Hypermutation signature reveals a slippage and realignment ...Hypermutation signature reveals a...

6
Hypermutation signature reveals a slippage and realignment model of translesion synthesis by Rev3 polymerase in cisplatin-treated yeast Romulo Segovia a , Yaoqing Shen b , Scott A. Lujan c , Steven J. M. Jones b,d , and Peter C. Stirling a,d,1 a Terry Fox Laboratory, BC Cancer Agency, Vancouver, BC, Canada V5Z1L3; b Michael Smith Genome Sciences Centre, Vancouver, BC, Canada V5Z4S6; c Genome Integrity and Structural Biology Laboratory, National Institute of Environmental Health Sciences, National Institutes of Health, Research Triangle Park, NC 27709; and d Department of Medical Genetics, University of British Columbia, Vancouver, BC, Canada V6T1Z3 Edited by Rodney Rothstein, Columbia University Medical Center, New York, NY, and approved February 3, 2017 (received for review November 8, 2016) Genegene or genedrug interactions are typically quantified using fitness as a readout because the data are continuous and easily measured in high throughput. However, to what extent fitness cap- tures the range of other phenotypes that show synergistic effects is usually unknown. Using Saccharomyces cerevisiae and focusing on a matrix of DNA repair mutants and genotoxic drugs, we quantify 76 genedrug interactions based on both mutation rate and fitness and find that these parameters are not connected. Independent of fit- ness defects, we identified six cases of synthetic hypermutation, where the combined effect of the drug and mutant on mutation rate was greater than predicted. One example occurred when yeast lacking RAD1 were exposed to cisplatin, and we characterized this interaction using whole-genome sequencing. Our sequencing re- sults indicate mutagenesis by cisplatin in rad1Δ cells appeared to depend almost entirely on interstrand cross-links at GpCpN motifs. Interestingly, our data suggest that the following base on the tem- plate strand dictates the addition of the mutated base. This result differs from cisplatin mutation signatures in XPF-deficient Caeno- rhabditis elegans and supports a model in which translesion synthe- sis polymerases perform a slippage and realignment extension across from the damaged base. Accordingly, DNA polymerase ζ ac- tivity was essential for mutagenesis in cisplatin-treated rad1Δ cells. Together these data reveal the potential to gain new mechanistic insights from nonfitness measures of genedrug interactions and extend the use of mutation accumulation and whole-genome se- quencing analysis to define DNA repair mechanisms. mutator | mutation signature | translesion synthesis | genedrug interaction | DNA repair G enome maintenance pathways suppress the accumulation of mutations derived from chemical lesions or mismatches in DNA that arise during normal metabolic processes. Despite thou- sands of potentially mutagenic lesions occurring per cell per day, mitotic cell division exhibits extremely low rates of mutation (10 8 10 10 per bp per generation) under normal conditions (1). Cells with defects in DNA repair pathways are generally more permissive for mutation accumulation (MA), and this phenomenon likely underlies the predisposition of individuals inheriting cellular DNA repair de- fects to cancer formation (2). Increasing the rate of mutations in a cell population makes it more likely that the necessary mutations in oncogenes and tumor-suppressor genes will arise in a given lineage, leading to cellular transformation and proliferation (3). Although DNA repair defects can predispose to cancer for- mation, when acquired somatically they may serve as an Achilles Heel that can be exploited for cancer therapy. This apparent di- chotomy is because chemotherapeutic chemicals often work by damaging DNA, stalling DNA replication, or disrupting mitosis, all of which could be potentially deleterious to a cell with an ac- quired DNA repair deficiency (2). Several well-documented ex- amples of this type of genedrug interaction exist for the BRCA1 and 2 genes, where their mutational status can predict sensitivity of tumors to cisplatin and its derivatives (4). Indeed there are now multiple efforts underway to inhibit additional DNA repair pro- teins themselves to further sensitize cancers to killing with geno- toxic agents or to overwhelm tumor cells with already debilitated DNA repair capacity (2). In cancer, as in model systems, there are typically surviving cells after a genotoxic insult. These cells bear a signature of mutations associated with the genotoxin they survived (5). Mutation signature analysis of tumor genomes has been refined in the past 5 years, and a set of 30 canonical mutation signatures is now maintained in the Catalogue of Somatic Mutations in Cancer database (5, 6). In some cases, the etiology of mutation patterns is strongly linked to specific genetic mutations or environmental exposures, whereas in other cases the etiology remains unknown. Studies in model organisms have sought to dissect which aspects of a mutation signature are due to specific deficiencies in genome maintenance factors or to specific chemical treatments (79). Indeed, the largest such study to date in Caenorhabditis elegans characterized both a panel of mutant strains and the effects of Aflatoxin B1, mechlorethamine, and cisplatin (9). The intention of genotoxin treatments clinically is to kill cells rather than mutagenize them. Model organism studies have also provided a means to map genetic networks underlying genotoxin sensitivity. The systematic identification of synthetic lethal in- teractions or chemicalgenetic interactions has been led by studies in budding yeast, Saccharomyces cerevisiae. Indeed, a full Significance Cancer cells often have defects in DNA repair and are killed effectively by drugs that damage DNA. However, surviving cells can acquire additional mutations after treatment with these genotoxic chemicals. Here we apply a simple model system to reveal synergy between specific DNA repair muta- tions and genotoxic drugs that occurs independently of fitness defects. Moreover, by analyzing the entire genome of a mutagenized cell population, we identify a signature of mu- tations that informs the mechanism of the translesion synthe- sis DNA damage tolerance pathway. Our work establishes a conceptual framework for predicting the mutational burden of cells surviving genotoxin treatment and adds to a growing list of examples supporting the utility of model organism mutation signature analysis for generating mechanistic insights. Author contributions: R.S. and P.C.S. designed research; R.S. performed research; S.A.L. and S.J.M.J. contributed new reagents/analytic tools; R.S., Y.S., S.A.L., S.J.M.J., and P.C.S. analyzed data; and R.S. and P.C.S. wrote the paper. The authors declare no conflict of interest. This article is a PNAS Direct Submission. Data deposition: The sequence reported in this paper has been deposited in the Sequence Read Archive database (accession no. SRP091984). 1 To whom correspondence should be addressed. Email: [email protected]. This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. 1073/pnas.1618555114/-/DCSupplemental. www.pnas.org/cgi/doi/10.1073/pnas.1618555114 PNAS | March 7, 2017 | vol. 114 | no. 10 | 26632668 GENETICS Downloaded by guest on May 24, 2021

Transcript of Hypermutation signature reveals a slippage and realignment ...Hypermutation signature reveals a...

Page 1: Hypermutation signature reveals a slippage and realignment ...Hypermutation signature reveals a slippage and realignment model of translesion synthesis by Rev3 polymerase in cisplatin-treated

Hypermutation signature reveals a slippageand realignment model of translesion synthesis byRev3 polymerase in cisplatin-treated yeastRomulo Segoviaa, Yaoqing Shenb, Scott A. Lujanc, Steven J. M. Jonesb,d, and Peter C. Stirlinga,d,1

aTerry Fox Laboratory, BC Cancer Agency, Vancouver, BC, Canada V5Z1L3; bMichael Smith Genome Sciences Centre, Vancouver, BC, Canada V5Z4S6;cGenome Integrity and Structural Biology Laboratory, National Institute of Environmental Health Sciences, National Institutes of Health, Research TrianglePark, NC 27709; and dDepartment of Medical Genetics, University of British Columbia, Vancouver, BC, Canada V6T1Z3

Edited by Rodney Rothstein, Columbia University Medical Center, New York, NY, and approved February 3, 2017 (received for review November 8, 2016)

Gene–gene or gene–drug interactions are typically quantified usingfitness as a readout because the data are continuous and easilymeasured in high throughput. However, to what extent fitness cap-tures the range of other phenotypes that show synergistic effects isusually unknown. Using Saccharomyces cerevisiae and focusing on amatrix of DNA repair mutants and genotoxic drugs, we quantify 76gene–drug interactions based on bothmutation rate and fitness andfind that these parameters are not connected. Independent of fit-ness defects, we identified six cases of synthetic hypermutation,where the combined effect of the drug and mutant on mutationrate was greater than predicted. One example occurred when yeastlacking RAD1 were exposed to cisplatin, and we characterized thisinteraction using whole-genome sequencing. Our sequencing re-sults indicate mutagenesis by cisplatin in rad1Δ cells appeared todepend almost entirely on interstrand cross-links at GpCpN motifs.Interestingly, our data suggest that the following base on the tem-plate strand dictates the addition of the mutated base. This resultdiffers from cisplatin mutation signatures in XPF-deficient Caeno-rhabditis elegans and supports a model in which translesion synthe-sis polymerases perform a slippage and realignment extensionacross from the damaged base. Accordingly, DNA polymerase ζ ac-tivity was essential for mutagenesis in cisplatin-treated rad1Δ cells.Together these data reveal the potential to gain new mechanisticinsights from nonfitness measures of gene–drug interactions andextend the use of mutation accumulation and whole-genome se-quencing analysis to define DNA repair mechanisms.

mutator | mutation signature | translesion synthesis | gene–druginteraction | DNA repair

Genome maintenance pathways suppress the accumulation ofmutations derived from chemical lesions or mismatches in

DNA that arise during normal metabolic processes. Despite thou-sands of potentially mutagenic lesions occurring per cell per day,mitotic cell division exhibits extremely low rates of mutation (10−8–10−10 per bp per generation) under normal conditions (1). Cells withdefects in DNA repair pathways are generally more permissive formutation accumulation (MA), and this phenomenon likely underliesthe predisposition of individuals inheriting cellular DNA repair de-fects to cancer formation (2). Increasing the rate of mutations in acell population makes it more likely that the necessary mutations inoncogenes and tumor-suppressor genes will arise in a given lineage,leading to cellular transformation and proliferation (3).Although DNA repair defects can predispose to cancer for-

mation, when acquired somatically they may serve as an AchillesHeel that can be exploited for cancer therapy. This apparent di-chotomy is because chemotherapeutic chemicals often work bydamaging DNA, stalling DNA replication, or disrupting mitosis,all of which could be potentially deleterious to a cell with an ac-quired DNA repair deficiency (2). Several well-documented ex-amples of this type of gene–drug interaction exist for the BRCA1and 2 genes, where their mutational status can predict sensitivityof tumors to cisplatin and its derivatives (4). Indeed there are now

multiple efforts underway to inhibit additional DNA repair pro-teins themselves to further sensitize cancers to killing with geno-toxic agents or to overwhelm tumor cells with already debilitatedDNA repair capacity (2).In cancer, as in model systems, there are typically surviving cells

after a genotoxic insult. These cells bear a signature of mutationsassociated with the genotoxin they survived (5). Mutation signatureanalysis of tumor genomes has been refined in the past 5 years, anda set of 30 canonical mutation signatures is now maintained in theCatalogue of Somatic Mutations in Cancer database (5, 6). In somecases, the etiology of mutation patterns is strongly linked to specificgenetic mutations or environmental exposures, whereas in othercases the etiology remains unknown. Studies in model organismshave sought to dissect which aspects of a mutation signature are dueto specific deficiencies in genome maintenance factors or to specificchemical treatments (7–9). Indeed, the largest such study to date inCaenorhabditis elegans characterized both a panel of mutantstrains and the effects of Aflatoxin B1, mechlorethamine, andcisplatin (9).The intention of genotoxin treatments clinically is to kill cells

rather than mutagenize them. Model organism studies have alsoprovided a means to map genetic networks underlying genotoxinsensitivity. The systematic identification of synthetic lethal in-teractions or chemical–genetic interactions has been led bystudies in budding yeast, Saccharomyces cerevisiae. Indeed, a full

Significance

Cancer cells often have defects in DNA repair and are killedeffectively by drugs that damage DNA. However, survivingcells can acquire additional mutations after treatment withthese genotoxic chemicals. Here we apply a simple modelsystem to reveal synergy between specific DNA repair muta-tions and genotoxic drugs that occurs independently of fitnessdefects. Moreover, by analyzing the entire genome of amutagenized cell population, we identify a signature of mu-tations that informs the mechanism of the translesion synthe-sis DNA damage tolerance pathway. Our work establishes aconceptual framework for predicting the mutational burden ofcells surviving genotoxin treatment and adds to a growing listof examples supporting the utility of model organism mutationsignature analysis for generating mechanistic insights.

Author contributions: R.S. and P.C.S. designed research; R.S. performed research; S.A.L.and S.J.M.J. contributed new reagents/analytic tools; R.S., Y.S., S.A.L., S.J.M.J., and P.C.S.analyzed data; and R.S. and P.C.S. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Data deposition: The sequence reported in this paper has been deposited in the SequenceRead Archive database (accession no. SRP091984).1To whom correspondence should be addressed. Email: [email protected].

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1618555114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1618555114 PNAS | March 7, 2017 | vol. 114 | no. 10 | 2663–2668

GEN

ETICS

Dow

nloa

ded

by g

uest

on

May

24,

202

1

Page 2: Hypermutation signature reveals a slippage and realignment ...Hypermutation signature reveals a slippage and realignment model of translesion synthesis by Rev3 polymerase in cisplatin-treated

pairwise gene–gene interaction study is now complete for bothessential and nonessential yeast genes (10). In addition severalthousand small molecules have been profiled for sensitivity andresistance across the yeast knockout (YKO) collections (11).These approaches are being combined to understand the effectsof chemical perturbations on genetic interaction networks andidentify gene–gene synergies in drug sensitivity (12). In each ofthese studies, the primary readout for synergy between genes andchemicals is fitness, as it is quantitative, simple to measure inhigh throughput, and informative. Nevertheless, other quantita-tive phenotypic readouts are possible, and the YKO collectionhas been profiled by numerous biochemical, cytological, andfunctional phenotypes (13).Reasoning that DNA repair deficiencies would result in cell

death, mutagenesis of survivors, or both after a genotoxic insult,we assessed the overlap of fitness and mutagenesis for repre-sentative chemical genotoxins in yeast cells defective for allmajor DNA repair pathways. Quantifying growth and mutationrates showed little overlap between these parameters and furtherrevealed cases of unexpected hypermutation. We predicted thatthere would be a pattern of mutations associated with hyper-mutagenesis and characterized that of rad1Δ and cisplatin bywhole-genome sequencing, uncovering evidence of a translesionsynthesis (TLS) mechanism for yeast DNA polymerase ζ (Rev3).Together these data define gene–drug interactions, underscore amutation signature in yeast, and apply MA and whole-genomesequencing to suggest DNA repair mechanisms.

ResultsA Network of Genotoxin-Induced Fitness Defects and Mutation Rates.To investigate gene–drug interactions, we first established a panelof DNA repair mutants representing the major DNA repairpathways in yeast (Table S1). Haploid yeast bearing gene deletionsimpairing homologous recombination (HR), nonhomologous endjoining (NHEJ), nucleotide-excision repair (NER), base-excisionrepair (BER), mismatch repair (MMR), TLS, Fanconi Anemia-like (FA), and postreplication repair (PRR) pathways were chosento represent loss or deficiencies of each pathway. To account forthe multiple steps of, and routes for, DNA double-strand break(DSB) repair by HR, we also included exo1Δ, rad52Δ, sgs1Δ,mus81Δ, and mph1Δ lines, as these mutants specifically impairend resection, strand exchange, double Holliday junction disso-lution, resolution, or other HR steps, respectively (14). This panelof wild-type (WT) and 12 mutant strains (Fig. 1) was exposed tofive classes of DNA damaging agents, represented by the bi-functional alkylator cisplatin (Cis), the antimetabolite fluorouracil(5FU), the topoisomerase I inhibitor camptothecin (Cpt), thetopoisomerase II inhibitor etoposide (Etp), and the monofunc-tional alkylator methyl methanesulfonate (MMS). We screened 78pairwise gene–drug combinations plus WT for changes in fitnessand, in 76 cases, were also able to measure mutation rates (rad5Δgrowth was impaired to a degree that prevented fluctuationanalysis).Growth was measured over 24 h, and the area under the curve

was calculated and normalized to the untreated WT to measurefitness (Fig. 1A). Overall, we found that Cpt and Etp had nosignificant additional effects on any strain using the concentrationstested, and although 5FU inhibited growth of all lines irrespectiveof the genetic background, it only showed a synergistic interactionwith apn1Δ. We also observed that cisplatin had synergistic gene–drug interactions with rad1Δ, rad5Δ, rad52Δ, rev3Δ, and sgs1Δ andthat MMS had synergistic interactions with rad5Δ and sgs1Δ (P <0.05; Fig. 1A). These interactions are consistent with, for example,the known roles for BER in repair of uracil in DNA or the knowncomplexity of repairing cisplatin interstrand cross-links (ICLs),which involves at least HR, NER, FA, and TLS. Moreover, thisscreen possibly underrepresents chemical sensitivities because very

low concentrations of drugs were used to allow comparison withfluctuation assays below.Mutation rates of the same matrix were quantified at CAN1

using a well-plate fluctuation assay (Fig. 1B and Table S2) (15). Inuntreated cells, the baseline mutation rates matched previouslyreported rates (Table S3) (7, 15–20). Again we observed that Cptand Etp had no major mutator effects regardless of the geneticbackground at the given drug concentrations. On the contrary,cisplatin, 5FU, and MMS increased the mutation rates of specificmutants. When this increase in mutation rate exceeded theexpected effects of multiplying the effect of the repair deficiencyand the effect of the chemical on WT, we defined it as SyntheticHypermutation (SHyp). We chose to apply the multiplicativemodel, both because it performs best across a range of geneticinteractions based on fitness (21) and because it creates a morestringent definition of a SHyp interaction (we compare with anadditive definition in Fig. S1). Applying this metric, we identifiedsix cases of SHyp: mph1Δ-Cis, rad1Δ-Cis, apn1Δ-5FU, mlh1Δ-5FU, exo1Δ-MMS, and mph1Δ-MMS (Fig. 1B and Fig. S1). Ofthese cases, rad1Δ-Cis, apn1Δ-5FU, and mlh1Δ-5FU showed themost severe SHyp phenotype (17.8-, 45.1-, and 33.4-fold increaseover WT, respectively). Importantly, cases of SHyp did not overlapwith fitness defects except for rad1Δ-Cis and apn1Δ-5FU, andconversely many strains with fitness defect exhibited no increase inmutation rate. Indeed, survival analysis of cisplatin sensitivity of

BERMMRHR MMR

FAHR HR

NERFA PRR TLS TLS HR NHEJ

No drugWT ap

n1Δ

exo1Δ

mlh1Δ

mph1Δ

mus81Δ

rad1Δ

rad5Δ

rad30Δ

rad52Δ

rev3Δ

sgs1Δyku80Δ

WT apn1Δ

exo1Δ

mlh1Δ

mph1Δ

mus81Δ

rad1Δ

rad5Δ

rad30Δ

rad52Δ

rev3Δ

sgs1Δyku80Δ

CPT

CIS

ETP

5FU

MMS

No drug

CPT

CIS

ETP

5FU

MMS

B

C

A

*

* ***

* *

* *

** * * *

SSAHR

0

10

20

30

40

50

60

70

WT rad1Δ rad2Δ rad23Δ rad26Δ

CA

N1R

rate

(x10

-7) p

er g

ener

atio

n

Nucleotide Excision Repair mutant

No cisplatin10μM cisplatin

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1

High fitness Low fitness

3.5 21 38 56 73 91 108 126 143 160

Low rate High rate

Fig. 1. Gene–drug interactions measured by fitness and mutation rates.Heat maps illustrating (A) fitness and (B) CAN1 mutation rate relative to WT.Interactions significantly greater than expected (P < 0.05) are indicated *.The transition from yellow to blue indicates greater fitness defects or highermutation rates. The first yellow box is set at the WT rate; any lower rates(i.e., in some rev3Δ- and 5FU-treated conditions) are given the same color.Black boxes for rad5Δ indicate not done. (C) CAN1 mutation rates of otherNER-deficient strains in cisplatin. rad1Δ and rad2Δ exhibit SHyp (P < 0.05).

2664 | www.pnas.org/cgi/doi/10.1073/pnas.1618555114 Segovia et al.

Dow

nloa

ded

by g

uest

on

May

24,

202

1

Page 3: Hypermutation signature reveals a slippage and realignment ...Hypermutation signature reveals a slippage and realignment model of translesion synthesis by Rev3 polymerase in cisplatin-treated

sgs1Δ, mus81Δ, and rad1Δ shows nearly identical viability curvesover cisplatin concentration but different mutational outcomes(Fig. S2). Together our data show that decreased fitness and in-creased mutability by genotoxins in DNA repair-deficient cells arenot perfectly linked. This observation is likely both an issue of doseand DNA repair mechanisms wherein some gene–drug combina-tions are more likely to elicit toxic or irreparable intermediates.

Cisplatin Hypermutation in Specific NER-Deficient Contexts. One ofthe outcomes of identifying unexpected increases in mutation rateis the potential to use these observations to uncover previouslyunrecognized modes of mutagenesis. The strongest SHyp inter-actions, apn1Δ-5FU and mlh1Δ-5FU, likely reflect known links ofBER and MMR to removing fluorouracils paired with A or Gfrom DNA (22). Whether a specific mutagenesis mode and sig-nature was responsible for SHyp in rad1Δ-Cisplatin was less clear,and we decided to pursue analysis of this interaction further. Rad1is a nuclease involved in cleaving DNA flanking bulky lesionsduring NER and ICL repair. Recognition of these lesions canoccur during transcription or globally due to the action of up-stream NER components that recognize distorted DNA helices(23). To determine which aspects of NER exhibit SHyp with cis-platin, we tested mutation rates in three other NER mutants—rad2Δ, rad26Δ, and rad23Δ. Deletion of RAD2, which, like Rad1,encodes a nuclease required for DNA incision and is related tohuman XPG, also exhibited SHyp with cisplatin (Fig. 1C). How-ever, deletion of RAD26, which selectively impairs transcription-coupled (TC) NER, or RAD23, which partially impairs global andTC-NER but allows efficient DNA incision, had no SHyp phe-notype with cisplatin (Fig. 1C). Because Rad1 acts first and Rad2is required in a structural role to position Rad1 (23), loss of eitherwould block incision; consequently, alternative mechanisms ofunhooking of the cisplatin damage must be error prone.

MA and Whole-Genome Sequencing. It is possible that CAN1 showedvariable mutation rates in repair-deficient strains due to a locus-specific effect. To explore the genomic spectrum of mutations andidentify any evidence of such bias, we pulsed diploid WT, rad1Δ,sgs1Δ, and mus81Δ with a higher dose of cisplatin and sequencedthe whole genomes of 12 independent survivors for each strain (Fig.2A). Consistent with the CAN1 fluctuation analysis data, the rad1Δmutant strain acquired significantly more mutations than WT or theother mutants (Fig. 2B). As expected, based on the preference ofcisplatin for cross-linking guanine residues, the most common mu-tation types were single-nucleotide variants (SNVs) at C:G basepairs. Moreover, indels seen in rad1Δ, which were primarily –1deletions, occurred at C:G base pairs, and rare dinucleotide sub-stitutions were also evident at motifs containing C:G base pairs (Fig.2 C and D). To ensure that the rate and type of mutations were dueto cisplatin and not simply rad1Δ, we passaged diploid WT andrad1Δ cells through 40 single-cell bottlenecks and sequenced 12independent whole genomes after this 1,000 generation MA ex-periment (Fig. 2A). The mutation rate and type evident in therad1Δ MA lines were similar to the CAN1 rate in untreated rad1Δcells and showed proportionally more mutations at T:A base pairsand no dinucleotide substitutions (Fig. S3). No correlations betweenrad1Δ-Cis mutations with transcription, replication timing, orleading-strand character were seen, although a weakly significantenrichment (P = 0.0103) was seen between nucleosome-boundDNA and mutations (24). Thus, the specific combination of RAD1deletion and cisplatin treatment engages an error-prone repairprocess that causes a global increase in specific types of mutations.

Cisplatin-rad1Δ Mutation Signatures Suggest Templated BaseSubstitutions. The flanking sequences around mutations can de-fine a signature that is associated with a particular genetic orchemical perturbation. Indeed, previous work in C. elegans has ex-amined cisplatin treatment of several mutants including XPF-1 (i.e.,

C. elegans RAD1)-deficient animals (9). We extracted flanking se-quences for the predominant C > N mutations in cisplatin-treatedgenomes and performed pLogo and enrichment analysis (Fig. 3Aand Fig. S4A) (25). This analysis revealed that GpCpT motifs werefavored when all C > N mutations were considered. Because cis-platin is expected to preferentially cross-link guanine residues, thissignature suggests that ICLs are the mutagenic lesion in yeast RAD1mutants. The observed GpCpT motif differs from C. elegans, whereCpCpT motifs were preferentially mutated in xpf-1 worms (9).Because intrastrand cross-links are much more common than ICLs(26), our data suggest either that rad1Δ yeast are able to repairthese in an error-free manner, even when lacking a key NER pro-tein, or that they die and are removed from the experiment. FurtherpLogo analysis with fixed positions at GC showed that, for eachmutation type, adenine was significantly enriched at the –2 position(i.e., ApGpCpT) (Fig. 3A, Center and Fig. S4A). When we analyzedthe 3′ position, a different picture emerged: The mutation typevaried with the +1 base, such that C > T mutations preferentiallyoccurred in GpCpT motifs, C > Amutations in GpCpA motifs, andC > G mutations in GpCpG motifs (Fig. 3B and Fig. S4B). Thus,the base that is erroneously inserted at the damaged site is selectedby base-pairing with the 5′ base on the opposite template strand.Interestingly, mutations at GpCpC, which would be error free ifusing the 5′ template-strand base for synthesis, occurred at only 13of 327 C > N mutations, much less frequently than for GpCpT,GpCpA, and GpCpG. There was also no evidence that indels

0

40

80

120

160

C>A C>G C>T T>A T>C T>G

Num

ber o

f SN

Vs

Type of base substitution

WTrad1Δ/Δsgs1Δ/Δmus81Δ/Δ

B C

02468

101214

CpACpGCpTGpCGpTDin

ucle

otid

e su

bstit

utio

ns

Reference sequence

MutationsTpTTpGTpCTpAGpCCpGApTApGApCApA

D

Repeat 40x

G1000Mutation accumulation

Genotoxictreatment

12x

Single clone expansion

Single-cell bottlenecks

100μMcisplatin/2hrs

wt, rad1Δ/Δ,sgs1Δ/Δ,

mus81Δ/Δ

...

wt, rad1Δ/Δ

WGS

12x

A

0

5

10

15

20

25

30

rad1Δ/Δ (cisplatin) rad1Δ/Δ (MA) WT (MA)

Num

ber o

f ind

els

Indel pattern

InsertionsDeletions

Fig. 2. MA and whole-genome sequencing. (A) Experimental design ofdrug treatment and 1,000 generation (G1000) approaches for the indicatedstrains (WGS, whole-genome sequencing). (B) Summary of WGS results ofSNVs in cisplatin-treated WT, rad1Δ, sgs1Δ, and mus81Δ homozygous dip-loid strains. SNV number and type are summarized in the bar graphs color-coded as indicated. (C) Dinucleotide substitutions in rad1Δ-Cis genomes. Theparental dinucleotide sequences are on the x axis, and the mutated se-quence is represented by the color of the bar. (D) Pattern of small indels inrad1Δ-Cis and MA genomes. Indel size and sequence are on the x axis, andindel types are color-coded.

Segovia et al. PNAS | March 7, 2017 | vol. 114 | no. 10 | 2665

GEN

ETICS

Dow

nloa

ded

by g

uest

on

May

24,

202

1

Page 4: Hypermutation signature reveals a slippage and realignment ...Hypermutation signature reveals a slippage and realignment model of translesion synthesis by Rev3 polymerase in cisplatin-treated

occurred at GpCpC motifs, as only 2 of 12 deletions at GpCpNoccurred at GpCpC (within Dataset S1). These observations furthersupport a model of templated TLS as discussed below.

DNA Polymerase ζ Drives Rad1Δ-Cisplatin SHyp. Our mutation sig-nature analysis suggests error-prone DNA replication is likely caus-ing mutations in rad1Δ cells treated with cisplatin. To identify therelevant polymerase, we deleted all enzymatic TLS components in-dividually in the rad1Δ background. Loss of RAD30, which slightlypositively affected fitness compared with the rad1Δ single mutant,showed no effect on rad1Δ hypermutation (Fig. 4 A and B), howeveradditional loss of REV1 or REV3 led to inviability when treated with10 μM cisplatin (Fig. 4A). This observation is consistent with theDNA damage tolerance role of REV3 and the idea that TLS is bothsupporting viability in RAD1-deficient cells and causing a hyper-mutation signature, as was previously reported for complex frame-shifts at spontaneous DNA lesions (27). A lower dose of cisplatinstill revealed SHyp with rad1Δ but showed complete suppression ofSHyp in rad1Δrev3Δ and rad1Δrev1Δ cells (Fig. 4B), indicating thatPolζ TLS was responsible for the mutations. To determine if en-dogenous DNA lesions could promote 5′-templated mutations (i.e.,+1 3′ on the strand being replicated by the 5′ base on the oppositetemplate strand), we scored a published dataset of URA3 mutationssequenced in unperturbed NER-deficient (rad14Δ) cells with or

without TLS polymerase deletions (28). Remarkably, downstreamtemplating was observed at 43% of SNVs, deletion of RAD30 had noeffect on this frequency, whereas REV3 deletion reduced the fre-quency of SNVs matching the downstream base to 29% (Fig. 4C).Furthermore, analysis of another published Rev3 mutation setbrought on by a mutant replisome (pol3-Y708A) (29) also showed apattern of templated insertions [i.e., in REV3 37% of SNVs matchedthe 3′ base (n = 126), whereas in rev3Δ 22% of SNVs matched the 3′base (n = 85), P = 0.0431 Fisher’s exact test]. Together these datalead us to propose a model in which slippage and realignment of theDNA template by Rev3 led to the observed mutation signature andmay be generalizable to both endogenous lesions and possible alsowhen Rev3 is called to replicate undamaged DNA (Fig. 4D andDiscussion).

DiscussionSHyp as an Alternative Measure of Phenotypic Enhancement inGenome Maintenance. As fitness datasets on gene–gene and gene–drug interactions accumulate (10, 11), measurements of trigenicinteractions (30) or the ways in which chemicals shift genetic-interaction networks (12) are enhancing the wiring diagram of thecell. Damage repair and mutation is key to the chemotherapeuticaction of alkylating agents, and changes in DNA damage repair canhave profound effects on therapeutic efficacy. Although there isprior evidence that ultrahigh mutation rates can drive severe fitnessconsequences, even lethality (31), the overlap of these phenotypesbroadly is not known. By systematically ablating the major DNArepair pathways in yeast and measuring the effects of diverse gen-otoxic chemicals, our data suggest that fitness and chemically in-duced hypermutation are not necessarily linked.SHyp interactions between 5FU and mlh1Δ and apn1Δ are

largely consistent with the role of MMR and BER, respectively,

A

B

02468

10

GCGCG

AGCG

Fold

enr

ichm

ent

C>G

01234567

GCGCT

AGCT

Fold

enr

ichm

ent

C>T

02468

1012

GCGCT

AGCT

Fold

enr

ichm

ent

C>N

012345678

GCGCA

AGCA

Fold

enr

ichm

ent

C>A

n = 155

n = 39

n = 133

n = 327

+3.49

-3.490

unde

rov

er

+3.51

-3.510

unde

rov

er

-3 -2 -1 0 +1 +2 +3 -3 -2 -1 0 +1 +2 +3

-3 -2 -1 0 +1 +2 +3-3 -2 -1 0 +1 +2 +3

-3 -2 -1 0 +1 +2 +3 -3 -2 -1 0 +1 +2 +3

-3 -2 -1 0 +1 +2 +3 -3 -2 -1 0 +1 +2 +3

+3.49

-3.49

0

unde

rov

er

+3.51

-3.510

unde

rov

er

+3.49

-3.49

0

unde

rov

er

+3.51

-3.51

0

unde

rov

er

+3.49

-3.490

unde

rov

er

+3.51-3.510

unde

rov

er

Fig. 3. Mutation signature analysis induced by cisplatin in rad1Δ diploidyeast. (A) Flanking nucleotides for C > N substitutions are presented aspLogo plots and fold enrichment. The mutated C is centered in each plotwith fixed positions at C (Left) and GC (Center). The red line indicates sig-nificant enrichment at P < 0.05, and the font size indicates the magnitude ofenrichment. Overrepresented motifs (over) are above, whereas un-derrepresented (under) are below. The increasing fold enrichment for 2–3-and 4-nucleotide motifs containing a mutated C is shown (Right). All analysiswas performed with a set of 41 mers and cropped for visualization (Materialsand Methods). (B) Flanking analysis for C > A, C > G, and C > T mutations aspresented in A. Note the +1 base changing with the mutation type.

rad14Δ rad14Δrad30Δ

rad14Δrev3Δ

A C

D

0

10

20

30

40

50

60

70

80

WT rad1Δ rad1Δ rad30Δ

WT rad1Δ rad1Δ rev1Δ

rad1Δ rev3Δ

CAN1

mut

atio

n ra

te*

No cisplatin

non-canonical unhooking

5μM cisplatin10μM cisplatin

0.000

0.009

0.002

0.030

0

0.2

0.4

0.6

0.8

1

1.2

WT rad1Δ rad1Δ rad30Δ

rad1Δ rev1Δ

rad1Δ rev3Δ

Rel

ativ

e fit

ness

com

pare

d to

WT

Expected fitness Observed fitness in 10μM cisplatin

GCTCGA

?

5′

5′

5′3′

3′

3′

5′3′ C A

GG

5′5′3′

3′G TC A

GG

T

Polζ

realignment

?

5′5′3′

3′G TC A

GG

·

Polζ

slippage

0%10%20%30%40%50%

% P

oten

tial t

empl

ated

SN

Vs

p=0.033

n.s.

B

Fig. 4. Role of TLS in hypermutation of cisplatin-treated rad1Δ cells.(A) Fitness measurements and (B) CAN1 mutation rates (* × 10−7 per genera-tion) in rad1Δ cells lacking the indicated TLS polymerase subunit. (C) Frequencyof SNVs identical to 3′ base in URA3 from the indicated strains from ref. 28.(D) Model for slippage and realignment of DNA polymerase ζ (Rev3) at a cis-platin-damaged base. Noncanonical and presumably not optimal unhookingof the ICL creates a substrate for error-prone TLS that promotes slippage ofPolζ at the lesion and subsequent realignment upon lesion bypass leading to atemplated SNV using the 5′ base on the template strand to insert the SNV.

2666 | www.pnas.org/cgi/doi/10.1073/pnas.1618555114 Segovia et al.

Dow

nloa

ded

by g

uest

on

May

24,

202

1

Page 5: Hypermutation signature reveals a slippage and realignment ...Hypermutation signature reveals a slippage and realignment model of translesion synthesis by Rev3 polymerase in cisplatin-treated

in removing uracil and 5-fluorouracil from DNA (22). Althoughwhether these are separate roles in repairing different lesions ordue to potential cooperation of MMR and BER is unclear (22,32). Regardless, these data show that 5FU can overwhelm therepair capacity of MMR or BER-deficient cells without addi-tionally affecting the fitness of MMR impaired cells. These dataare consistent with evidence that 5FU has multiple modes ofaction interfering with both DNA and RNAmetabolism (22). Onthe other hand, although both NER and HR are required forrepair of cisplatin ICLs and mutants in these pathways had fit-ness defects, rad1Δ but not sgs1Δ or other major HR mutantsshowed SHyp with cisplatin. This difference suggests that al-though survival in rad1Δ cells is supported by an error-pronemechanism, that same pathway is insufficient to support sgs1Δcells, which either die, likely due to toxic recombination inter-mediates, or repair cisplatin damage in an error-free way.

A Mutation Signature of TLS Polymerase Slippage and Realignment.Whole genome sequencing analysis of mutagenized model organismgenomes is a powerful way to link mutation signatures to specificcausal phenomena (33). This approach has been used to shed newlight on DNA repair mechanisms (7), the effects of genotoxicchemicals (9, 34), and the role of cytidine deaminase enzymes incancer (25). Characterization of our sequencing data revealed that inrad1Δ yeast ICLs were the predominant source of mutations, asmost variants were found at GpC motifs. This signature differs fromthe cisplatin mutation signature of both C. elegans and humans,where intrastrand cross-links seem to drive mutagenesis (i.e., muta-tions at CpC motifs) (9, 35). There are several possible reasons forthis difference, including the complexity of the FA pathway for ICLrepair in humans that is only partly represented in yeast or the highdivergence of REV3 itself (i.e., human REV3L protein is >twice aslong and exhibits only 25% similarity to yeast Rev3). Although Rev1plays an important structural role, we predict Rev3 is catalyzing themutations that we observed in our study, as the catalytic role of Rev1to insert C opposite the damaged guanine would be error-free andthus not detectable in our assay. However, another possibility is thatintrastrand cross-links are not repaired in rad1Δ cells and lead to celldeath, effectively masking the mutations seen at these sites in othersystems. We also found evidence of Rev3 slippage and realignmentin mutations from untreated cells, suggesting that endogenous,noncisplatin-induced lesions can promote the same mechanism (28).Our mutation signature analysis revealed not only a different

type of mutagenic lesion in yeast but also a mechanism involvingPolζ, Rev3, which indicates a slippage and realignment eventduring TLS. Previous work in vitro has identified this behavior inanother TLS polymerase, human DNA polκ (36). Indeed, thismechanism may also be similar to that proposed for Dpo4 basedon structural and in vitro studies (37). The concept of slippage asa source of replication errors has been around for many years,although never, to our knowledge, applied to slippage and re-alignment in B-family DNA polymerase (38). We propose that(Fig. 4D), at least in the specific context of cisplatin adducts, theguanine lesion is bypassed by Rev3 in a manner that uses Wat-son–Crick base-pairing with the undamaged template strandbase. This phenomenon could lead to an indel, but more oftenthe newly inserted base shifts to mispair with the damaged Gresidue and is extended with the correct base at the 3′ positionon the newly synthesized strand. Whether suboptimal DNA in-cision by compensatory nucleases favors this misincorporationevent is an active area of research. Indeed, the nature of DNAincision in the absence of Rad1 is unclear, although structure-specific nucleases such as Slx4, regulated by Slx1, Mus81, and itspartner Mms4, and Yen1 have activities that could in principlecomplement Rad1 loss at some structures (39). Alternatively,new data in Xenopus extracts implicate the N-glycosylase ac-tivity of NEIL3, an upstream BER factor, in unhooking psor-alen and other ICLs (40). Therefore, it is also possible that

yeast N-glycosylases, such as Ntg1 and Ntg2, could act on cisplatincross-links in the absence of the preferred Rad1 pathway andcreate a unique chemical moiety that drives slippage-realignmentTLS by Polζ in vivo.

SHyp and Cancer. The framework created here sought to establishthe gene–drug relationships driving hypermutation in cells sur-viving genotoxin treatment beginning with canonical DNA repairpathways and drugs of clinical relevance. There remains a largespace to be explored; in yeast, there are many hundreds of geneswhose loss of function causes genome instability and hundredsof others whose increased dosage causes genome instability (8,41, 42). The ways in which these will combine with each otherand with various chemicals to permit mutagenesis remainincomplete.Hypermutation of surviving cancer cells after genotoxic che-

motherapy could be viewed as negative if it permits the acquisitionof chemoresistant mutations. However, current evidence suggeststhat resistance mutations are often preexisting in tumor cellpopulations, and their emergence is a clonal selection processrather than chemotherapy-induced mutations (43). Indeed, evenfor therapy-related leukemias, p53 mutations driving leukemo-genesis were recently found to be preexisting when chemotherapyoccurred (44). Thus, SHyp may not be a major concern for ac-quiring chemoresistant mutations when treating repair-deficientcancers with genotoxic chemotherapies. Although chemotherapyhas been shown to minimally impact chemoresistant mutations,the DNA-damaging nature of chemotherapeutic agents can in-crease the overall mutational load (35). We propose that thishypermutation of surviving cells could ultimately be beneficial incontexts where an immune response to tumor neo-antigens isdesired, as is the case for therapies targeting immune checkpoints,where higher mutational loads are known to improve outcomes(45). Understanding more about the interactions of genotoxicchemotherapies with genome maintenance defects in cancer couldultimately support interventions where it is desirable to increasethe mutational load or reshape the mutational profile of cancercells following frontline therapy.

Materials and MethodsYeast Strains and Growth Curve Analysis. Yeast strains are listed in Table S4.Yeast were grown on SD minimal media plus histidine, uracil, leucine, lysine,and methionine + 2% (wt/vol) dextrose (SD+5) or YPD at 30 °C. Growthcurves were conducted as described (41). The observed fitness, by area underthe curve, was compared with the expected fitness value using a t test (41).Unless otherwise indicated, drug concentrations used were 1 μM Cpt, 10 μMcisplatin, 200 μM Etp, 10 μM 5FU, and 0.0005% MMS.

Fluctuation Analysis and SHyp Analysis. Fluctuation analysis at CAN1 wasperformed with ≥18 replicates per condition as described (15), except thatcells were grown to saturation in the presence of drugs in SD+5 for 48 h at30 °C. Cell numbers were determined using a TC-20 cell counter (BioRad),and CANR colony counts were converted to mutation rates using the Ma–Sandri–Sarkar maximum-likelihood estimator calculator (46). To calculateexpected mutation rates, we multiplied or added (Fig. S1) the fold increaseof a gene deletion or a chemical on WT cells over untreated WT cells, thenexpressed this value as CANR×10−7 per generation by multiplying by theuntreated WT rate. The observed rates and associated confidence intervalswere measured as described (46). Expected values below the 95% confi-dence interval of the observed mutation rate were considered candidateSHyp interactions.

MA and Whole-Genome Sequencing. Diploid strains were pulsed with 100 μMcisplatin for 2 h before plating on YPD. Diploids were used to buffer delete-rious mutations, and a pulse of cisplatin ensured that mutations altering themutation spectrum did not accumulate. For MA lines, single colonies wereplated on YPD and passaged 40 times by restreaking to fresh YPD every 2–3 d(∼1,000 generations). Twelve independent lines for each condition were out-grown for genomic DNA extraction (8). DNA was subjected to 125 bp paired-end whole-genome sequencing at Canada’s Michael Smith Genome SciencesCentre using the Illumina HiSeq2000 platform. The average sequencing

Segovia et al. PNAS | March 7, 2017 | vol. 114 | no. 10 | 2667

GEN

ETICS

Dow

nloa

ded

by g

uest

on

May

24,

202

1

Page 6: Hypermutation signature reveals a slippage and realignment ...Hypermutation signature reveals a slippage and realignment model of translesion synthesis by Rev3 polymerase in cisplatin-treated

coverage compared with the haploid reference was 129× (i.e., >60× for eachdiploid genome). Sequence files are deposited at the Sequence Read Archive(www.ncbi.nlm.nih.gov/sra) (accession no. SRP091984). Reads were aligned tothe Saccer3 reference genome using Burrow–Wheeler Aligner 0.5.7 (47). Var-iants were identified exactly as described previously (8). Mutations wereextracted by comparing evolved to parental genomes. Variant calling wascarried out using the mpileup utility of the SAMtools package 0.1.18 (48) witha threshold of 20. Copy-number variants were analyzed using a locally modi-fied version of CNAseq (49). Additionally, we manually verified variant callswith the Integrated Genomics Viewer (Broad Institute).

Mutation Signature Analysis. Flanking sequence was extracted from theSaccer3 genome based on the mutation positions (Dataset S1). The enrichmentof motifs within the diploid rad1Δ-cisplatin genomes was calculated exactly asdescribed based on a set of 41 mers (±20 bp around the SNV) (25). The sig-nificance of motif enrichments was calculated using the hypergeometric

distribution. pLogos were generated online at plogo.uconn.edu/ (50). Asforeground input n(fg), 41-mer sequences from the rad1Δ-cisplatin genomeswere used. These sequences were centered at the mutated C for all substitu-tions with fixed positions at C and GC. Unique 41-mer sequences automaticallygenerated by pLogo from the Saccer3 yeast genome were used as backgroundinput n(bg). Reanalysis of literature variants was performed by manuallyscoring whether the SNV matched the base immediately 3′ of the mutation(28, 29). The proportions of these templated SNVs were compared with a two-tailed Fisher exact test.

ACKNOWLEDGMENTS. We thank N. O’Neil and the P.C.S. laboratory forreading the manuscript and Philip Hieter for strains. The work is fundedby Canadian Institutes of Health Research (CIHR) Grant MOP-136982, Nat-ural Sciences and Engineering Research Council of Canada Grant RGPIN2014-04490, and the British Columbia Cancer Foundation. P.C.S. is a CIHRNew Investigator.

1. Farlow A, et al. (2015) The spontaneous mutation rate in the fission yeast Schizo-saccharomyces pombe. Genetics 201(2):737–744.

2. Lord CJ, Ashworth A (2012) The DNA damage response and cancer therapy. Nature481(7381):287–294.

3. Stratton MR, Campbell PJ, Futreal PA (2009) The cancer genome. Nature 458(7239):719–724.

4. Lord CJ, Ashworth A (2013) Mechanisms of resistance to therapies targeting BRCA-mutant cancers. Nat Med 19(11):1381–1388.

5. Alexandrov LB, et al.; Australian Pancreatic Cancer Genome Initiative; ICGC BreastCancer Consortium; ICGC MMML-Seq Consortium; ICGC PedBrain (2013) Signatures ofmutational processes in human cancer. Nature 500(7463):415–421.

6. Forbes SA, et al. (2010) COSMIC (the Catalogue of Somatic Mutations in Cancer): Aresource to investigate acquired mutations in human cancer. Nucleic Acids Res38(Database issue):D652–D657.

7. Serero A, Jubin C, Loeillet S, Legoix-Né P, Nicolas AG (2014) Mutational landscape ofyeast mutator strains. Proc Natl Acad Sci USA 111(5):1897–1902.

8. Stirling PC, Shen Y, Corbett R, Jones SJ, Hieter P (2014) Genome destabilizing mutatoralleles drive specific mutational trajectories in Saccharomyces cerevisiae. Genetics196(2):403–412.

9. Meier B, et al. (2014) C. elegans whole-genome sequencing reveals mutational sig-natures related to carcinogens and DNA repair deficiency. Genome Res 24(10):1624–1636.

10. Costanzo M, et al. (2016) A global genetic interaction network maps a wiring diagramof cellular function. Science 353(6306):aaf1420.

11. Lee AY, et al. (2014) Mapping the cellular response to small molecules using che-mogenomic fitness signatures. Science 344(6180):208–211.

12. Li X, O’Neil NJ, Moshgabadi N, Hieter P (2014) Synthetic cytotoxicity: Digenic inter-actions with TEL1/ATM mutations reveal sensitivity to low doses of camptothecin.Genetics 197(2):611–623.

13. Giaever G, Nislow C (2014) The yeast deletion collection: A decade of functional ge-nomics. Genetics 197(2):451–465.

14. Jasin M, Rothstein R (2013) Repair of strand breaks by homologous recombination.Cold Spring Harb Perspect Biol 5(11):a012740.

15. Lang GI, Murray AW (2008) Estimating the per-base-pair mutation rate in the yeastSaccharomyces cerevisiae. Genetics 178(1):67–82.

16. Huang ME, Rio AG, Nicolas A, Kolodner RD (2003) A genomewide screen in Saccha-romyces cerevisiae for genes that suppress the accumulation of mutations. Proc NatlAcad Sci USA 100(20):11529–11534.

17. Doerfler L, Harris L, Viebranz E, Schmidt KH (2011) Differential genetic interactionsbetween Sgs1, DNA-damage checkpoint components and DNA repair factors in themaintenance of chromosome stability. Genome Integr 2:8.

18. Kokoska RJ, Stefanovic L, DeMai J, Petes TD (2000) Increased rates of genomic dele-tions generated by mutations in the yeast gene encoding DNA polymerase delta or bydecreases in the cellular levels of DNA polymerase delta. Mol Cell Biol 20(20):7490–7504.

19. Schürer KA, Rudolph C, Ulrich HD, Kramer W (2004) Yeast MPH1 gene functions in anerror-free DNA damage bypass pathway that requires genes from homologous re-combination, but not from postreplicative repair. Genetics 166(4):1673–1686.

20. Ang JS, Duffy S, Segovia R, Stirling PC, Hieter P (2016) Dosage mutator genes inSaccharomyces cerevisiae: A novel mutator mode-of-action of the Mph1 DNA heli-case. Genetics 204(3):975–986.

21. Mani R, St Onge RP, Hartman JL, 4th, Giaever G, Roth FP (2008) Defining geneticinteraction. Proc Natl Acad Sci USA 105(9):3461–3466.

22. Wyatt MD, Wilson DM, 3rd (2009) Participation of DNA repair in the response to5-fluorouracil. Cell Mol Life Sci 66(5):788–799.

23. Marteijn JA, Lans H, Vermeulen W, Hoeijmakers JH (2014) Understanding nucleotideexcision repair and its roles in cancer and ageing. Nat Rev Mol Cell Biol 15(7):465–481.

24. Clausen AR, et al. (2015) Tracking replication enzymology in vivo by genome-widemapping of ribonucleotide incorporation. Nat Struct Mol Biol 22(3):185–191.

25. Chan K, et al. (2015) An APOBEC3A hypermutation signature is distinguishable fromthe signature of background mutagenesis by APOBEC3B in human cancers. Nat Genet47(9):1067–1072.

26. Eastman A (1987) The formation, isolation and characterization of DNA adductsproduced by anticancer platinum complexes. Pharmacol Ther 34(2):155–166.

27. Harfe BD, Jinks-Robertson S (2000) DNA polymerase zeta introduces multiple muta-tions when bypassing spontaneous DNA damage in Saccharomyces cerevisiae. MolCell 6(6):1491–1499.

28. Stone JE, Lujan SA, Kunkel TA, Kunkel TA (2012) DNA polymerase zeta generatesclustered mutations during bypass of endogenous DNA lesions in Saccharomycescerevisiae. Environ Mol Mutagen 53(9):777–786.

29. Northam MR, Robinson HA, Kochenova OV, Shcherbakova PV (2010) Participation ofDNA polymerase zeta in replication of undamaged DNA in Saccharomyces cerevisiae.Genetics 184(1):27–42.

30. Braberg H, et al. (2014) Quantitative analysis of triple-mutant genetic interactions.Nat Protoc 9(8):1867–1881.

31. Herr AJ, Kennedy SR, Knowels GM, Schultz EM, Preston BD (2014) DNA replicationerror-induced extinction of diploid yeast. Genetics 196(3):677–691.

32. SenGupta T, et al. (2013) Base excision repair AP endonucleases and mismatch repairact together to induce checkpoint-mediated autophagy. Nat Commun 4:2674.

33. Segovia R, Tam AS, Stirling PC (2015) Dissecting genetic and environmental mutationsignatures with model organisms. Trends Genet 31(8):465–474.

34. Tam AS, Chu JS, Rose AM (2015) Genome-wide mutational signature of the chemo-therapeutic agent mitomycin C in Caenorhabditis elegans. G3 (Bethesda) 6(1):133–140.

35. Huang KK, et al. (2016) Exome sequencing reveals recurrent REV3L mutations in cis-platin-resistant squamous cell carcinoma of head and neck. Sci Rep 6:19552.

36. Mukherjee P, Lahiri I, Pata JD (2013) Human polymerase kappa uses a template-slippage deletion mechanism, but can realign the slipped strands to favour basesubstitution mutations over deletions. Nucleic Acids Res 41(9):5024–5035.

37. Ling H, Boudsocq F, Woodgate R, Yang W (2004) Snapshots of replication through anabasic lesion; structural basis for base substitutions and frameshifts. Mol Cell 13(5):751–762.

38. McCulloch SD, Kunkel TA (2008) The fidelity of DNA synthesis by eukaryotic replica-tive and translesion synthesis polymerases. Cell Res 18(1):148–161.

39. Muñoz-Galván S, et al. (2012) Distinct roles of Mus81, Yen1, Slx1-Slx4, and Rad1 nu-cleases in the repair of replication-born double-strand breaks by sister chromatidexchange. Mol Cell Biol 32(9):1592–1603.

40. Semlow DR, Zhang J, Budzowska M, Drohat AC, Walter JC (2016) Replication-dependent unhooking of DNA interstrand cross-links by the NEIL3 glycosylase. Cell167(2):498–511 e414.

41. Stirling PC, et al. (2011) The complete spectrum of yeast chromosome instability genesidentifies candidate CIN cancer genes and functional roles for ASTRA complex com-ponents. PLoS Genet 7(4):e1002057.

42. Duffy S, et al. (2016) Overexpression screens identify conserved dosage chromosomeinstability genes in yeast and human cancer. Proc Natl Acad Sci USA 113(36):9967–9976.

43. Bhang HE, et al. (2015) Studying clonal dynamics in response to cancer therapy usinghigh-complexity barcoding. Nat Med 21(5):440–448.

44. Wong TN, et al. (2015) Role of TP53 mutations in the origin and evolution of therapy-related acute myeloid leukaemia. Nature 518(7540):552–555.

45. Schumacher TN, Schreiber RD (2015) Neoantigens in cancer immunotherapy. Science348(6230):69–74.

46. Hall BM, Ma CX, Liang P, Singh KK (2009) Fluctuation analysis CalculatOR: A web toolfor the determination of mutation rate using Luria-Delbruck fluctuation analysis.Bioinformatics 25(12):1564–1565.

47. Li H, Durbin R (2009) Fast and accurate short read alignment with Burrows-Wheelertransform. Bioinformatics 25(14):1754–1760.

48. Li H, et al.; 1000 Genome Project Data Processing Subgroup (2009) The SequenceAlignment/Map format and SAMtools. Bioinformatics 25(16):2078–2079.

49. Jones SJ, et al. (2010) Evolution of an adenocarcinoma in response to selection bytargeted kinase inhibitors. Genome Biol 11(8):R82.

50. O’Shea JP, et al. (2013) pLogo: A probabilistic approach to visualizing sequencemotifs. Nat Methods 10(12):1211–1212.

2668 | www.pnas.org/cgi/doi/10.1073/pnas.1618555114 Segovia et al.

Dow

nloa

ded

by g

uest

on

May

24,

202

1