Mutation signatures specific to DNA alkylating agents in yeast and cancers

Abstract

Alkylation is one of the most ubiquitous forms of DNA lesions. However, the motif preferences and substrates for the activity of the major types of alkylating agents defined by their nucleophilic substitution reactions (S_N1 and S_N2) are still unclear. Utilizing yeast strains engineered for large-scale production of single-stranded DNA (ssDNA), we probed the substrate specificity, mutation spectra and signatures associated with DNA alkylating agents. We determined that S_N1-type agents preferably mutagenize double-stranded DNA (dsDNA), and the mutation signature characteristic of the activity of S_N1-type agents was conserved across yeast, mice and human cancers. Conversely, S_N2-type agents preferably mutagenize ssDNA in yeast. Moreover, the spectra and signatures derived from yeast were detectable in lung cancers, head and neck cancers and tumors from patients exposed to S_N2-type alkylating chemicals. The estimates of mutation loads associated with the S_N2-type alkylation signature were higher in lung tumors from smokers than never-smokers, pointing toward the mutagenic activity of the S_N2-type alkylating carcinogens in cigarettes. In summary, our analysis of mutations in yeast strains treated with alkylating agents, as well as in whole-exome and whole-genome-sequenced tumors identified signatures highly specific to alkylation mutagenesis and indicate the pervasive nature of alkylation-induced mutagenesis in cancers.

INTRODUCTION

Sequencing of a vast number of cancer genomes and exomes has demonstrated that many tumors carry a heavy mutation burden (1,2). Such mutations in tumors are often non-random and can be diagnostic of the mutagen that the tumor genome may have historically encountered. Information pointing toward activity of a mutagen can be due to the preference of the agent to mutate certain nucleotides, resulting in the non-uniform mutation spectrum. However, mutation spectra of different agents are limited to just six types of base substitutions, leading to large overlaps. Classic examples of such overlaps include C→T changes that are characteristic of the activity of cytidine deaminases including APOBEC3 enzymes, activation-induced cytidine deaminase (AID), ultra violet radiation (UV)-induced mutations and mutations caused by spontaneous deamination of methylated cytosines within CpG dinucleotides (3). In recent years, analyses focused on understanding the etiology of mutations in cancers have greatly benefited from inclusion of the flanking nucleotides in addition to the mutated residues leading to the identification of mutation signatures in cancers (2). Subsequent studies of mutation catalogs of thousands of tumors led to the identification of mutation signatures attributable to specific DNA damaging agents that further showed preference for double-stranded (ds) and/or single-stranded (ss) DNA (1,4–5). Comparisons of groups of tumors with differential exposures to cancer-causing agents as well as studies in model organisms have indicated the etiology of numerous mutation signatures. However, many important classes of mutagens have no discrete signature(s) assigned.

Alkylation damage is one of the most common sources of DNA lesions. Human cells are exposed to endogenous sources of alkylation primarily in the form of S-adenosylmethionine and nitroso compounds generated by nitrosation of amines (6,7). Many environmental sources of DNA alkylation including components of air and water pollution and tobacco smoke are also responsible for DNA damage and mutagenesis in human cells (8–10). The alkylation reactions on DNA bases involve both the N- and O-atoms. The chemical reaction type (S_N1 or S_N2) defines the atom and DNA base specificity of the alkylating agent. In addition, the specificity of alkylating agents for the N- or O-atom in the damaged base is contingent on whether the reacting DNA base is ss or ds (11–14) and summarized in Table 1). Alkyl-DNA lesions also represent blocks to the replication machinery and can result in mutations upon bypass by translesion synthesis (TLS) polymerases or in double-strand breaks (DSBs) when closely spaced lesions are not repaired (15) leading to gross chromosomal rearrangements. In addition to these mutagenic pathways, various error-free repair pathways exist in mammalian cells to prevent mutagenesis by alkylation damage (16). Recent findings based on model systems and on analyses of mutation catalogs in thousands of cancer genomes revealed that transient ssDNA intermediates formed during DNA replication and DSB repair are hypermutable. While the bulk of mutagenesis occurs in the major ds part of the genome, the hypermutation in ssDNA can significantly contribute in genome-wide mutation load (1,17). As such, it is possible that alkylating agents contribute to hypermutation in cancers through mutagenesis in ssDNA.

Table 1.

The preference for DNA strandedness and the prevailing base changes associated with the lesions induced by DNA alkylating agents analyzed in this study

Alkylation reaction	Agents	Lesions	Locationa	Base changesb
SN1-type alkylation	EMS, MNU, MNNG, Temozolamide	O6-alkylguanine	dsDNA	G→A
SN2-type alkylation	EMS, MMS, tobacco-specific alkylating agents, 1,2-DCP	N1-alkyladenine	ssDNA	A→T
		N3-alkyladenine	dsDNA	A→T
		N7-alkyladenine	dsDNA	A→G
		N7-alkylguanine	ssDNA and dsDNA	G→T and G→C
		N3-alkylcytosine	ssDNA	C→T and C→A
		O4-alkylthymine	NDd	T→A, T→C and T→Gc
		O2-alkylthymine	ND	T→A, T→C and T→Gc
		N3-alkylthymine	ND	T→A, T→C and T→Gc

^aData is obtained from (11).

^bUnless otherwise noted, the data is obtained from (34).

^cIndicates the base change identified in this study.

^dIndicates the location preference has not been determined.

Moreover, while it is clear that base specificity of alkylating agents depends on DNA strandedness, their signatures in single versus dsDNA have not been ascertained. For this purpose, we utilized yeast strains engineered to generate large stretches of regulatable and persistent ssDNA (18). We defined the mutation spectra and signatures of alkylation damage in ssDNA and dsDNA after treating these strains with model S_N1 and S_N2-type DNA alkylating agents, ethyl methanesulfonate (EMS), N-Nitroso-N-methylurea (MNU), Methylnitronitrosoguanidine (MNNG) and methyl methanesulfonate (MMS) followed by whole genome sequencing of yeast isolates. The information gathered in yeast allowed us to probe cancer genomes and exomes wherein we identified mutation patterns ascribable to either S_N1 or S_N2-type alkylation damage.

MATERIALS AND METHODS

Yeast strains

The yeast strains used in this study were derived from CG379 with the following genotype—MATα his7-2 leu2-3,112 trp1-289, cdc13-1, CAN1, ADE2, URA3 and LYS2 were deleted from their original positions and the lys2::ADE2-URA3-CAN1 array was reintroduced near the de novo telomere on the left arm of chromosome V (18). The rev1-AA allele carrying strain were derived from the yKC387 strain (19). REV3 was replaced with the NatMX cassette and RAD30 and MAG1 were replaced with the HphMX cassette (20).

Schematics of the yeast experimental system

The cdc13-1 temperature-sensitive (ts) allele renders the telomeres unprotected upon shifting yeast strains to 37°C. Resection at these unprotected telomeres leads to generation of long stretches of ssDNA and arrest of the yeast culture in G2 phase of the cell cycle. Releasing the cultures into fresh media at the permissive temperature 23°C allows resynthesis of the resected strands and the cell cycle can progress.

We grew yeast strains to ∼2 to 5 × 10⁸ cells/ml. The cultures were diluted 10-fold and shifted to 37°C for 6 h. The strains were then washed, resuspended in water and treated with low doses of MMS (0.5 mM) and EMS (2 mM) for another 24 h at 37°C. For treatment with MNNG and MNU, the strains were washed, resuspended in water and treated with either 30 μg/ml MNNG dissolved in dimethyl sulfoxide, or 40 mM MNU dissolved in dimethyl sulfoxide for 30 min. Appropriate dilutions of the cultures prior to shift at 37°C, 6 h after arrest at 37°C and 24 h or 30 min after treatment with the mutagens at 37°C were plated on complete media to assess the viability of the treated cultures, and on 60 mg/ml L-canavanine and 20 mg/ml adenine sulfate containing media to select for Can^RAde⁻ mutants. The mutants were tested for the presence of the URA3 gene to avoid selection of isolates that have lost the chromosome V left arm via a gross-chromosomal rearrangement. DNA was extracted from independent Can^RAde⁻ isolates, for whole genome sequencing.

DNA sequencing of yeast strains

Whole genome DNA libraries were prepared using the KAPA hyper kit (KAPA Biosystems, Wilmington, MA, USA) and paired end 150 bp reads from the HiSeq 4000 (Illumina, San Diego, CA, USA) were obtained. The reads were aligned to the reference genome ySR127 (21) using the CLC-genomics workbench (Qiagen, Redwood City, CA, USA). The sequences were realigned using the variants detected by the ‘low frequency variant detection’ tool in CLC-genomics workbench and duplicate mapped reads were removed. The final variants were detected using the ‘fixed ploidy’ tool for haploid strains with ploidy = 1 and variant probability = 90% and were filtered for single nucleotide variants (SNVs). From each Can^RAde⁻ isolate sequenced, SNVs that were also present in the parent strains that were not treated with MMS or EMS, were removed from analysis. For Δrad30 strains, we used Sanger sequencing to sequence the reporter region on Chr 5 (1–12 969 bp).

Analysis of mutation spectra

SNVs present within 30 kb of the chromosome ends were annotated as ‘subtelomeric’ changes and the remaining changes that were more than 30 kb away from the chromosome ends were annotated as ‘mid-chromosome’ changes. SNVs were annotated based on pyrimidines, and the reverse complements for each change was accounted for. Ninety-six channel plots were generated using SomaticSignatures (22). For base substitution in ssDNA, the mutations were annotated based on the strand that was rendered single stranded upon telomere uncapping. The complements of the mutations on the left arm of the chromosomes were analyzed since, the bottom strand was single stranded in this configuration. The number of A, T, G or C bases in subtelomeric or mid-chromosome regions were calculated using SeqinR. An example of the mutation frequency calculation is provided below

pLogo evaluation of preferences in nucleotides flanking mutation position

To identify over- or under-represented nucleotides in the ±1 and ±2 positions of the mutated residues, we utilized PLogo, an open access web-based tool. PLogo statistically calculates the over- or under-representation of DNA motifs as compared to the background sequence composition (23). The background sequence provided was a 40mer comprising of the mutated residue and 20 bases of flanking DNA on either side. The tool provides a visual representation of the data wherein the height of the nucleotide correlates with the log odds of the significance of over-representation of the residues.

Knowledge-based statistical evaluation of enrichment and mutation load associated with trinucleotide-centered mutation signatures

To analyze mutation signatures in trinucleotide motifs identified in our experiments or derived from prior knowledge, we calculated enrichment of mutations in either scattered, clustered or all mutations as described in (4,24). Mutation clusters were defined as 2 or more mutations in a span of 10 kb. P-values for clusters were calculated as defined in (4,24). Two or more mutations that were present within 10 bases of each other were denoted as complex events, likely due to Pol ζ activity (25). Such groups of mutations were counted as a single event. Mutations that did not conform to the definition of ‘clusters’ were denoted as ‘scattered’ mutations. To determine enrichment, the fraction of base substitutions of a given type in a signature-defining trinucleotide motif among all base substitutions of this type was calculated. This ratio was further compared the prevalence of the trinucleotide motif versus the single base in the 40 bases flanking the mutated residue. An example of enrichment calculation for the nCy→nTy mutation signature is provided below:

Fishers exact test was utilized to determine if enrichment was statistically significant. We further used the R-command p.adjust() to correct the P-values accounting for multiple testing. Given a set of P-values, this command provides adjusted P-values using either Benjamini–Hochberg method or the Bonferroni method. The minimum estimate of mutation loads for each mutation signature was only calculated for those samples whose enrichment >1 and the false discovery rate-corrected P-value < 0.05. For samples not meeting these criteria, the minimum estimate of mutation loads was denoted as 0. The calculation for minimum estimate of mutation loads is shown below using nCy→nTy mutation signature as an example.

To statistically assess the preference for aCy→aTy changes induced by S_N1 alkylating agents, as compared to bCy→bTy changes (b = t, g or c), we compared the odds-ratios of the two mutation signatures using Breslow-Day test in R.

Determination of transcription-dependent strand-bias in mutation spectra

To determine if the mutations in cancers were present on the transcribed or non-transcribed strand, we used GenomicRanges (26) to annotate the mutations on either strand of the transcripts. A one-sided Chi square goodness of fit analysis was performed with the expected ratios of mutations on the transcribed and non-transcribed strands to be 1:1. The resulting P-values were corrected for multiple testing using p.adjust().

Analysis of mutation distribution around bidirectional promoters was done using bedtools closestBed to annotate the distance of the mutations with respect to the center of the promoters. Mutations 5′ (left) relative to the promoter region's center were annotated as negative distances, while mutations to the 3′ (right) were annotated as positive distances. The positions of the bidirectional promoters were obtained from a previously published study (27). The number of A→T and T→A substitutions for the bias in T→A substitutions, as well as the number of A→G and T→C changes for the bias in T→C changes were calculated in non-overlapping bins of 1 kb up to ranging from −20 to 20 kb. To determine if the transcription strand-bias was statistically significant, a one-sided Fishers exact test was performed for the total number of base substitutions in A and T, respectively within a window of 20 kb. The resulting P-values for each cancer type were corrected to adjust for false discovery rates.

RESULTS

Experimental system

We used a haploid yeast strain with the cdc13-1 ts allele that allows for controlled generation of ssDNA, enabling assessment of mutagenesis in ssDNA and dsDNA (18,21). At the restrictive temperature (37°C), telomeres in these strains get uncapped and resection at their chromosomal ends ensues yielding long stretches of ssDNA (28,29). Shifting strains to the permissive temperature (23°C) allows them to resynthesize the resected DNA strand and restores functional telomeres. Treatment of these strains with DNA damaging agents after the shift to 37°C generates lesions in both subtelomeric ssDNA (within ∼30 kb from the telomeres as defined in (21)) and in dsDNA distant from telomeres (mid-chromosome). Base damage in ssDNA cannot be repaired by either base- or nucleotide excision repair pathway due to the absence of an undamaged second DNA strand for templated repair synthesis. As such, mutations in the subtelomeric ssDNA reflect the specificity of lesion formation and of bypass of the lesion by error-prone TLS. To select mutations introduced in ssDNA, we have utilized a haploid strain harboring a cassette of the three reporter genes CAN1, URA3 and ADE2 near the left telomere of chromosome V (18). The location of these reporters near the telomeric region increases the likelihood of the reporter genes being included in the transient stretch of ssDNA. Similar to Chan et al. (18), isolates with clustered mutations at the left subtelomeric region of chromosome V were selected by simultaneous incidence of resistance to the arginine analog L-canavanine due to a loss of function mutation in the CAN1 gene, and a mutation in the closely located ADE2 gene, which renders yeast colonies red (Figure 1A). To identify mutation signatures in yeast ssDNA and dsDNA we used model alkylating agents—MMS which preferably alkylates DNA via S_N2-type alkylation, EMS which acts through both an S_N1- and S_N2-type reactions and MNNG and MNU which act via S_N1-type of reactions (Table 1). Previously known base specificity of the reactions (Table 1) allowed to assign the mutations generated by these agents to one of the two types of the alkylating reactions. MMS treatment led to a 46-fold increase in the frequency of Can^RAde^- mutants, and EMS led to a 39-fold increase in the Can^RAde^- clustered mutation frequency compared to strains that were not treated with any mutagen. Treatment with MNNG led to an average of 19-fold increase in Can^RAde^- mutation frequencies, and MNU led to a 12-fold average increase in Can^RAde^- mutation frequencies (Figure 1B and Supplementary Table S1). Further, by whole-genome sequencing we were able to detect clustered mutations in subtelomeric regions of other chromosomes thereby increasing the sample size for signature analysis in ssDNA. Analyzing mutations in subtelomeric ssDNA formed by 5′→3′ resection for signature analysis allows unambiguous strand assignment of base lesions resulting in a mutation (Figure 1A). Throughout this paper the base changes in ssDNA have been annotated to refer to the unresected intact strand (e.g. C→T wherein lesions were formed on the cytosine in ssDNA). Mutation spectrum in most of the genome, which is in a predominantly dsDNA form, is a manifestation of lesion formation, efficacy of excision repair pathways and/or error-prone TLS of unrepaired lesions. Although the density of mutations induced in dsDNA distant from telomeres is expected to be much lower than in subtelomeric ssDNA, whole-genome sequencing allowed us to collect enough mutations for assessing dsDNA mutation signatures. Mutations in dsDNA could originate from DNA damage on either strand (e.g. C→T or G→A), therefore, in agreement with the common format of signature presentation, only the mutated pyrimidine will be displayed for denoting mutation signatures in dsDNA. This will be noted in the text and/or figure legends, where appropriate.

Figure 1.

Experimental system for identifying mutation signatures in yeast ssDNA and dsDNA. (A) The experimental design in yeast to obtain DNA damage-specific mutation spectra in ssDNA and dsDNA. The three mutation reporters URA3, ADE2 and CAN1 were placed in the subtelomeric region of ChrV. The yeast strains carried the ts cdc13-1 allele. Upon shift to 37°C, telomeres are uncapped and resection exposes ssDNA at the chromosome ends. Treatment with the mutagen results in lesion formation (solid black triangles) on exposed DNA bases either in ssDNA or in dsDNA (open circles with base annotated). An example of lesion formation in adenines is shown here. Resynthesis of the telomeres on shifting cultures to permissive temperature (23°C) results in mutations (solid circles with the base substitution annotated) in ssDNA due to the lack of repair in ssDNA. An example of A→G changes due to lesion bypass is shown here. On the other hand, damage in dsDNA can be repaired and results in a lower number of mutations. In ssDNA the mutagenic lesions can be unambiguously annotated to the intact strand during DNA end resection (open triangles), while in dsDNA the lesion may be formed on either strand making it difficult to identify the strand-specificity of the mutation. Mutations that inactivate CAN1 and ADE2 yield Can^RAde^- (red) yeast colonies, while mutations that only inactivate CAN1 result in formation of Can^RAde⁺ (white) yeast colonies. Solid gray circles depict to capped telomeric ends. (B) Can^RAde^- mutation frequencies upon treatment with no mutagen, MMS, EMS, MNNG or MNU. For each strain tested, the mean mutation frequency and the standard error of the mean are depicted. Source data is presented in Supplementary Table S1.

Alkylating agents induce mutations in yeast ssDNA and dsDNA

Whole-genome sequencing of 93 MMS-induced and 88 EMS-induced Can^RAde^- yeast isolates confirmed the presence of at least two mutations near chrV left telomere, one in CAN1 and another in the ADE2 gene of each isolate. Can^RAde^- isolates induced by MMS often contained additional mutations near the chrV left telomere as well as near other telomeres, predominantly within 30 kb from the chromosome ends (Supplementary Figure S1). This was in agreement with the previously determined size of subtelomeric regions to be prone to hypermutation in cdc13-1 yeast strains (21) and will be referred to as subtelomeric in this paper. A total of 974 of 1290 bases substitutions in all the MMS-treated isolates were within the subtelomeric regions of the chromosomes. About 85% of these mutations (829 base substitutions) were found to be spaced more tightly than mutations in mid-chromosome regions as defined by spatial clustering analysis (4,24) (Supplementary Table S2). This analysis considers the average density of mutations in the genome of an isolate and the density of mutations in each putative cluster. One hundred and sixty-eight clusters were found in subtelomeric regions, and only five clusters were in the mid-chromosome regions in MMS-treated strains. These data imply that MMS-induced mutations were formed on ssDNA at the resected chromosomal ends. The clusters ranged in size from 2 to 15 mutations in subtelomeric regions (Supplementary Table S2). Overall, our data indicate that MMS-induced mutations were predominantly formed on ssDNA at the resected chromosomal ends.

Conversely, EMS-mutagenized strains had mutations in both mid-chromosome and the subtelomeric regions (Supplementary Figure S2). 632 base substitutions of the total 1170 mutations were found in mid-chromosome locations, while, 538 changes were subtelomeric. A total of 329 subtelomeric mutations fell into the 108 total clusters ranging from 2 to 13 mutations each. Only 10 mutations were present in clusters in mid-chromosome locations. Whole genome sequencing of MNNG- and MNU-treated isolates further demonstrate that these agents also substantially mutagenized mid-chromosome regions. A total of 916 of the MNNG-induced base substitutions, and 1322 MNU-induced base substitutions were located mid-chromosomally, while 579 MNNG-induced mutations, and 563 MNU-induced mutations were subtelomeric (Supplementary Table S2).

In summary, MMS mutagenesis has the strongest preference to ssDNA, while EMS, MNU and MNNG are mutagenic to both ssDNA and dsDNA.

S_N2-type alkylation reactions performed by the alkylating agents in ssDNA, lead to the formation of N7-alkylguanine, N3-alkylcytosine and N1-alkyladenine in ssDNA as well as N7-alkylguanine and N3-alkyladenine in dsDNA (11,30–31). N7-akylguanine is one of the major products of SN2-type alkylation reactions in both ss- and dsDNA. However, this lesion would be accurately and efficiently copied by cytosine and would only lead to mutations infrequently (Table 1 and (32)). N3-alkylcytosines and N1-alkyladenines formed in ssDNA can be copied only by error-prone TLS and thus often result in mutations (4–5,33). In yeast, 5′→3′ resection of uncapped telomeres would render the bottom strand single stranded on the left arm of chromosomes and the top strand on the right chromosome arms. We accounted for this strand switch in ssDNA formation by presenting mutations in nucleotides to be mutated in the strand that would be stay intact after 5′→3′ resection in each subtelomeric region (Figure 2). In agreement with prior studies (summarized in Table 1), MMS treatment predominantly led to mutagenesis at C, G and A nucleotides of ssDNA (Figure 2A and Supplementary Figure S2). We calculated the density of base substitutions in the subtelomeric regions of all chromosomes. The strand rendered single stranded upon 5′ to 3′ resection (top strand for right and bottom strand for left telomeres) was used as the reference strand to determine the damaged residue. Most of the changes were in cytosines (C→G, C→A and C→T), while a smaller fraction was in adenines (A→G, A→T) and guanines (G→T and G→A) in ssDNA. Mutations in T residues made up a very small fraction of the changes in concordance with previous studies (Figure 2A and Supplementary Table S3). These data, in combination with prior knowledge of the chemical specificity of DNA alkylation indicate that the predominant mutagenic lesions formed by MMS in ssDNA were N3-methyl cytosine, N7-methyl guanines and N1-methyl adenine, with a minor fraction of mutations resulting from methyl thymines.

Figure 2.

EMS and MMS-induced mutation density in yeast ssDNA and dsDNA. The mutation density for MMS, EMS, MNNG and MNU-induced mutations in (A) ssDNA in (B) dsDNA. Mutation density per nucleotide for ssDNA is calculated as the ratio of number of base substitutions in the DNA strand rendered single stranded per strain and the number of mutable bases in the subtelomeric regions. For dsDNA, the mutation density per base pair is calculated as the number of base substitutions in mid-chromosome regions per sequenced strain versus the number of mutable base pairs in mid-chromosome regions (Supplementary Table S3).

Analysis of the spectra of changes by EMS, MNNG and MNU revealed that unlike subtelomeric regions, C→T (and G→A) changes made up the majority of the base substitutions in mid-chromosome regions (Figure 2B and Supplementary Figure S2). It has previously been shown that S_N1-type alkylating agents can efficiently form O6-alkylguanine lesions. Frequent error-prone copying of this lesion with thymine by a replicative DNA polymerase leads to G→A transitions (30,34–35).

Overall, we show that mutagenesis by S_N2-type alkylation preferably occurs in ssDNA, while, mutations due to S_N1-reaction-formed O6-ethylguanine are abundant in dsDNA.

Alkylation-induced mutagenesis by S_N2 agents is partially prevented by the activity of Mag1

In bacteria and mammalian cells N1-methyladenine and N3-methycytosine lesions in ssDNA can be reversed to normal bases by AlkB enzymes (reviewed in (36)). Recently it was demonstrated that the Mag1 glycosylase in yeast has an expanded repertoire of substrates including N1-methyladenine and N3-methycytosine, thereby facilitating base excision repair of these lesions and compensating for the absence of AlkB-type direct reversal activity (37). As such, we deleted MAG1 in our strains to identify its involvement in repair of alkylation damage by EMS or MMS in ssDNA and dsDNA. Treatment of the Δmag1 isolates with MMS and EMS for 24 h at 37°C led to a reduction in viability of the isolates as compared to the wild-type strains (wild-type isolates treated with either MMS or EMS = 56% viability as compared to untreated isolates, Δmag1 strains treated with MMS = 20% viability and Δmag1 strains treated with EMS = 27% viability as compared to untreated isolates). This result is in agreement with the known role of Mag1 in repair of a variety of MMS-induced lesions. Deletion of the MAG1 gene, led to a 4-fold increase in the ssDNA-associated MMS-induced Can^RAde^- mutation frequency and a 2-fold increase in EMS-induced Can^RAde^- mutation frequency. Whole genome sequencing of the isolates demonstrated that the mutation spectra of Δmag1 isolates treated with MMS was similar to the wild-type strains in both subtelomeric and mid-chromosome regions (Figure 2, 3 and Supplementary Tables S1–3). However, both MMS and EMS-induced mutation frequencies of all base substitutions in mid-chromosome regions were increased from 1.18- to 10.84-fold in Δmag1strains than the wild-type isolates (Supplementary Table S3). The mutation frequencies for all base substitutions in the subtelomeric regions were increased by 1.79- to 10.02-fold in the Δmag1 isolates treated with MMS, and by 1.38- to 4.86-fold in strains treated with EMS, as compared to wild-type strains. These data corroborate the previous findings that Mag1 glycosylase can act upon the substrates of AlkB in yeast (37), also see ‘Discussion’ section). We conclude that Mag1 activity can prevent alkylation mutagenesis, however as we still see strong increase in mutagenesis in ssDNA of the MAG1 wild-type strains treated with MMS or EMS, this prevention is only partial.

Figure 3.

Can^RAde^- mutation frequencies and spectra in Δmag1 yeast strains. (A) Can^RAde^- mutation frequencies in wild-type or Δmag1 strains treated with MMS or EMS. * denotes P-values < 0.05 for an unpaired t-test comparing the mutation frequencies in the wild-type and Mag1-deficient strains. Data are presented in Supplementary Table S1. (B) The mutation density for MMS or EMS-induced mutations in ssDNA and dsDNA in Δmag1 strains. The source data for this figure is presented in Supplementary Table S3.

Alkylation mutagenesis by S_N2 agents in ssDNA relies on translesion synthesis

It was previously established in yeast that mutagenesis by MMS relies on conversion of damaged bases into AP-sites and Polζ -dependent translesion synthesis (TLS) (38,39). We previously showed that functional Polζ is also required for MMS-induced mutagenesis in ssDNA, where AP-sites are not formed (33). EMS mutagenesis in dsDNA does not require Polζ function (40) presumably because error-prone copying of O6-alkyl guanine is performed by a replicative DNA polymerase. In addition to Polζ, the other polymerases Polη and Rev1 also participate in TLS. Although Rev1 is required for maintaining the structural integrity of the Polζ complex, it also provides independent TLS activity by inserting cytosines across DNA lesions (41–45). On the other hand, Polη (Rad30) is separate from the Polζ branch of TLS and is required for error-free bypass of a variety of DNA lesions (46–49). To assess the roles of each of these TLS polymerases in alkylation-induced mutagenesis in ssDNA, we measured frequencies of MMS or EMS-induced Can^RAde^- mutation in subtelomeric ssDNA (Figure 4A and Supplementary Table S1) in strains deficient in the different branches of TLS. Deletion of REV3, the catalytic subunit of Polζ eliminates TLS dependent on the catalytic activity of Polζ as well as Rev1. As such, Δrev3 strains displayed a strong decrease in MMS as well as of EMS-induced mutagenesis in ssDNA (Figure 4A). These observations agree with our conclusion based on mutation spectra data (see above) that like MMS-induced mutagenesis, EMS mutagenizes ssDNA via S_N2-type reactions.

Figure 4.

Can^RAde^- mutation frequencies, genome-wide mutation load and spectra in wild-type isolates and TLS mutant strains. (A) Can^RAde^- mutation frequencies in wild-type, and TLS mutant strains treated with no mutagen, MMS or EMS. The mean and standard error of the mean for the mutation frequencies is shown (Supplementary Table S1). An unpaired t-test was used to compare the mutation frequencies in wild-type strains and TLS-deficient strains. * denotes P-value < 0.05, ** denotes P-values < 0.001. (B) The mutation loads in subtelomeric (subtel) and mid-chromosome (mid-chr) regions of the genome in wild-type or rev1AA yeast strains treated with MMS or EMS (Supplementary Table S2). The median number of mutations in the strains are denoted with the black rectangle, and the error bars indicate the 95% confidence intervals. A Mann–Whitney U-test was performed to determine if the increase in mutation loads in rev1AA yeast strains were statistically relevant. For samples with statistically significant increase, the resulting P-value is depicted on the graph. (C) The spectra for base substitutions in cytosines and adenines in ssDNA are provided for WGS wild-type and rev1AA strains treated with MMS and for the reporter region sequenced in Δrad30 strains (Supplementary Table S3). A one-sided Fisher's exact test was performed to test the hypothesis that the ratio of C to G substitutions versus other substitutions in cytosines would be lower in rev1AA strains as compared to the wild-type strains and the ratio of A to G substitutions versus other substitutions in adenines would be lower in rev1AA strains than the wild-type strains treated with MMS. The resulting P-values for mutations in cytosines in rev1AA strains is 0.03 and for the mutations in adenines is 0.0011. The P-values for mutations in cytosines in the Δrad30 isolates is <0.0001 and for adenines is 0.178.

Deletion of RAD30 (Pol η) did not impact mutation frequency in EMS-treated strains and caused a 3-fold increase in MMS-induced Can^RAde^- mutation frequency as compared to the wild-type strains (Figure 4A). Also, the catalytically dead rev1AA allele (D647A and E648A) did not impact mutation frequencies in EMS treated cultures, while increasing Can^RAde^- mutation frequency 7-fold in MMS treated cultures as compared to the wild-type MMS-treated strains (Figure 4A). We speculate that these increases may reflect error-free bypass of a subset of alkylated bases by either of these TLS activities.

We sequenced the genomes of 31 Can^RAde^- MMS-treated rev1AA isolates and 24 Can^RAde^- EMS-treated rev1AA isolates. In rev1AA isolates, there were 23 mutations per isolate on average in the subtelomeric regions as compared to an average of 13 mutations in subtelomeric regions per isolate in wild-type MMS-mutagenized Can^RAde^- strains. Furthermore, analysis of base substitutions in rev1AA strains demonstrated that these isolates have a 2.86- to 5.1-fold increase in the density of each of the three guanine residue substitutions as compared to the MMS-treated wild-type strains (Supplementary Table S3), implying a role for Rev1 in accurately bypassing MMS-induced guanine lesions (likely N7-methylguanines). This increase in mutations in rev1AA strains was not detectable for the MMS-induced mutations in mid-chromosome regions, or for EMS-induced mutations throughout the genome (Figure 4B). The lack of increase in EMS-induced mutations in rev1AA isolates might reflect differences in bypass of methylated and ethylated bases in DNA (also see ‘Discussion’ section). Analysis of the base substitution spectra in these isolates demonstrated that in the rev1AA strains, the proportion of A→G and C→G changes versus all other base substitutions in cytosines and adenines in ssDNA were reduced as compared to the wild-type strains (Figure 4C). C→T changes made up 41% of the changes in cytosines, while C→G changes were 26% of the changes in cytosines in wild-type isolates (233 C→T changes, 149 C→G changes), while in rev1AA strains, C→T changes were 49% and C→G were 21% of the changes in cytosines (151 C→T changes, 64 C→G changes). A→T changes were equal to A→G changes in the wild-type isolates (101 mutations each, 44% of the mutations in adenines each), while in the rev1AA isolates, A→G changes only made up 27% of the changes in adenines (36 A→T substitutions), while A→T changes made up 57% of the substitutions in adenines (75 A→G substitutions) (Figure 4C and Supplementary Table S3). A→G and C→G changes would likely occur due to the insertion of a cytosine opposite the lesion by TLS. These observations are in agreement with the Rev1 terminal dCTP tranferase being capable of inserting only cytosines across lesions (19,38). We also analyzed mutations induced by MMS in the reporter region in Δrad30 strains. We did not see an impact of Rad30 deficiency on the MMS-induced mutation spectra in adenines in ssDNA, however, we did notice a decrease in the relative proportion of C→G changes as compared to wild-type strains (7% C→G changes, 10 C→G base substitutions, and 47% C→T changes, 67 C→T base substitutions) (Figure 4C and Supplementary Table S3). These data indicate that Polη may play a role in bypass of N3-methylcytosines or abasic sites created from N3-methylcytosines in ssDNA.

S_N1-type alkylating agents induce C→T changes in nCy context with a preference for aCy→aTy changes

PLogo analysis of EMS-induced C→T changes demonstrated that the mutated cytosine was frequently flanked by an adenine in the −1 position and by a thymine in the +1 position (Figure 5A). It is important to note that the predominant lesion formed by S_N1-type alkylating agents is O6-alkyl guanine leading to G→A in purine containing strand, however, to maintain nomenclature, we denote the mutations in the pyrimidine containing strand. We compared EMS-induced mutations in wild-type yeast strains to other S_N1-type alkylation agents—MNU and MNNG, and those seen in cancers treated with the S_N1-type alkylation agent temozolomide (2,50–51) as well as in mice embryonic fibroblasts mutagenized by MNNG (52), and lung tumors in mice exposed to MNU (53). The signature identified in yeast was highly similar to non-negative matrix factorization (NMF)-derived Signature 11 (2) and in COSMIC (https://cancer.sanger.ac.uk/cosmic/signatures)) with the eight major peaks pertaining to nCy→nTy changes, which include the aCt→aTt motif suggested by PLogo. We detected a statistically significant enrichment with the nCy→nTy signature in yeast strains treated with EMS, MNU and MNNG, while MMS-treated yeast genomes did not show such an enrichment (Supplementary Figure S3 and Table S4).

Figure 5.

Mutation signatures associated with S_N1-type alkylation in yeast, mice and cancers. (A) PLogo graphical output depicting the +1 and −1 base preference for mid-chromosome C→T changes in EMS-treated yeast strains. The red line at log odds of the binomial probability ±2.5 indicates P-value < 0.05. (B) The enrichment for C→T changes in the aCy motif or in the bCy motif, where ‘b’ denotes c, t or g. The enrichment values are provided for yeast strains treated with EMS (EMS yeast), MNNG (MNNG yeast), MNU (MNU yeast), hypermutated glioblastomas treated with temozolomide (TMZ GBM), mouse embryonic fibroblasts treated with MNNG (MNNG MEFs) and mouse lung tumors treated with MNU (MNU mouse lung tumors). * denotes P< 0.05 for a Breslow Day test for comparison of odds ratios for the two signatures in each cohort (Supplementary Tables S5 and 6). (C) Comparison between the enrichment values of the UV-specific mutation signature (yCn→yTn) and the temozolamide (S_N1-type alkylation damage)-induced mutation signatures (nCy→nTy and aCy→aTy) in melanomas. The enrichment for the specified mutation signatures was obtained for each whole exome sequenced melanoma sample in The Cancer Genome Atlas (TCGA) (Supplementary Table S7). Correlation between the two signatures was determined by a two-sided Pearson correlation analysis.

We further analyzed if the preference for adenines in the −1 position or thymines in the +1 position was a hallmark of S_N1-type alkylation mutagenesis. While, we did not see higher enrichments for C→T changes in the nCt motif than the nCc motif in any of the other S_N1-type alkylating agents-treated samples (Supplementary Table S5), measurement of enrichment for C→T changes in either aCy, or bCy motifs (b = t, c or g) in EMS, MNU or MNNG treated yeast genomes, temozolomide-treated hypermutated recurrent glioblastomas, MNNG-treated mice embryonic fibroblasts and MNU-treated mice lung tumors demonstrates a bias toward mutations in the aCy context (Figure 5B; Supplementary Figure S4 and Tables S5-6). Such a bias toward the aCy context is not visible in the NMF-derived 96-channel signatures, possibly because NMF-based analyses do not take into account the background nucleotide composition around the mutated residues. Interestingly, in MNU and MNNG-treated yeast isolates, we also see high enrichment of C→T changes at the gCy motif (Supplementary Figure S4 and Table S5). This change was not enriched in samples from mice treated with the same agents, indicating that this may be due to differences in yeast and mammalian DNA repair systems (Supplementary Table S5). Nonetheless, the statistically significance enrichment of C→T changes in the aCy motif is prevalent in all samples treated with S_N1-type alkylating agents. These data indicate that the preference for C→T changes in the aCy motif, revealed by our analysis is a universal feature of S_N1-type alkylating agents irrespective of the chemical and the organism impacted.

Since, the temozolomide signature nCy→nTy, overlaps UV-mutation signature (yCn→yTn), we see a strong correlation between enrichments for the UV-specific yCn→yTn signature and the nCy→nTy signature in melanoma samples present in The Cancer Genome Atlas (TCGA). However, comparison of enrichments for aCy→aTy changes in melanomas versus yCn→yTn changes does not indicate a positive correlation between the two mutation signatures (Figure 5C and Supplementary Table S7). Thus, the aCy→aTy signature is more specific for alkylation damage leading to formation of O6-alkylguanines in DNA. Limiting the S_N1-type alkylation signature to its more specific (aCy→aTy) component can help differentiate between UV-mutagenized samples and alkylating drug-mutagenized tumor samples.

Strand-biased MMS-like mutations in adenines are seen in cancers with a history of alkylation damage

We noted that MMS led to more A→T and A→G changes than T→A and T→C changes and similarly, more mutations in cytosines than in guanines in DNA rendered single stranded upon resection (Figure 2). Strand-biased mutations in cytosines are found in many different cancers and are characteristic of a variety of mutagenic processes (1,2). As such, to specifically address alkylation-induced mutagenesis, we focused on mutations in adenines. The strand-biased nature of A→T and A→G changes in our study echoed previous work that demonstrated that treatment of yeast strains with MMS leads to transcriptionally-biased mutations in adenines, likely due to the preferred formation and escape from base-excision repair of N1-methyladenines on the ss non-transcribed strand during transcription (54). Therefore, we proposed that S_N2-type alkylating agents would have a transcriptionally biased mutation spectrum wherein A→T and A→G mutation loads would be higher on the non-transcribed strands of genes than the transcribed strand. We analyzed the strand-bias for A→T and A→G changes in a 20 kb window to the left and right of bi-directional promoters in whole exome-sequenced cancer samples from TCGA (55). Only samples with at least 500 mutations in adenines in the selected windows were analyzed, to avoid samples with very low A→T and A→G changes that would confound the study. In lung adenocarcinomas (LUAD), lung squamous cell carcinomas (LUSC), head-neck squamous cell carcinoma (HNSC) and liver hepatocellular carcinoma (LIHC), A→T changes were less than T→A and A→G changes were less than T→C to the left of the bi-directional promoter, while A→T were greater than T→A substitutions and A→G were more than T→C substitutions to the right of the bi-directional promoters (Figure 6A-B and Supplementary Table S8). These observations are consistent with the bottom strand being ss and amenable to S_N2-alkylation damage to the left of the bi-directional promoters whereby transcription occurs using the top strand as the template strand, and vice-versa to the right of the bi-directional promoter. We did not see this pattern for strand-biased mutagenesis in other cancers. Additionally, we analyzed the TCGA mutation catalogs of whole genome sequenced (WGS) lung cancers (56). The specificity for A→T and A→G changes on the non-transcribed strand around bidirectional promoters was maintained in these samples as well (Supplementary Table S8). We also saw an overall increase in transcriptome-wide A→T and A→G changes on the non-transcribed strands in a variety of cancers including LUAD, LUSC, and HNSC cancers (Figure 6C and Supplementary Table S9)

Figure 6.

Transcription associated strand-bias for A→T and A→G mutations. (A) A pictorial representation of transcription bubbles originating from a bidirectional promoter (black circle, BDP). The transcripts are portrayed as black lines originating from the RNA polymerase machinery (orange circles). Lesions accumulated in the non-transcribed strand are denoted as red stars. (B) The distribution of A→T or A→G changes around a BDP in cancers. The red dots indicated the number of A→T or A→G changes per kb. The blue dots indicate T→A or T→C changes per kb. The X-axis denotes the distance from the center of the BDP (denoted as 0). The number of mutations per kb are plotted up to 20 kb to the right (20000 on the X-axis) and up to 20 kb to the left (−20000 on the X-axis) of the BDP. The cancer cohorts shown here are—LUAD, LUSC and HNSC. The analysis for other TCGA whole exome and WGS samples are present in Supplementary Table S8. P-values are calculated using a Fisher's exact test predicting that mutations in thymines are more to the left of the BDP than to the right of the BDP, while mutation loads in adenines are higher to the right of the BDP than to the left. (C) Exome-wide transcriptional strand-bias for A→G and A→T mutations cancers. * denote P-values < 0.01 as determined by a binomial test to assess the prediction that mutations in adenines are prevalent on the non-transcribed strand. Only data from LUAD, LUSC and HNSC cohorts from TCGA and cholangiocarcinomas from individuals exposed to 1,2-DCP (Chol) are shown here. The analysis for other TCGA tumors is present in Supplementary Table S9.

S_N2-type alkylating agents in tobacco smoke may also contribute to the mutation spectra seen in lung and head and neck cancers. We also saw a significant strand-bias for A→T and A→G changes on the non-transcribed strand in whole-exome sequenced tumors obtained from 4 patients with cholangiocarcinomas, who had a history of exposure to the haloalkanes 1,2-dichloropropane (1,2-DCP) and/or dichloromethane (57)- known S_N2 type alkylating agents (Figure 6C and Supplementary Table S9). Taken together, these data imply that the propensity for A→T and A→G changes originating from lesions on ssDNA is a feature that is characteristic of S_N2-type alkylating agents.

S_N2-type alkylating agents lead to T→G changes in the hTg motif in yeast as well as in cancers

Inspection of the 96-channel mutation spectra and subsequently PLogo analysis of MMS-induced mutations revealed that T→G changes were flanked by a guanine on the 5′ end (Figure 7A and Supplementary Figure S2). This substitution was found to be consistent with the slippage and realignment activity of Polζ and likely indicates the signature of bypass of methylated thymines by translesion polymerases (58). However, we also saw T→G changes in the gTg context in yeast strains that were not exposed to alkylating agents (Supplementary Figure S2 and Supplementary Table S2). To avoid overlap with these background mutations in ssDNA, we only used hTg→hGg (h is either a, t or c) signature. We statistically confirmed a 2.5-fold enrichment of the hTg→hGg signature in in MMS-mutagenized strains (Figure 7B and Supplementary Table S10). Interestingly, T→G changes flanked by a guanine in the +1 position are also visible in mid-chromosome regions (Supplementary Table S2 and Figure S2), therefore, we speculate that MMS-induced alkylthymine lesions are likely formed in both ssDNA and in dsDNA. Although we did not detect the hTg→hGg signature in TMZ-treated glioblastomas, or MNNG- and MNU-treated mice tissue, we detected this mutation signatures in ssDNA of samples treated with S_N1-type alkylating agents, indicating that these agents may also have S_N2-type activity on ssDNA (Figure 7B and Supplementary Table S10). We also detected this signature in whole exome sequenced glioblastomas in TCGA (Supplementary Table S10). However, since these tumors do not have a known history or treatment with TMZ, nor are they hypermutated as seen in the TMZ-treated glioblastomas with MGMT defects, it is possible that this signature is reminiscent of endogenous alkylation damage in these cancers.

Figure 7.

Mutation signatures associated with S_N2-type DNA alkylation in yeast, and cancers. (A) PLogo shows an overrepresentation of G in the +1 position of T→G changes (fixed position) induced by MMS in ssDNA in yeast. The red line at log odds of the binomial probability ±2.5 indicates P-value < 0.05. (B) Enrichment values for T→G changes in the hTg motif (h denotes a, t or c) in yeast strains treated with MMS, EMS, MNU, MNNG or no mutagen. The * denotes P-values < 0.05 (see ‘Materials and Methods’ section). (C) The enrichment and minimum mutation load for hTg→hGg changes in WGS cancers. Each pie chart represents WGS samples from a given TCGA cohort. The colors indicate hTg→hGg fold enrichment. Samples with hTg→hGg enrichment < 1 or with Benjamini-Hoechberg corrected P-values > 0.05 are represented in black and are excluded from the scatter graph below. The scatter graph depicts the minimum mutation loads for hTg→hGg base substitutions in each sample of the represented WGS cancers. In addition to the TCGA samples, the pie chart and scatter graph also depict whole exome sequenced cholangiocarcinoma samples from patients exposed to haloalkanes. (D) The hTg→hGg minimum mutation loads for WGS lung tumors obtained from never-smokers (blue circles) or smokers (red circles). The median values and the 95% confidence intervals are plotted as black lines on the graphs. P-value was calculated using a two-sided Mann–Whitney U test. The data for this figure is presented in Supplementary Table S10.

We further probed the previously published WGS samples with the signature identified in yeast to identify cancers with statistically enriched hTg→hGg changes. Seven of the 14 WGS cancer types demonstrated an overall enrichment value >1 with a Benjamini–Hoechberg corrected P-value < 0.05 (Figure 7C and Supplementary Table S10). Interestingly, LUAD and LUSC cancers wherein tobacco smoke is as a major carcinogen, demonstrate high enrichment with this mutation signature suggesting a role of the S_N2-type alkylating agents in tobacco smoke in mutagenesis. The hTg→hGg signature was also enriched in two of the whole exome sequenced cholangiocarcinoma samples from patients exposed to 1,2-DCP (Figure 7C and Supplementary Table S10), further implying the role of S_N2-alkylating agents T→G base substitutions. The low but detectable levels of hTg→hGg changes in cancers that do not share tobacco smoke as a common etiology attest to the widespread nature of endogenous and environmental DNA alkylating agents capable of damaging DNA.

Since, hTg→gGg mutations were enriched in smoking-associated cancers, we hypothesized that these changes were likely due to alkylation damage by components of tobacco smoke. Using the data on smoking status for the WGS LUSC and LUAD samples, we divided the samples into tumors obtained from either lifelong never-smokers or from smokers. Analysis of the hTg→hGg signature in these samples demonstrated that this signature was not statistically enriched in samples from never-smokers (P-value > 0.05 in 11/15 tumors from never smokers) and the minimum-mutation loads associated with such samples were 0. On the other hand, the hTg→hGg signature was significantly enriched across most tumors from current or past smokers (enrichment >1 and P-value < 0.05 in 56/72 tumors) with a higher minimum mutation load in smokers compared to never-smokers (Figure 7D and Supplementary Table S10).

To further determine that tobacco smoke did not increase mutagenesis at thymines irrespective of the trinucleotide motif, we analyzed the gTh→gGh mutation signature in WGS lung cancers. The gTh→gGh motif is orthogonal to the tobacco smoke-associated hTg→hGg mutation signature. We did not see an enrichment of the gTh→gGh signature in any of the lung cancer samples (Supplementary Table S11). Therefore, our analysis indicates that the hTg→hGg signature is specific to DNA alkylation damage associated with tobacco smoke exposure.

Alkylation-induced mutation signatures have recently been derived by unsupervised mathematical signature extraction from whole genome sequencing of human cell lines treated with various alkylating agents (59). We probed these mutation calls with the trinucleotide-centered mutation signatures ascertained in our study. We found that the samples treated with S_N1-type agents often depict enrichment of nCy→nTy and aCy→aTy signatures, while one sample MMS and two samples treated with Dimethyl sulfate, S_N2-type agents, as well as the samples treated with metabolically activated 1,2-Dimethylhydrazine (S_N1 and S_N2-type agent) demonstrated enrichments of the hTg→hGg and the more relaxed nTg→nGg signatures (Supplementary Table S12). This analysis provides an additional support to our conclusions that the mutation signatures for S_N2 as well as for S_N1-type alkylating agents identified in yeast are conserved in higher eukaryotes.

DISCUSSION

Although signatures for many mutagenic processes are currently known, they are still missing for many environmental DNA damaging agents. Deciphering the underlying causes of signatures often requires unraveling a mixture of mutations originating from a variety of DNA damaging agents potentially present in complex chemical carcinogens. This task is especially difficult when analyzing the impacts of weak mutagens in repair proficient backgrounds. Our approach relies on obtaining mutation spectra for mutagens acting within either ssDNA or dsDNA from sequenced yeast strains. Mutation spectra and signatures gleaned from subtelomeric ssDNA wherein excision repair pathways are not efficient in preventing mutagenesis, likely represent damage specificity and replicative bypass of the lesion. Conversely, active repair pathways in mid-chromosome dsDNA are expected to remove damage, accurately fill the gap using the second strand as a template and thereby prevent mutagenesis. As such, mutations in these genomic regions represent unrepaired or erroneous repair of damage culminating into mutations. Our methodology combines yeast-based mutation spectra data with prior mechanistic studies as well as motif preferences of the mutagen deciphered from studies in model organisms or cancers as inputs to identify trinucleotide-centered mutation signatures in yeast. This method relying on the formulation of a statistically provable knowledge-based hypothesis allows to reveal and statistically evaluate even minor components of mutation signatures that are undetectable by classical NMF-based techniques. By extrapolating our approach to analyze published mutation catalogs from cancer genomes or exomes, we further determined the impacts of S_N1- and S_N2-type DNA alkylation-induced DNA damage on mutagenesis in cancers.

In agreement with biochemical data regarding motif preferences for O6-alkylguanine formation, as well as the NMF-derived signature, mutations in dsDNA of yeast strains treated with EMS also demonstrated an enrichment for nCy→nTy changes (2,44,51,60–62). However, unlike earlier studies that do not indicate a preference for the nucleotide in the -1 position of C→T changes, our knowledge-based analysis in yeast, human tumors treated with TMZ, or mouse tissue treated with MNNG or MNU confirmed the preference for aCy→aTy changes due to mutagenesis by S_N1-type alkylating agents in yeast, humans and mice (Figure 4B). Since most approaches using mathematical deconvolution of signatures do not consider the background nucleotide composition of the mutated region, they may not be able to accurately identify motifs that are preferentially mutagenized within a signature. Thus, detection of the aCy→aTy mutation signature as the diagnostic component of S_N1-type alkylation demonstrates the utility of our methodology.

S_N2-type alkylating agents like MMS can frequently form N7-methylguanine on both ssDNA and dsDNA, N3-methyladenines on dsDNA as well as N1-methyladenine, N3-methylcytosines on ssDNA (12,31). Replicative bypass of unrepaired lesions in adenine and cytosines has been shown to lead to A→T transitions and C→T and C→A base substitutions in Escherichia coli, respectively (63,64) and Table 1). We detected a similar mutation spectrum in yeast ssDNA upon exposure to MMS or EMS, with a prevalence of C→T, C→A likely due to mutagenesis in both cytosines and guanines, as well as A→T and A→G changes (Figure 2).

Deletion of MAG1 in our strains led to a slight increase in mutations in cytosines and adenines in the subtelomeric and mid-chromosome regions of MMS-treated isolates. This agrees with previous studies that demonstrated that Mag1 can initiate repair of N1-methyladenine and N3-methycytosine (37). The increase in mutations in subtelomeric regions indicates that Mag1 is responsible for removal of lesions that may occur in partially ds regions of the subtelomeric ends formed during re-synthesis of the second strand in our strains. As such, Mag1 proficiency would allow for base excision repair to function in these partially ds stretches, thereby reducing mutagenesis.

Although both EMS and MMS form S_N2-type lesions in ssDNA, in rev1AA isolates, mutagenesis was only increased in the MMS-treated isolates. Considering that Rev1 is only capable of inserting cytosines opposite DNA lesions (19,38), it is possible that Rev1 is involved in the error-free bypass of N7-methylguanine-derived lesions in ssDNA, as insertion of C opposite the lesion would not be mutagenic. In the absence of the catalytic activity of Rev1, we note an increase in the total mutation loads in subtelomeric regions in MMS-treated strains, which is further reflected in the increased mutation densities for guanine base substitutions suggesting erroneous bypass of guanine lesions in ssDNA (Figure 4A and B; Supplementary Table S3). Interestingly, this increase in Can^RAde^- mutation frequencies, mutation loads in ssDNA as well as increased guanine base substitutions is not as prominent in EMS-treated rev1AA isolates. These discrepancies in the role of Rev1 in bypass of lesions induced by EMS or MMS in ssDNA might be explained by differences in TLS capability for bypass of ethylated or methylated residues. In addition, EMS also results in the formation of O6-methylguanines, which are usually copied by replicative polymerases resulting in G→A changes (30,34–35). This phenomenon might further mask the impact of the rev1AA allele in altering EMS-induced mutagenesis in guanine residues.

Altogether, analysis of S_N1 and S_N2 mutagenesis in yeast ss- and dsDNA as well as in mammalian systems demonstrate the capacity of our yeast-based system to highlight conserved trinucleotide-centered mutation signatures for specific classes of chemical agents. While many components of the DNA repair machinery differ between yeast and mammalian cells, the similarity of mutation signatures between yeast, mice, human tumors and iPSC-lines that were treated with S_N1 or S_N2 type alkylating agents indicate that the processes yielding alkylation-induced mutations are conserved across species. In this study, we demonstrate that both S_N1 and S_N2-type alkylation-induced DNA damage due to exogenous and endogenous agents is prevalent in human cancers. Cigarette smoke as well as unburned tobacco products like oral snuff or chewing tobacco carry a variety of carcinogenic compounds. Among these, tobacco-specific nitrosamines and ethylene oxide are a very potent DNA alkylating agents and have been directly linked to carcinogenesis (65–68). Products of DNA alkylation have been detected in urine and tissues of smokers (69,70). In keeping with these observations, we detected the S_N2-type mutation signature comprising of strand-biased A→T and A→G changes as well as hTg→hGg substitutions, primarily in lung cancers (Figure 6 and 7C). Moreover, in lung cancers, the enrichment and mutation load of the hTg→hGg signature were significantly higher in smokers than in non-smokers, pointing toward S_N2-type DNA alkylation by tobacco products as a prominent source of the hTg→hGg mutation signature (Figure 7D). We also detect the MMS-induced mutation signature in BRCA, UCEC and GBM samples which might be indicative of endogenous sources of alkylation damage in these cells. Additionally, the presence of alkylation signatures in skin samples might indicate the activity of environmental alkylating agents commonly found as air pollutants in mutagenesis of skin cells (Figure 6C).

Whole genome sequencing of human cell lines treated with various alkylating agents led to the identification of various mutation signatures associated with DNA alkylation (59). The mutation signature associated with the S_N2-type agent, dimethyl sulfate in this study was similar to the MMS-induced mutation pattern in yeast, and we could also detect the aCy→aTy signatures in the S_N1-type agent N-ethyl-N-nitrosourea-treated samples. However, contrary to data from sequenced tumors, mouse models and our yeast results presented here, in the above cited experiments with cell lines, the S_N1-type agents TMZ and MNU predominantly led to T→C changes. It is possible that such differences in prevailing mutation signatures result from differences in repair of alkylation lesions in induced pluripotent cell lines as compared to primary tissues obtained from mice and tumor samples.

In summary, we demonstrated that alkylation-induced DNA damage occurs across a variety of tissue types, and mutation signatures associated with these DNA damaging agents are prevalent in cancers. We noted a distinct DNA strandedness specificity for the different mutagen classes in our yeast studies. The S_N1-type alkylation mutation signature was predominantly associated with dsDNA whereas the mutation spectra and signatures associated with S_N2-type of alkylation were mutagenic mostly to yeast ssDNA. The S_N1-associated signatures identified in yeast were readily detectable in DNA repair-deficient cancers treated with alkylating chemotherapeutic drugs, while the S_N2-signatures could be detected in numerous cancer types, indicating their widespread presence in the environment and endogenously.

DATA AVAILABILITY

The FASTQ files for all the WGS yeast samples in this study have been deposited in the NCBI Sequence Read Archive database under accession number PRJNA596426. The R codes for analysis of the mutation signatures will be provided upon request.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

FUNDING

US National Institute of Health Intramural Research Program Project [Z1AES103266 to D.A.G.]. Funding for open access charge: US National Institute of Health Intramural Research Program Project [Z1AES103266].

Conflict of interest statement. None declared.