Increased Mutagenesis and Unique Mutation Signature Associated with Mitotic Gene Conversion

See allHide authors and affiliations

Science  02 Jul 2010:
Vol. 329, Issue 5987, pp. 82-85
DOI: 10.1126/science.1191125

This article has a correction. Please see:


To examine the fidelity of DNA synthesis during double-strand break (DSB) repair in Saccharomyces cerevisiae we studied gene conversion in which both strands of DNA are newly synthesized. The mutation rate increases up to 1400 times over spontaneous events, with a significantly different mutation signature. Especially prominent are microhomology-mediated template switches. Recombination-induced mutations are largely independent of mismatch repair, by DNA polymerases Polζ, Polη, and Pol32, but result from errors made by Polδ and Polε. These observations suggest that increased DSB frequencies in oncogene-activated mammalian cells may also increase the probability of acquiring mutations required for transition to a cancerous state.

A substantial increase in mutation rates has been proposed to explain how a single cell can sustain multiple mutations within a finite life span, resulting in malignancy (1). The origin of this mutator phenotype remains unknown, although an interesting correlation exists between increased DNA damage levels observed in oncogene-activated precancerous cells (2) and the timing of this necessary mutation rate increase. Most oncogene-induced DNA damage appears to be double-strand breaks (DSBs), which pose a severe challenge to maintaining genome integrity. Various homologous recombination (HR) mechanisms (3) repair DSBs, among which gene conversions (GCs) pose the least genomic threat. GCs have been associated with increased mutation rates (46), largely dependent on the error-prone translesion polymerase, Polζ; however, these errors were measured in presumably single-stranded DNA adjacent to a DSB.

To directly examine GC-associated mutation rates occurring in repaired DNA consisting of two newly synthesized strands, we took advantage of budding yeast mating-type (MAT) switching initiated by an HO endonuclease-induced DSB and repaired by synthesis-dependent strand annealing (SDSA) (Fig. 1A) (7). We modified the heterochromatic donor HMRa, replacing the HMRa1 open reading frame (ORF) with Kluyveromyces lactis URA3 (Kl-URA3) (fig. S1). DSB repair efficiency was ~99%, and nearly all repair events yielded Ura+ (i.e., mata1::Kl-URA3) cells. However, Ura mutants were selected using 5-fluoroorotic acid. Cells acquiring Ura mutations during recombination become Ura+ if hmr::Kl-URA3 remains unaltered and is desilenced (8). Among exponentially growing cells, recombination-induced Ura mutagenesis was 240 times as high as the spontaneous rate of mata1::Kl-URA3 (Table 1) and was specific to the repaired MAT locus. The GC-associated mutation rate in cells synchronously released from G1 was 5.9 times as high as asynchronous cultures (P = 0.0001) and 1400 times as high as spontaneous events (Table 1).

Fig. 1

(A) MATα switching occurs through SDSA (4). DNA synthesis beginning in HMR-Z copies Y sequences and progresses into HMR-X. The nascent strand dissociates and anneals to MAT-X sequences on the other side of the DSB. Nonhomologous sequences are clipped off, and second-strand synthesis initiates using the first strand as a template. Nonprocessive DNA synthesis results in premature dissociation from HMR, which can result in error-prone reannealing to the HMR donor template or to other ectopic templates. To complete the GC, template switch events must dissociate again and reanneal to HMR. (B) GC-associated quasipalindrome template switch at position 584-585 within the Kl-URA3 ORF. (C) GC-associated ectopic template switch into Sc-ura3-52 and then back to hmr::Kl-URA3 in which a frameshift is introduced due to misalignment at a shared 5-bp sequence. Underlined sequences indicate Kl-URA3 microhomology (nucleotides 423 to 427).

Table 1

Spontaneous and GC-dependent mutation rates. Relative changes from the corresponding WT rate are in parentheses. *, P < 0.05 (t-test) versus the corresponding WT rate; **, P < 0.01; ***, P < 0.001; ±, SEM; n/a, not applicable.

View this table:

Sequences of independent Ura GC events (Fig. 2 and Table 2) reveal that both synchronous and asynchronous wild-type (WT) cells produce mainly single base-pair substitutions (BPSs) with a significantly larger (P = 0.006) fraction being –1 frameshifts; these data were combined for subsequent statistical comparisons. The proportions of BPSs, –1 deletions, +1 insertions, and complex mutations differ significantly (P = 0.015) from our analysis of spontaneous Kl-URA3 mutations, as well as from a recent study on spontaneous mutations of Saccharomyces cerevisiae URA3 (Sc-URA3) (9) (Table 2). Nearly all GC-associated –1 frameshifts occur in homonucleotide runs (≥2 bases), compared with about half among spontaneous events (P = 0.04) (9).

Fig. 2

GC-associated Kl-URA3 mutant spectrum. Single letters above the Kl-URA3 ORF indicate BPSs. Black boxes with white text indicate nonsense mutations. Insertions and deletions are shown below the ORF; insertions are green and deletions are red. A large deletion occurring between direct repeats is indicated with repeats underlined in red. Blue lines below the ORF mark two ectopic events where Kl-URA3 sequences were replaced by Sc-ura3-52 sequences; double blue lines depict microhomology junctions. The template switch marked as “(2)” contained two independent jumps to Sc-ura3-52, one of which contained a frameshift. Quasipalindrome-mediated template switch mutations are within blue boxes (Fig. 1C and fig. S3 depict these mechanisms). Recurrent mutations are indicated by an X followed by the number of times that mutation was observed.

Table 2

Spontaneous and GC-dependent spectra for K. lactis URA3. For 1-bp deletions, 1-bp insertions, BPSs, and complex mutations, numbers in parentheses indicate percentage of total mutations represented by each respective mutation. –1 deletions and +1 insertions are divided into events occurring in either homonucleotide runs ≥2 nucelotides (nt) or nonrepeating sequences. spont., spontaneous; comb., combined data from asynchronous and G1 cells.

View this table:

Ten out of 12 GC-associated complex mutations can be explained by template switches occurring in the context of quasipalindromes (Fig. 1, A and B, and Fig. 2); similar mutations have been observed from Escherichia coli to human p53 tumors (1012). Two other complex GC events resulted from interchromosomal template switches between the hmr::Kl-URA3 donor, on chromosome 3, and the endogenous Sc-ura3-52 allele, on chromosome 5, which share only 73% identity. In these events the nascent DNA strand must prematurely dissociate from its donor template, hmr::Kl-URA3, invade a new donor, Sc-ura3-52, and eventually come back to hmr::Kl-URA3 to complete MAT switching (Fig. 1A). Homology at the junctions within these Kl-Sc-Kl-ura3 chimeras ranges from 2 to 11 base pairs (bp). Of 455 GC-dependent Ura mutants sequenced in WT and several mutant backgrounds, 10 such interchromosomal template switches and 27 presumptive quasipalindrome events were identified (Table 2 and table S1). No such events were detected in spontaneous mat::Kl-URA3 mutations (Table 2). All 10 ectopic template switches used a shared 5-bp microhomology that misaligns during strand invasion and introduces a frameshift, –1 if used as the entry point or +1 if used as the exit point (Fig. 1C); all DNA synthesis after the template switch was error free. Without this misalignment, these events would apparently result in Ura+ chimeric Kl-Sc-Kl-ura3 GC products and avoid detection. To generate Ura outcomes not requiring the 5-bp misaligning sequence, we inserted two 4-bp sequences into Sc-URA3 such that in-frame template switches would generate Ura outcomes (fig. S4A). This modification increased interchromosomal template switch rates by a factor of about 3. Junctions contained 2 to 17 bp microhomologies, and only one of six used the 5-bp misaligning sequence (fig. S4B). We conclude that GC is considerably less processive than normal DNA replication, even though GC requires the proliferating cell nuclear antigen clamp that should anchor DNA polymerases to the template (13).

These ectopic template switches are reminiscent of spontaneous segmental duplications in yeast and copy number variation in human cancers, which are proposed to occur through microhomology-mediated break-induced replication (BIR) (14, 15) and in yeast require the nonessential Polδ subunit, Pol32 (16). GC-associated mutation rates and mutant spectra do not change in the absence of Pol32 (Table 1, table S1, and fig. S2F). Moreover, ectopic template jumps proved to be Pol32-independent (fig. S2C); thus, we suggest that they arise by a novel microhomology-mediated SDSA pathway.

Mismatch repair (MMR) effectively repairs mismatches in heteroduplex DNA formed during MAT-switching strand invasion (17). However, MMR plays a minor role in discouraging mutations arising during repair DNA synthesis (Table 1). Moreover, mutant spectra of mlh1Δ and msh6Δ strains were not different from WT (figs. S2B and S2C and table S1). Thus, either the extending DNA strand does not recruit MMR proteins during GC or MMR cannot distinguish old from new strands, and, hence, mutations are as likely to be introduced as removed.

Error-prone Rev3 (DNA Polζ) appears to play a central role in generating BPSs but not frameshift mutations in presumably single-stranded DNA adjacent to a repaired DSB (46). We investigated the role for REV3 where both DNA strands are newly synthesized. REV3 does not significantly affect recombination efficiency or the GC mutation rate or spectrum (Table 1, table S1 and fig. S1D). We suggest that Rev3 plays a nonessential role in filling-in single-strand gaps adjacent to the repaired locus but not within the gap-repair region. We also created a rev3Δ rad30Δ double mutant, lacking both Polζ and the error-prone polymerase Polη (18). MAT-switching efficiency and mutant spectra were not affected, but there was a modest, statistically significant increase in the GC mutation rate (Table 1 and fig. S1E), suggesting that Polη plays an error-free role during GC. Further evidence that translesion synthesis (TLS) polymerases do not greatly affect GCs is illustrated by the observation that GC mutation rates are not changed by pol30-K127,164R, an allele that prevents posttranslational modification by ubiquitylation or the addition of the small ubiquitin-like modifier, SUMO, modifications that recruit TLS polymerases (Table 1).

Both Polδ and Polε DNA polymerases are important for HR (13, 16, 1923). To test for preferential usage during GC, we examined proofreading deficient, error-prone alleles, pol3-01 and pol2-4, respectively. The GC mutation rate for pol2-4 is 4.3 times as high as WT (P = 0.004), indicating that Polε plays an important replicative role during GC (Table 1). The spontaneous and GC mutation spectra for pol2-4 were significantly different from WT with a large increase in +1 frameshifts (Table 2 and Fig. S1, G and H).

The GC-dependent mutation rate for pol3-01 was only 2.2 times as high as WT (P > 0.05), suggesting that Polδ plays a less substantial role during GC than previously suggested (22); however, the Ura mutation spectrum revealed that Polδ plays a major role. Most striking, pol3-01 eliminated all –1 frameshifts and complex events (P < 0.0001) (Table 2 and fig. S1I), which also occurred among spontaneous mutations (P < 0.05) (Table 2 and fig. S1J). The lack of –1 frameshifts and other template switches may be explained by the fact that a Polδ proofreading-defective protein is actually more processive than WT (24). Much of the increased spontaneous mutation rate in pol3-01 appears to be a consequence of activating a Dun1 kinase-dependent S-phase or DNA damage checkpoint (25). It is possible that the GC-associated effects with pol3-01 could be the result of activating a Dun1-dependent checkpoint; however, we find that deleting Dun1 alone has no effect on the rate of spontaneous or GC mutation rates or mutant spectra (Table 1, table S1, and fig. S2K).

We conclude that Polε and Polδ are both involved in DNA synthesis during GC, with no apparent preference for using one polymerase over the other. Whether first- and second-strand syntheses are carried out by different DNA polymerases, we cannot ascertain. The presence of frequent template switches suggests that DNA synthesis during GC has decreased processivity compared with S-phase replication. During a nonprocessive GC where there is premature dissociation of the nascent DNA strand from the template, followed by reinitiation, there may be an opportunity to switch between polymerases.

Although GC is generally thought to be the most error-free pathway of DSB repair, it exhibits a significantly elevated mutation rate. Weinberg (26) argues that mammalian cells must suffer ≥6 mutations before transitioning to a fully cancerous state. How cells experience such high mutation levels during limited life spans remains unknown, except when MMR genes or proofreading activity of replicative polymerases are mutated (2729). The observation that activated oncogenes promote dramatic increases in DSB formation (2) suggests that some mutations required for carcinogenesis may arise by increased recombination-associated mutation rates.

Supporting Online Material

Materials and Methods

Figs. S1 to S4

Table S1


References and Notes

  1. Kl-URA3 (GenBank: Y00454) was a gift from O. Zill and J. Rine. pol30-K127,164R was a gift from H. Ulrich. NIH Genetics Training Grant GM007122 supported W.M.H. NIH grant GM20056 to J.E.H. supported the research.

Stay Connected to Science

Navigate This Article