphates, dATP, dTTP, dGTP and dCTP, are essential precursors for the DNA synthesis, which is required for replication, recombination and repair. Because these DNA transactions are needed to maintain genome stability, perturbations in the absolute and relative concentrations of the four dNTPs increase mutation rates by reducing the fidelity of DNA synthesis (1). Changes in dNTPs concentration are mutagenic and may occur due to mutations in enzymes involved in dNTP metabolism or changes in the environment. The mutational mechanisms induced by imbalanced dNTP pools within cells have not been extensively investigated. In vitro studies performed with purified DNA polymerases have revealed several mechanisms by which dNTP perturbations reduce fidelity [reviewed in (2)]. One mechanism predicts that imbalanced dNTP pools increase the probability that a DNA polymerase will misinsert an incorrect dNTP [reviewed in (3)]. For example, an abnormally high ratio of dTTP as compared to dGTP can promote misinsertion (MI) of dTTP opposite a template C (Figure 1, top left). Additional studies in vitro indicate that imbalanced dNTP concentrations can also increase the rate of formation of insertion–deletion (indel) errors during DNA synthesis (4,5). For instance, when the ratio of the correct dNTP to the incorrect dNTP needed for synthesis within a mononucleotide run strongly favors the incorrect dNTP, the probability of DNA strand misalignment (MA) (Figure 1, top right) is increased, thereby increasing indel error rates. Alternatively, a dNTP imbalance may induce an MI that, in the appropriate sequence context, can result in primer relocation [(PR), Figure 1, second pathway from left] to create an indel intermediate with one or more correct terminal base pairs. Another possibility supported by evidence in vitro (5–10) is MA followed by correct incorporation (IN), then realignment (RA), thereby creating a base–base mismatch that was initiated by MA (Figure 1, rightmost pathway). Finally, in vitro studies show that the probability that mismatches will eventually result in substitutions or indels are increased if the nucleotides to be incorporated immediately following the mismatch (colored green in Figure 1) are present at high enough concentrations to promote rapid extension (RE) from the mismatch prior to proofreading and/or RA [reviewed in (3,7)].
The goal of the present study is to determine the extent to which the concepts outlined in Figure 1 may explain the specificity of mutagenesis observed in vivo when dNTP pools are perturbed. In previous studies, mutations induced by dNTP pool imbalances were analyzed in phage T4 (rII and thymidine kinase genes), Escherichia coli (lacI gene), Saccharomyces cerevisiae (SUP4-o gene), mouse (integrated bacterial gpt gene), Chinese hamster (aprt gene) and human (HPRT gene) cells (11–22). The majority of the mutations produced by dNTP pool imbalances were found to be single base pair events dominated by substitutions, whereas indels were recovered less frequently and not in each investigation. Some of these studies suggested that dNTP pool imbalances promote mutagenesis by suppressing proofreading activity of DNA polymerases (2). In this study, we took advantage of recent structure–function studies that have advanced our understanding of how dNTP pools are regulated by ribonucleotide reductase (RNR) (23). RNR is responsible for catalyzing the rate-limiting step in the synthesis of all four dNTPs and is comprised of multiple subunits. Yeast RNR is highly regulated on many levels: at the G1/S boundary and in response to DNA damage, RNR genes are transcriptionally upregulated (24–27), the levels of the RNR inhibitor Sml1 are post-transcriptionally reduced (28) and the small subunit Rnr2/Rnr4 proteins are re-distributed to the cytoplasm, where the large subunit Rnr1/Rnr3 proteins are localized (29–31). Additionally, the activity of RNR is controlled allosterically (32). The large subunit contains an allosteric specificity site responsible for controlling the balance of the four dNTPs. The specificity site influences the specific ribonucleoside diphosphate reduction reaction within the catalytic site, which is also present in the large subunit (33).
In this study, we utilized yeast strains with single amino acid substitutions in loop 2 of Rnr1 (23). Loop 2 is highly conserved throughout the evolutionary ladder from yeast through humans (34–36). The residues of loop 2 are important for the connection between the specificity site and the catalytic site (34,36). Specifically, three yeast strains were created that encode a different single amino acid change in one of two conserved residues in loop 2, tyrosine 285 (Y285) or glutamine 288 (Q288). As previously reported, strains with a change at Y285 have no noticeable proliferation defects, whereas strains with mutations at Q288 are slow proliferating and exhibit a cell cycle checkpoint response (23). Each of the three mutations studied here (Y285F, Y285A and Q288A) have a different effect on dNTP pool balance, and spontaneous mutation rates. Here, we present an analysis of mutational specificity at the CAN1 locus in these strains, and interpret the results in relation to the different dNTP pool imbalances in these strains and the models in Figure 1.
Cells were grown without selection in YPDA medium (1% yeast extract, 2% bacto-peptone, 2% dextrose, 250mg/l adenine, 2% agar for plates). Canr colonies were scored on synthetic complete medium containing 2% dextrose (SCD) lacking arginine and supplemented with 60µg/ml canavanine. All growth was at 30°C.
At least two independent isolates for each strain were used for both mutation rate analysis and mutation spectra assembly. Mutation rate protocol was previously described (23). Cultures were inoculated with single colonies and grown without selection in 5ml YPDA to saturation (~2×108cells/ml). Cells were washed with sterile double distilled water (ddH2O) and resuspended in 1ml of sterile ddH2O. To calculate cell viability and mutation rates, 50–100µl of the appropriate dilution was plated on YPDA and on SCD–Arg with canavaine, respectively. Colonies were counted after 2–4 days of growth. Mutation rates and associated 95% confidence intervals were determined by a modified method of the median (37,38).
A single canavanine-resistant colony from each culture was used to generate the spontaneous mutation spectrum to guarantee that each mutant resulted from an independent event. Mutant colonies were purified and genomic DNA was extracted. The sequence of the CAN1 locus was amplified from the isolated DNA in two separate polymerase chain reactions (PCRs), and the resulting PCR products underwent automated DNA sequence analysis (Macrogen Inc. Seoul, Korea). The first half of the CAN1 locus was amplified using primers PR1 5′-TCAGGGAATCCCTTTTTGCA and PR7 5′-TGAAATGTGAAGGCAGCGTT with the resulting PCR-amplified fragments sequenced by primers PR3 5′-CCAGTGGGCGCTCTTATA and PR8 5′-CGCCAGTGGAACTTTGTA. The latter half of the CAN1 locus was PCR amplified using primers PR2 5′-GAAATGGCGTGGAAATGTGA and PR4 5′-TTACCGGCCCAGTTGGAT, this PCR product was the sequenced by PR5 5′-CAACCATTATTTCTGCCG and PR9 5′-CCTGCAACACCAGTGATA. Sequences were analyzed using Sequence Manager Software (DNA STAR Inc.).
Monte Carlo statistical analysis was performed to compare the distribution (number and location) of mutation events between spectra (39,40). Chi-square analysis was used to determine if a specific event occurred at a rate that is significantly different from the wild-type strain. Probability (P) values of <0.05 were considered significantly different.
We studied yeast strains harboring three different mutations in loop 2 of Rnr1, rnr1-Y285F, rnr1-Y285A and rnr1-Q288A. The rnr1-Y285F and rnr1-Y285A mutant strains progress through the cell cycle normally, whereas the strain containing the rnr1-Q288A mutation proliferates slowly and has a prolonged S-phase (23). The dNTP concentrations of the three mutant strains were measured by high performance liquid chromatography as reported previously (23). Compared to the dNTP levels in a wild-type strain, the rnr1-Y285F mutant strain has a 3-fold increase in the concentration of dTTP and dCTP, 1.8-fold increase in the concentration of dATP but no change in dGTP (Figure 2). The rnr1-Y285A mutant strain has dTTP, dCTP and dATP levels that are increased by 20-, 17- and 2-fold, respectively, but no change in dGTP (Figure 2). Finally, the rnr1-Q288A strain has dATP and dGTP levels increased by 6.6- and 16-fold, respectively, no significant change in the concentration of dTTP, and a 12-fold decrease in dCTP compared to the level in the wild-type strain (Figure 2). The low dCTP concentration could cause a stall in replication fork progression, explaining the slow S-phase progression and S-phase checkpoint activation in the rnr1-Q288A strain (23). In addition to having alterations in dNTP pools, the three rnr1 mutant strains have different spontaneous rates of mutation. The mutation rates were measured at the CAN1 locus, using an assay that detects mutations that yield a non-functional protein, thereby allowing cells to grow on media containing canavanine. The mutation rate in the wild-type strain was 4×10−7. The rates in the rnr1-Y285F, rnr1-Y285A and rnr1-Q288A strains were higher, at 11×10−7, 57×10−7 and 17×10−7, respectively (23).
The location and identity of the sequence changes in the open reading frame that were responsible for resistance were determined by the isolation of independent canavanine-resistant colonies, preparation of the colonies genomic DNA and sequencing of the CAN1 locus (Figure 3A). The rates of specific types of mutations were then calculated by multiplying the percentage of that event by the total mutation rate. The sequence changes associated with the 93 mutants sequenced of the wild-type strain (Table 1) were distributed throughout the open reading frame (Figure 3B). Sixty-five of the mutations were single base substitutions, 11 were single base indels and the remaining mutations were larger deletions and complex events involving multiple base pairs. This specificity is similar to wild-type CAN1 mutation spectra reported by others (41,42).
For the rnr1-Y285F strain, 93 mutants were sequenced (Table 1). The spectrum (Figure 3C) was largely comprised of base substitutions (78 of 93), such that the substitution rate was 3-fold higher than in the wild-type strain (Table 1, 9.2×10−7 versus 2.9×10−7). Only 17% of the mutations in the rnr1-Y285F spectrum share a location in common with those in the wild-type spectrum. This difference is highly statistically significant (P<0.0001).
The spectrum in the rnr1-Y285A strain, which has elevated dTTP and dCTP and a higher mutation rate, is strikingly different. At 18 locations in CAN1 (red and blue symbols in Figure 3D), mutation rates were 20- to >300-fold higher than in the wild-type strain. Moreover, the spectrum is dominated by indels (101/173), yielding an average increase in the indel rate of 66-fold compared to the wild-type strain (Table 1, 0.5×10−7 versus 33×10−7). Additionally, 97% of the indel mutations (98 of 101, Table 1) are single base pair deletions (triangles in Figure 3D), most of which are in one of three hotspots between base pairs 780–880 (Figure 3D). Seventy-two base substitutions were also observed, corresponding to an 8-fold increase in substitution rate relative to the wild-type strain (Table 1, 24×10−7 versus 2.9×10−7). Interestingly, >80% of the base substitutions were transversions (Table 1), the majority of which (45 of 59) occurred at G•C base pairs.
The rnr1-Q288A strain has a mutation spectrum (Figure 3E) that differs significantly (P<0.0001) from the three other spectra by Monte Carlo analysis. In this strain, 124 of 169 mutants were single base substitutions, 65% of which were transitions (Table 1). The distribution of substitutions (circles and squares in Figure 3E) was non-random, such that substitution rates at five different locations in CAN1 were from 10- to 50-fold higher than the wild-type strain. In addition, 41 of the rnr1-Q288A mutants had indel mutations corresponding to an 8-fold increase in the overall indel error rate (Table 1, 4.1×10−7 versus 0.5×10−7). Similar to the base substitution mutations, the indels cluster into five mutation hotspots with mutation rates that are 10-fold higher than in the wild-type strain (triangles in Figure 3E).
Table 2 list the nucleotide positions within the CAN1 gene in the rnr1-Y285A and rnr1-Q288A strains where single base substitution and deletion rates are at least 10-fold higher than the rate in the wild-type strain. The events listed in the tables correspond to the red and blue hotspots displayed in Figure 3. The red hotspots are mutations that are predicted to have occurred when the non-coding strand is the template for replication, and the blue symbols represent mutations that are predicted to have occurred when the coding strand is the template for replication (Figure 3A). Also listed in Tables 2 and and33 are the number of occurrences of each event at the hotspot, the corresponding mutation rates and the fold-increase in rate over the mutation rate for that event in the wild-type strain. The last column in the tables shows the sequences of both DNA strands at the hotspots, with the presumptive error in red and the adjacent correct nucleotides that are in excess in green. The green letters in bold print correspond to correct bases with dNTP concentrations at least 5-fold greater than the dNTP concentration measured in the wild-type strain. The green letters in non-bold print represent dNTPs present at levels that are 2- to 5-fold greater than the wild-type levels. The DNA sequences are displayed in this manner to facilitate discussion of possible mutational mechanisms (see later).
The major goal of this study was to determine the extent to which the ideas outlined in Figure 1 can explain the specificity of mutagenesis observed in vivo when dNTP pools are selectively perturbed. Despite the fact that the rnr1-Y285A and rnr1-Q288A strains harbor very different dNTP pool imbalances and do not have a single mutational hotspot in common (red and blue symbols in Figure 3D/E), all 28 hotspots in CAN1 can be explained by dNTP pool induced increases in three processes that determine DNA replication fidelity: dNTP MI, DNA strand MA and mismatch extension at the expense of proofreading.
The highest mutation rate for any hotspot (12×10−7, Table 2) is for the loss of a G•C base pair at positions 857–859. This hotspot was observed in the rnr1-Y285A strain, where the dNTP concentrations are lowest for dGTP and highest for dCTP and dTTP. The deletion is consistent with two models depicted in Figure 4. One (Figure 4A, left branch) involves increased MA when replicating a CCC template, followed by RE of the misaligned intermediate via IN of 10 consecutive correct dTTPs and dCTPs, which are in excess. In fact, all six of the deletion hotspots in the rnr1-Y285A strain involve template cytosine run with adenine as a 5′neighbor (Table 2). Therefore, the high dGTP:dTTP ratio in the rnr1-Y285A strain could promote deletions via MI of dTTP opposite template C, followed by PR (second model, Figure 4A, right branch). It is also possible that an MA within the polymerase active site (6) may be stabilized by excess dTTP, in a model called dNTP stabilized MA (43,44). These models are not mutually exclusive, and each of them is consistent with the deletions observed in these sequence contexts resulting from the rnr1-Y285A induced dNTP pool imbalances.
The five deletion hotspots in the rnr1-Q288A strain are at different locations than the hotspots observed in the rnr1-Y285A strain (Figure 3 and Table 2). Two hotspots in the rnr1-Q288A strain (at 1247–1248 and at 1258–1259) involve deleting a G•C base pair from small G•C runs, and a third (964–969) involves deleting an A•T base pair from a 6bp A•T homopolymer run, one of the longest mononucleotide runs within the CAN1 locus. These hotspots have explanations similar to those in the rnr1-Y285A strain (see last column in Table 2), but reflect the different dNTP pool imbalance in the rnr1-Q288A strain. Note that in all three cases, one of the next correct nucleotides, dTTP, is not in large excess. This might result in less efficient extension and correspondingly more proofreading, perhaps contributing to the lower rate of mutation at these sites as compared to the deletion hotspots in the rnr1-Y285A strain.
The two hotspots in the rnr1-Q288A strain that occur at nucleotides 476 and 703 involve deleting a non-iterated G•C base pair (Table 2). This is particularly informative, in that non-iterated bases are not usually seen as hotspots for deletions. Commonly, strand MAs at non-iterated nucleotides do not generate intermediates that can be stabilized by the correct base pairing, which is possible at repetitive sequences [reviewed in (45)]. Because these two non-iterated sites accumulate multiple mutations in a strain with imbalanced dNTP pools, we suggest that the initiating event for these deletions is MI of dTTP opposite template G (e.g. site 476, Figure 4B), followed by PR to allow T to pair with its 5′template neighbor, an adenine. In both cases, the next correct nucleotides are in excess to promote extension of the MA.
Despite being completely different in the rnr1-Y285A strain as compared to the rnr1-Q288A strain (Table 3), all 18 base substitution hotspots can be rationalized by dNTP pools that increase dNTP MI and then also increase subsequent mismatch extension. For example, the C to A substitution hotspot at position 648 (Figure 4C) in the rnr1-Y285A strain would result from MI of dTTP opposite template C due to the high ratio of dTTP compared to dGTP, followed by the RE of approximately nine nucleotides from the mismatch, which is promoted by the high concentrations of dTTP and dCTP. Similarly, the G to A substitution hotspot at position 1018 in the rnr1-Q288A strain (Figure 4D) would result from MI of dTTP opposite template G due to the high ratio of dTTP compared to dCTP, followed by rapid IN of the next 15 correct nucleotides promoted by the excessively high concentrations of dATP and dGTP.
A non-exclusive alternative can be considered for two of the eight substitution hotspots in the rnr1-Y285A strain. At base pairs 314(a) and 1440 (Table 3), a G to T substitution occurs at the 5′-G in a run that in each case is flanked by a T. At these sites, the rightmost pathway in Figure 1, involving MA, correct IN, RA and mismatch extension, could also contribute to formation of these substitutions.
Competition by mass action for correct versus incorrect dNTP insertion at the polymerase active site also rationalizes a more general difference between the two rnr1 mutant strains involving base substitutions at G–C base pairs (Table 1). The rnr1-Y285A mutation enhances transversions to a greater extent than it enhances transitions, especially at G•C base pairs (Table 1). This is anticipated by the unusually high relative ratio of dTTP to dGTP (17:1) as compared to the more normal relative ratio of dATP to dGTP (2:1) (Figure 2). In comparison, the rnr1-Q288A mutation enhances G to A transitions to a slightly greater extent than it enhances G to T transversions (Table 1). The concentrations of both dTTP and dATP are significantly higher than the concentration of dCTP. The relative ratios of dTTP to dCTP and of dATP to dCTP are 15:1 and 83:1, respectively (Figure 2). Interestingly, even though dTTP is not as abundant as dATP it is the favored base substitution in the rnr1-Q288A strain.
As mentioned earlier, the six deletion hotspots in the rnr1-Y285A strain were all runs of G•C base pairs. There are 28 runs of three or more G•C base pairs in the CAN1 open reading frame, 14 with guanine and 14 with cytosine in the coding strand. Interestingly, five of the six hotspots involved coding strand guanine runs (Table 2).
This is a statistically significant difference by chi-square analysis (P=0.0005), indicating a strand bias in formation of single base deletions resulting from imbalanced dNTP pools. Further evidence for a strand bias comes from comparing the hotspots depicted in red versus blue colors in Figure 3. In the rnr1-Y285A spectrum, 13 of the 18 hotspots are predicted to have resulted from replication of the non-coding strand of the CAN1 gene (red), and five are predicted to have resulted from replication of the CAN1 coding strand (blue). In contrast, all 10 hotspots in the rnr1-Q288A strain are predicted to result from replication of the coding strand of the CAN1 gene. Additionally, the mutation hotspots within the CAN1 locus of a strain harboring and over-expressing a plasmid copy of rnr1-Q288A were also predicted to exclusively occur on the coding strand (data not shown). This difference between the rnr1-Y285A and rnr1-Q288A strains is highly significant (P=0.0008).
This putative asymmetric distribution of mutations on the coding strand in the rnr1-Q288A strain but not in the rnr1-Y285A strain suggests that mutagenesis can be differentially modulated during leading strand and lagging strand replication, depending on the nature of the pool imbalance. How might this occur? Previous analysis of early firing replication origins on the left arm of chromosome five of S. cerevisiae has shown that replication of the CAN1 gene originates from ARS507 and travels through CAN1 towards the telomere (46–48). This predicts that the non-coding strand should be the template for leading strand replication, and that the coding strand should be the template for lagging strand replication. Current evidence suggests that the leading strand template may primarily be replicated by Pol ε (49), whereas the lagging strand template may primarily be replicated by Pol δ (50). These facts are intriguing, given that (i) the rnr1-Y285A strain proliferates normally but the rnr1-Q288A strain proliferates slowly, has a defect in S-phase progression and elicits a checkpoint response (23) and (ii) Pol ε, but not Pol δ, is involved in the S-phase checkpoint response (51,52). Thus, the surprising absence of putative non-coding (i.e. leading) strand hotspots in the rnr1-Q288A strain despite the large dNTP pool imbalance may be related to Pol ε’s combined roles in leading strand replication and in the S-phase checkpoint response. In other words, as Pol ε attempts leading strand replication using too little dCTP and excess dATP and dGTP, certain mismatches that are known to be particularly difficult to extend [e.g. G-dAMP and G-dGMP, (53)] may be prevented from yielding mutations via Pol ε’s checkpoint function. Conceptually, this is similar to the idea that checkpoints initiated by DNA lesions stall replication to provide more time for DNA repair. In the present case, the checkpoint response may provide more time for error correction from proofreading by Pol ε, whose 3′-exonuclease is processive and highly active (54,55). A checkpoint might even allow more time for mismatch repair to correct leading strand replication errors, especially if replication and mismatch repair are coupled (56). Because there are still many coding (i.e. putative lagging) strand mutational hotspots in the CAN1 spectrum in the rnr1-Y285A strain, a checkpoint mechanism to reduce mutagenesis might not apply to lagging strand replication errors made by Pol δ, a polymerase with no reported role in the S-phase checkpoint response (Figure 5). It should also be noted that an activated S-phase checkpoint inhibits replication origin firing (57–59), which could be relevant to the strand biases observed here. Alternatively, it is also possible that differences in the distribution of mutations on the two strands might be related to differences in the efficiency with which pols α/δ versus pol ε extend mismatches driven by high dTTP/dCTP versus those that are driven by high dATP/dGTP. In the future, we plan to further examine this unexpected strand specificity, using rnr1 mutants in combination with mutator alleles of DNA polymerases δ and ε, in a system previously used to study leading and lagging strand replication fidelity in yeast (49,50).
The rnr1-Y285A strain has a higher dNTP pool imbalance and a higher mutation rate than does the rnr1-Y285F mutant (23). This correlation, and the fact that the observed mutational specificity can be rationalized by the models in Figure 1, implies that the elevated mutation rates for the rnr1 mutant strains are indeed the result of reduced DNA replication fidelity in vivo. It then follows that the dNTP concentrations that were measured here in extracts reflect the relative biologically relevant ratios of dNTPs that are available to a replication fork in vivo. This suggests that selective compartmentalization or channeling of dNTPs is not occurring in vivo, at least in budding yeast (1,60). This validates the continued use of HPLC analysis of cell extracts to measure biologically relevant dNTP pools.
Because dNTP pool imbalances reduce genome stability to different extents and with specificities that are highly dependent on the nature and degree of the imbalance, it follows that different dNTP pool imbalances may put specific genes at increased risk of inactivation by substitution and indel mutations. In this study, this risk is apparent despite the fact that the two major replication error correction mechanisms, proofreading and DNA mismatch repair, are genetically intact. Thus, it is possible that the mutagenesis observed here is merely the ‘tip of the iceberg’. Much of the mutagenic potential of dNTP pool imbalances may be offset by error correction, which could explain why some sites exist in CAN1 where we might expect to see mutations were not hotspots. As one example, no deletions were observed at a run of three G•C base pairs at positions 387–389, even though the sequence context (5′-GGGTTC) is remarkably similar to that of the deletion hotspot at positions 857–859 (5′-GGGTTTC) observed in the rnr1-Y285A strain. The mutational specificity observed here (Tables 2 and and3)3) indicates that at least one of the two well-known error correction mechanisms, proofreading, is being suppressed via rapid mismatch extension due to high concentrations of the next correct nucleotides to be incorporated after a mismatch is generated. However, we were surprised to observe such a high proportion of single base deletion mutations in the rnr1-Y285A strain, because the mismatch repair machinery normally corrects single base indel mismatches very efficiently. In fact, indels are rarely observed in CAN1 spectrum of mutator strains (41,61,62), unless mismatch repair is inactivated (41). The deletion and substitution mutations observed here obviously escaped mismatch repair, perhaps because the dNTP pool imbalance result in so many replication errors that mismatch repair is partially saturated (63,64), or alternatively, because the hotspots occur in sequence contexts where mismatch repair is intrinsically less efficient (65).
Division of Intramural Research of the National Institutes of Health (Project Z01 ES065070 to T.A.K.); Stiftelsen J.C. Kempes Minnes Stipendiefond (to D.K. and J.V.); Swedish Foundation for Strategic Research (to A.C.); Swedish Research Council (to A.C.); The Swedish Cancer Society (to A.C.). Funding for open access charge: The Swedish Research Council.
Conflict of interest statement. None declared.
We thank Roel Schaaper and Jessica Williams for critically reading and editing the manuscript.