A comparative analysis of the mutagenicity of platinum-containing chemotherapeutic agents reveals direct and indirect mutagenic mechanisms

Abstract Platinum-based drugs are a mainstay of cancer chemotherapy. However, their mutagenic effect can increase tumour heterogeneity, contribute to the evolution of treatment resistance and also induce secondary malignancies. We coupled whole genome sequencing with phenotypic investigations on two cell line models to compare the magnitude and examine the mechanism of mutagenicity of cisplatin, carboplatin and oxaliplatin. Cisplatin induced significantly more base substitution mutations than carboplatin or oxaliplatin when used at equitoxic concentrations on human TK6 or chicken DT40 cells, and also induced the highest number of short insertions and deletions. The analysis of base substitution spectra revealed that all three tested platinum drugs elicit both a direct mutagenic effect at purine dinucleotides, and an indirect effect of accelerating endogenous mutagenic processes, whereas the direct mutagenic effect appeared to correlate with the level of DNA damage caused as assessed through histone H2AX phosphorylation and single-cell agarose gel electrophoresis, the indirect mutagenic effects were equal. The different mutagenicity and DNA-damaging effect of equitoxic platinum drug treatments suggest that DNA damage independent mechanisms significantly contribute to their cytotoxicity. Thus, the comparatively high mutagenicity of cisplatin should be taken into account in the design of chemotherapeutic regimens.


Introduction
The formation of new somatic mutations increases tumour heterogeneity and can contribute to treatment resistance. Therefore, there is much interest in assessing the mutagenicity of DNA-damaging cytotoxic treatments. Mutagenic chemotherapy can also contribute to the formation of secondary cancers through inducing mutations in normal somatic tissue.
The platinum drugs cisplatin, carboplatin and oxaliplatin are amongst the most commonly used cytotoxic agents, employed in the treatment of a wide range of cancer types. Cisplatin, which received US Food and Drug Administration (FDA) approval in 1978, is effective in treating, e.g. bladder, head and neck, lung, ovarian, testicular cancer (1)(2)(3)(4)(5). Cisplatin is also a widely used agent in the treatment of paediatric tumours such as neuroblastoma, germ cell tumours, osteosarcoma, retinoblastoma, hepatoblastoma, brain tumours, spinal cord tumours and lymphomas (6). Cisplatin therapy is limited by serious side effects such as neurotoxicity, ototoxicity and nephrotoxicity, and the development of resistance (7)(8)(9). Carboplatin, an analogue of cisplatin selected for reduced nephrotoxicity, obtained FDA approval in 1989. Carboplatin is mainly used for ovarian, lung, and head and neck cancers (10)(11)(12). Neoadjuvant therapy based on either cisplatin or carboplatin has shown benefits in the treatment of triple-negative breast cancer (13,14). Carboplatin is particularly effective compared to docetaxel in the management of germline mutant BRCA1/2 breast cancer (15). Carboplatin is as effective as cisplatin in the treatment of ovarian cancer, but it is better tolerated (16), whereas it shows no clear tolerability benefit over cisplatin in the treatment of lung cancer (17). The third-generation platinum-derived antitumour drug oxaliplatin, which obtained FDA approval in 1996, is mainly used in colorectal cancer in combination therapy (18)(19)(20). Oxaliplatin has lower toxicity than cisplatin (21); it shows very limited activity against cisplatin-resistant tumours (22).
Platinum drugs damage DNA by forming covalent adducts. Cisplatin comprises a platinum atom surrounded in a plane by two ammonia groups and two chloride ligands. After entering the cell, the chloride ligands are sequentially displaced by water molecules, and the resulting positively charged platinum complexes can react with nucleophilic sites of the DNA (23,24). Cisplatin mainly reacts with purine bases, forming intrastrand crosslinks at GpG and ApG or GpGpG and ApGpG sequences (90% of all adducts), and rarer interstrand crosslinks (3-5%) (25,26). Cisplatin can also form DNAplatinum-protein complexes, for instance, it can cross-link chromosomal proteins to DNA (27,28). Carboplatin possesses a similar chemical structure to cisplatin: a platinum atom surrounded in a plane by two ammonia groups and two leaving carboxylate groups in the cis position. The dicarboxylate unit of carboplatin hydrolyses much more slowly than the chloride groups of cisplatin (24), but the adduct-forming moiety is identical. Oxaliplatin has a similar dicarboxylate leaving group to carboplatin, but it also possesses a 1,2 diamino-cyclohexane ligand. The active form of oxaliplatin, therefore, differs from that of the previous two drugs, but it also forms DNA adducts in sequence-and region-specific manner. Oxaliplatin is less reactive with DNA compared to cisplatin (29), but nevertheless more cytotoxic (30).
The mutagenicity of cisplatin has been investigated and reported using prokaryotic and eukaryotic reporter gene assays (reviewed by 31,32) and by whole genome sequencing of animals or cell lines (33)(34)(35)(36). The mutagenic footprint of platinum treatment was subsequently also found in the genomes of primary tumours (35,37,38) and metastases (39,40). In comparative studies, cisplatin and carboplatin were found to be similarly mutagenic in the Salmonella histidine reversion assay when a higher concentration of carboplatin was used (41), and cisplatin was found to be more mutagenic than oxaliplatin in the HPRT gene of mammalian cells at equimolar concentrations (42). Finally, 3.125 µM cisplatin was similarly mutagenic to 5 µM carboplatin on human induced pluripotent stem cells (36). In cancer genomes, somatic mutations attributed to each of the three discussed platinum agents showed varied contribution to the total mutational burden in different tissues (38). Detailed analysis of cisplatin-induced mutational spectra suggested that the most common types of base substitution and indel mutations occur at the sites of intrastrand crosslink adducts due to translesion DNA synthesis (34).
To achieve a controlled comparison of the mutagenicity of these three common platinum-containing cytotoxic agents and to analyse the mechanisms that give rise to their mutation spectra, we assayed mutagenesis in the genomes of two different cultured cell lines upon treatments at carefully determined equitoxic concentrations. We were able to show that all three platinum drugs were mutagenic and induced both single-nucleotide variations (SNVs) and short insertions/deletions (indels). Carboplatin and oxaliplatin caused fewer mutations than cisplatin, although induced similar overall mutational spectra that resembled the spectrum of cisplatin-induced mutations in cancer. We also identified a novel component of the platinum-induced SNV spectra that showed an equal contribution in case of equitoxic treatments with different platinum drugs. The cause of the different mutagenic effects of the tested platinum drugs is likely related to differences in their mechanisms of adduct formation and cell killing.

Platinum treatments at equitoxic concentrations
We employed two cell lines for testing the mutagenicity of platinum chemotherapeutics: the chicken DT40 lymphoblastoma cell line, which was used previously to establish the mutagenicity of cisplatin (34), and the human TK6 lymphoblastoid cell line, which has been extensively used for genotoxicity and mutagenicity assays (43). The karyotypic stability and low spontaneous mutation rate of DT40 make this cell line especially suited for whole genome mutation detection (44). As the first aim of this study was to directly compare the mutagenicity of platinum-containing therapeutic drugs, it was important to establish comparable treatment conditions. We opted for 1-h treatments, which is in the range of the plasma elimination times of free platinum drugs (45,46). We carefully measured the cytotoxicity of 1-h treatments with cisplatin, carboplatin and oxaliplatin using colony formation assays on DT40 cells, and cell viability measurements on TK6 cells ( Figure 1). IC 25 and IC 50 concentrations, representing equitoxic doses that allow 75% or 50% cell survival, were determined using these curves and employed in all subsequent experiments ( Table 1). The chosen treatment conditions showed good agreement between the relative toxicities of the three platinum drugs in DT40 and TK6 cells, with the TK6 cells showing slightly greater sensitivity. Also, the established treatment concentrations correlated very well with published pharmacokinetic measurements; the IC 50 concentrations on the human TK6 cells were approximately 3-4 fold higher than the peak plasma concentrations of the respective platinum drugs (47)(48)(49).

Cisplatin is more mutagenic than carboplatin or oxaliplatin
To accurately determine and compare the mutagenicity of platinum agents, we submitted cell populations derived from single-cell clones to 1-h treatments with each agent at IC 50 or IC 25 concentrations. We performed four repeated treatments at 1-week intervals to mimic clinical conditions. We showed earlier that this treatment regimen at the IC 50 concentration of cisplatin does not select for resistance (34). In parallel, we kept 'mock-treated' cells for the same overall duration. Whole genome sequences of the ancestral clones and 3-5 post-treatment descendant clones were compared with the IsoMut bioinformatic tool (50); any unique mutations that were detected in descendant clones must have arisen during the platinum or mock treatment.
Mutations were classified as SNVs, short insertions and short deletions. This is the first study to determine the spontaneous mutation rate of the TK6 cell line, which was moderately higher than in DT40 cells, measured earlier as 0.46 per Gb diploid genome per cell cycle (34). Specifically, the mean spontaneous base substitution rate was 169 per diploid genome in 50 days, which is equivalent to 0.75 per Gb per cell cycle, calculating with a 16-h cell cycle time. This relatively low spontaneous mutation rate makes the TK6 cell line well suited for mutagenesis studies.
Compared to the mock treatment, the number of mutations significantly increased in all three categories in response to treatment with either platinum agent, in both DT40 and TK6 cells (Figure 2a

Cisplatin induces more DNA damage than carboplatin or oxaliplatin
The different mutagenicity of the platinum drugs could be due to differences in DNA-damaging activity, or differences in the mutagenicity of the repair of drug-induced lesions. To test the DNA damage caused by the treatments, we first assessed the accumulation of phosphorylated histone H2AX (γH2AX). H2AX is phosphorylated by ATM, ATR or DNA-PK, and serves as a marker for both DNA double-strand breaks (DSBs) and for accumulated single-stranded DNA at stalled replication forks (51,52). Twenty-four hours after 1-h treatments at IC 25 or IC 50 concentrations, which were equivalent to the treatments used for the mutagenesis experiments, western blots showed an elevated level of γH2AX in whole cell extracts of DT40 cells in response to each platinum drug ( Figure 3a). Cisplatin treatment induced significantly higher levels of γH2AX than either carboplatin or oxaliplatin at equitoxic levels. Treatment of TK6 cells with cisplatin also induced more γH2AX than carboplatin or oxaliplatin (Figure 3b). In a time-course following 1-h IC 50 treatments, the drug-induced elevation of γH2AX levels was already observable 2 h following the treatment in the case of all three drugs on both DT40 and TK6 cells (Figure 3c, d), suggesting that this is a direct effect due to the stalling and possible collapse of replication forks rather than to DNA breaks caused by apoptosis. γH2AX accumulation in subnuclear foci 24 h after platinum treatment was also apparent using immunofluorescence (Figure 3e of γH2AX foci was in general higher upon cisplatin treatment than upon treatment with carboplatin or oxaliplatin in both DT40 and TK6 cells, at either IC 25 or IC 50 concentrations (Figure 3f, h). The γH2AX focus counts confirmed the western blot results that cisplatin induces more damage than the other tested drugs at equitoxic concentration. We also measured DSB formation directly using single-cell agarose electrophoresis comet assays. We observed a noticeable formation of comet tails following treatment with platinum agents at IC 50 concentrations, although the signal was much weaker than that obtained with the positive control hydrogen peroxide, used at a sublethal concentration (Figure 4a, b). The quantification of tail moments revealed that cisplatin caused more excessive DNA breakage than carboplatin or oxaliplatin in both DT40 and TK6 cells ( Figure  4c, d), which may be a direct effect from collapsing replication forks or the consequence of an earlier start of apoptosis in cisplatintreated cells. In agreement with the induction of DNA breaks, cell cycle analysis using flow cytometry on DT40 cells showed an increased G2/M phase population, an increase in cells arrested in S phase, and an increase in apoptotic cells with a sub-G1 DNA content (Figure 4e, f). The G2 arrest and the apoptotic enrichment were more pronounced after cisplatin treatment, especially 16 h after treatment, though by 24 h, the oxaliplatin-treated cells also started entering apoptosis. Taken together, these results highlight a correlation between the mutagenic and the DNA break-inducing effect of the various platinum agents.

Platinum-induced mutation spectra reveal direct and indirect mutagenic effects
When base substitutions are viewed in the context of the neighbouring bases, the DT40-derived triplet mutation spectra of all three tested platinum agents are dominated by specific C>A and T>A peaks (Figure 5a). The common cisplatin-induced NCC > NAC mutations likely formed at intrastrand platinum crosslinks at GG dinucleotides, in agreement with the conclusions of our previous study (34). We also observed specific T>A substitutions and an increase in most triplet base substitution types. The mutation spectra of carboplatinand oxaliplatin-treated DT40 cells showed similarities with the cisplatin spectrum, but the specific peaks were less prominent ( Figure  5a). C>A and T>A mutations more commonly occurred on the transcribed strand, suggesting transcription-coupled repair of GG and AG intrastrand crosslink adducts (Supplementary Figure 1). Even though the platinum drugs induced a greater number of base substitutions in the TK6 cell genomes, the resulting spectra were relatively featureless (Supplementary Figure 2), although similarities to the DT40-derived spectra such as the emergence of CTN > CAN mutations were apparent.
In order to better understand the base substitution processes upon platinum treatment, we performed principal component analysis (PCA) on the experimental triplet spectra. We included mutation datasets of cisplatin-treated human cell lines and of tumours in which the effect of cisplatin-induced mutagenesis was detected, taken from a recent publication (35). The results were intriguing (Figure 5b). In DT40 cells, principal component 2 (PC2) correlated very well with platinum-induced base substitutions. A similarly good correlation between the PCA result and mutagenicity was apparent in the case of TK6 genomes, except that here a combination of PC1 and PC2 contributed to the result. The analysis of both cell lines suggested that the tested platinum drugs cause base substitutions by similar mechanisms, and the only important difference is the level of mutagenesis. This conclusion is supported by the strong similarity of the dinucleotide mutation spectra and the types and context of short indels induced by the three drugs ( Supplementary Figures 3-6). The mutation spectrum of cisplatin-treated MCF-10A and HepG2 cells, with many C>T mutations (35), seemingly better correlates with PC1 ( Figure 5b). This suggests that there are cell type-dependent differences between platinum-induced mutagenic spectra, with the human TK6 cell line spectrum falling between chicken DT40 cells and human MCF-10A and HepG2 cells. The somatic mutation spectra of cisplatin-treated hepatocellular carcinomas most closely resembled that of TK6 cells ( Figure 5b); therefore, this cell line appears to be a good model for platinum mutagenesis in human cancer. We attempted to extract the base substitution processes from our cell line dataset using non-negative matrix factorisation (NMF). NMF on all the DT40 and TK6 mutation data could fit two components with small errors (Figure 5c, d; Supplementary Tables S12-S14). One NMF component was only detected in platinum-treated samples, and this platinum-related signature A is dominated by mutation classes that are likely associated with platinum-induced intrastrand crosslink adducts (34). The second NMF component described spontaneous mutagenesis in mock-treated DT40 and TK6 cells (Figure 5c, d). The spectrum of this 'background' component resembled that of the 'featureless' COSMIC signatures SBS3, SBS5 and SBS40, of which SBS5 and SBS40 show correlation with the patient's age in some cancer types, and are thought to represent mutagenesis in normal somatic cells (37,53). Unexpectedly, upon platinum treatment, we observed a strong increase in the number of mutations that belong to this background signature, as compared to the matched mock treatment (Figure 5d). The increase was significant in TK6 cells upon treatment with any of the platinum agents. When we restricted the NMF analysis to the DT40 samples, the increase in 'background'-type mutagenesis was also significant with all three drugs in these genomes (Supplementary Figure 2c, Supplementary  Table S13). Whereas the three platinum drugs induced similar numbers of 'background' signature mutations at equitoxic concentrations, cisplatin induced significantly more 'platinum-related' signature SNVs than the other two drugs in both cell types (Figure 5d, Supplementary Table S13).
These novel findings suggest that in addition to causing mutagenic DNA adducts, platinum drugs also increase mutagenesis indirectly by accelerating the spontaneous mutagenic processes in treated cells. As suggested by the results seen in TK6 cells, the indirect mutagenic effect may even be the major contributor to the mutagenic effect of platinum treatment.

Discussion
In this study, we used two cell line-based mutagenesis models to investigate and compare the mutagenic effect of three commonly used platinum-based cytotoxic drugs. Whole genome sequencing revealed dose-dependent mutagenic effects of each tested drug. In our model systems, cisplatin was significantly more mutagenic than carboplatin or oxaliplatin and also caused higher levels of DNA damage at concentrations of equivalent toxicity. The analysis of mutation spectra suggested that in addition to a direct mutagenic effect resulting from intrastrand platinum adducts, the drugs also damage DNA indirectly through a non-specific effect that may be correlated with their toxicity.
Whole genome sequencing provides the ultimate tool for determining the mutagenicity of environmental insults, including natural mutagens such as UV radiation (54), and mutagenic drugs (34). The correlation of cancer-derived mutational signatures with such environmental impacts can help with establishing their mutagenic effects, but the analysis is hampered by different exposures, different genetic backgrounds and other confounders. Only controlled experimental settings with single-parameter differences between treated and control samples can give reliable information on the full mutagenic effect of treatments. Whenever available, it is therefore best to use experimentally obtained mutation spectra for deciphering the biological mechanisms of mutagenesis. Our results demonstrated that platinum drugs induce mutations with two distinct spectra, suggesting a direct and an indirect mutagenic mechanism. The direct mechanism gives rise to point mutations and short indels mostly at GG and AG dinucleotides, presumably sites of platinum-containing intrastrand purine-purine crosslinks (26), and the resulting specific mutation types have also been observed in the genomes of platinumtreated tumours. When comparing the 'direct' platinum-related mutational signature derived from our data to platinum-associated COSMIC signatures, we can spot close similarities to SBS35 with identical prominent C>A and T>A mutation types, but limited likeness to SBS31 that is dominated by C>T mutations. SBS35 appears to describe platinum-induced base substitution mutations in cancer better than SBS31 (37), whereas SBS31 resembles the spectra derived from two other cisplatin-treated cell lines (35). Establishing the causes of the different balance of C>A, C>T and T>A mutations in various platinum-treated cell line models and cancers will require further investigation. We inferred a second, indirect mechanism from the increased contribution of a broad-spectrum SNV signature to overall mutagenesis. As this signature matches the mutational spectrum of mock-treated control cells, its presence is not specific to platinum treatment. Therefore, the observed indirect platinum-induced mutational process would be difficult to detect in tumour genomes, supporting our cell line-based comparative approach. We can only speculate as to the cause of the platinum-induced 'background signature' mutations, which echo our previous observation of increased broad-spectrum mutagenesis upon etoposide treatment (34). The underlying mutagenic process also appears to operate in untreated cells and is therefore probably related to endogenous DNA damage, although we have little understanding of the mechanism of spontaneous mutagenic processes. Platinum treatment stressing the cells may enhance this effect by increasing the production of reactive oxygen species and other adduct-forming species, by reducing the fidelity of DNA synthesis through disturbing intracellular nucleotide availability, or by reducing the general efficiency of DNA repair.
The amount of treatment-induced DNA damage, indicated primarily by the γH2AX signal that appeared soon after treatment, correlated with the mutagenicity of the three tested drugs. Cisplatin appeared to cause more DNA damage and more mutations than carboplatin and oxaliplatin, and the extra mutations were primarily of the direct type at the site of lesions. It follows that at equitoxic concentrations cisplatin causes more DNA adducts, especially considering that cisplatin and carboplatin form identical crosslink adducts. Successful replicative bypass of these lesions then gives rise to the observed mutations, while the collapse of some stalled replication forks at lesion sites can cause DNA breaks. But how could the matched treatments have equal toxicity, if cisplatin caused more  (35) and of 12 cisplatin-treated hepatocellular carcinoma samples. The symbols indicate cell or tissue types, the colours specify platinum drug treatments. COSMIC signatures SBS31 and SBS35 were also included and are shown as labelled. (c, d) De novo NMF of SNV mutation spectra of all sequenced samples identified two components; triplet mutation spectra of NMF signature A (platinum-related) and NMF signature B (background) are shown as the contribution of each triplet mutation type. (d) Mean SNV counts for all sequenced DT40 and in TK6 genomes, split using NMF into 'platinum-related' and 'background' signatures and then averaged by cell type and treatment as indicated. The error bars indicate the SEM of the column below, based on five DT40 or three TK6 independent treated single-cell clones. Significant differences in the mean number of mutations attributed to the 'background' signature, as compared to the mock treatment, are indicated at the bottom of the columns (unpaired t-test, *P < 0.05, **P < 0.01).
DNA damage? It is important to note that platinum drugs form both intrastrand and interstrand DNA adducts, of which only the more common intrastrand adducts contribute significantly to mutagenesis (34), even though interstrand adducts may be more toxic. However, the ratios of various DNA adducts were found to be similar in the case of cisplatin and oxaliplatin (55). As platinum drugs readily react with proteins as well as nucleic acids, it is likely that their mechanism of cell killing also includes DNA-independent routes. Indeed, one such mechanism has been demonstrated for oxaliplatin, which exerts a cytotoxic effect via inducing ribosome biogenesis stress (56). The platinum agents may also stress cells via further, yet undiscovered mechanisms, as implied by the increased rate of background mutagenesis upon treatment with all three drugs. The differences may be due to the different relative access of the platinum drugs to DNA versus other cellular components. This could be related to their different rates of activation after cellular import (replacement of the chloride or carboxyl moieties with hydroxyl groups) and the different reactivity of the active forms. It is important to note that although cisplatin and carboplatin form the same cis-diammineplatinum crosslink adduct, their active forms and the initial adduct-forming reactions are different, as cisplatin only loses one of its chloride ligands prior to forming the first covalent bond with DNA (57), and carboplatin may react in an analogous sequential manner.
Differences in the mutagenicity of cytotoxic drugs are relevant for treatment-induced secondary cancers. In partial agreement with our results, a recent study that compared platinum drugs in Drosophila found that oxaliplatin was the least mutagenic, and carboplatin induced fewer recombination events than cisplatin (58). Although the results of various model systems cannot be directly applied to clinical use, the carcinogenicity of cisplatin-based therapy, recently reviewed by Liang and colleagues (59), is well documented: there is an increased risk of developing both solid tumours and leukaemia following cisplatin treatment of testicular or ovarian cancer. The appearance of secondary malignancies is a particularly significant issue in the treatment of childhood cancer (60). Our results raise the possibility that the replacement of cisplatin with carboplatin may reduce the mutagenicity-related carcinogenic effect of therapies. Mutagenic treatments may also accelerate the evolution of resistance in the primary tumours, as demonstrated by cisplatin-induced reversion mutations in BRCA2-deficient cancers (34,61), especially if DNA repair deficiencies in the tumours increase the mutagenicity of the treatment (62). The genomic imprint of defective homologous recombination is a positive predictor of platinum treatment in advanced or triple-negative breast cancer (14,63). In these cases, the choice of cisplatin or carboplatin may influence the evolution of resistance. Platinum mutations are generated by translesion DNA synthesis, therefore inhibitors of translesion polymerases could potentially be used to limit the mutagenic effects of chemotherapies (64).
Further clinical relevance of our findings may concern the doselimiting side effects of chemotherapeutic treatments. Platinum treatments show considerable peripheral neurotoxicity, which is much stronger in the case of cisplatin than carboplatin. The main proposed underlying mechanisms are nuclear DNA damage in dorsal root ganglia, or mitochondrial damage and oxidative stress (65). Our results suggest the dominance of the former mechanism based on greater nuclear DNA damage caused by cisplatin treatment. In contrast, the high nephrotoxicity of cisplatin could be independent of its DNA-damaging effect and rather be related to the reactivity of this halogenated compound with cellular thiols (66). It will require further studies to establish whether the indirect mutagenic effect of all three platinum drugs observed in this study is related to oxidative stress, and whether this phenomenon is related to any specific side effects.
In conclusion, this genomic study demonstrated greater mutagenicity of cisplatin compared to carboplatin and oxaliplatin in two cell lines. We showed that platinum drugs exert direct as well as indirect mutagenic effects, of which the direct mutagenic effects correlate with the amount of DNA damage caused by the treatments but not with cytotoxicity. Our results can contribute to a careful appraisal of the benefits versus the short-and long-term side effects of platinum-containing chemotherapeutics to guide therapeutic choices.

Platinum sensitivity assays and long-term treatments for mutagenesis experiments
Sensitivity to cisplatin, carboplatin and oxaliplatin was measured using colony survival assays on DT40 cells and cytotoxicity assays on TK6 cells. Ten thousand cells were treated for 1 h at different drug concentrations. After the treatments, the platinum drug-containing medium was removed. For the colony formation assay, the cells were plated onto 6-well plates in medium containing 1% methylcellulose using a 10-fold dilution series. After 10-12 days, the surviving colonies were counted. For cytotoxicity assays, treated cells were plated onto 96-well plates at a density of 5000 cells/well, and cell survival was measured 72 h later on a Perkin-Elmer EnSpire plate reader using the PrestoBlue cell viability reagent (Invitrogen).
For the mutagenesis experiment, single ancestral clones were isolated from the DT40 and TK6 wild type cell lines and expanded, then four rounds of drug treatments were performed in weekly intervals. In each treatment, 1 million cells were exposed for 1 h. Each treatment series lasted 50 days including the initial period of ancestral clone expansion. Mock-treated cells were handled in parallel with the treated cells without the addition of platinum drug. Following the fourth treatment, single-cell clones were isolated and expanded to 1 million cells prior to genomic DNA preparation using the Gentra Puregene Cell Kit (Qiagen).

Whole genome sequencing, mutation calling and mutation analysis
Library preparation and DNA sequencing was done at Novogene, Beijing, using Illumina HiSeq X Ten and NovaSeq instruments (2 × 150 bp PE) and at BGI, Hong Kong, using 2 × 100 bp BGISeq and 2 × 150 bp DNBSeq technology. A mean sequence coverage of 30× was targeted in all cases (Supplementary Tables S1 and S7). DT40 sequence reads were aligned to the chicken (Gallus gallus) reference genome Galgal4.73; TK6 sequence reads were aligned to the human reference sequence. Duplicate reads were removed and the aligned reads were realigned near indels as described (62).
For detection of the independent SNVs and indels, we used the IsoMut method developed for the detection of unique mutations in multiple isogenic samples (50). Briefly, after running IsoMut with default parameters, detected mutations were filtered using a probability-based quality score calculated from the mutated sample and one other sample with the lowest reference allele frequency. The quality score threshold was set such that no more than five false positive SNVs or one insertion or one deletion would be detected in pre-treatment starting clones. Detailed lists of detected mutations are presented in Supplementary Tables S2-S6, S8-S11. Individual SNV spectra were averaged for each treatment. De novo NMF decomposition and analysis of transcriptional strand bias was performed using the R package MutationalPatterns (67). Custom scripts are available at https://github.com/szutsgroup/platinum_mutagenesis.

Detection of DNA damage markers
For western blot analyses, whole cell extracts were separated by sodium dodecyl sulphate-polyacrylamide gel electrophoresis and transferred to polyvinylidene difluoride membranes. After blocking in Tris-buffered saline (TBS) containing 3% bovine serum albumin for 1 h, the membranes were incubated with anti-phospho-histone H2AX (Ser139), clone JBW301 primary antibody (Millipore, 05-636) in 1:2000 dilution at 4°C overnight, washed, and incubated with anti-mouse IgG-peroxidase secondary antibody (Sigma, A9044) in 1:20 000 dilution for 1 h at room temperature. Chemiluminescent imaging using the Clarity Western ECL Blotting Substrate was performed with a ChemiDoc MP system and the band intensities were quantified using the ImageLab software (all from Bio-Rad Laboratories). Band intensities were normalised to α-tubulin detected on the same membrane with primary antibody T6199 (Sigma) at 1:500 dilution. Before averaging of measurements, the signals were also normalised to the untreated samples.
For immunofluorescence analysis, 1 million treated and recovered cells were pelleted onto poly-l-lysine (Sigma) coated coverslips and fixed with 4% paraformaldehyde in PBS. The samples were blocked with 0.1% Tween 20 and 0.02% SDS (sodium dodecyl sulphate) in phosphate-buffered saline (PBS), then incubated with anti-phosphohistone H2AX primary antibody in 1:1000 dilution and Alexa Fluor 488 anti-mouse secondary antibody (Thermo Fisher Scientific) in 1:1000 dilution for 1 h each at 37°C followed by Hoechst 33342 (Thermo Fisher Scientific) in 1:10 000 dilution at room temperature for 10 min. A Zeiss LSM710 confocal microscope was used to detect the fluorescent signal.

Alkaline single cell gel electrophoresis
For alkaline single cell gel electrophoresis 0.5 million DT40 and TK6 cells were treated with platinum drugs at IC 50 and IC 25 concentrations for 1 h and then allowed to recover for various periods. Samples were prepared using the Trevigen Comet Assay Kit, following the manufacturer's protocols and recommendations. After the alkaline gel electrophoresis, the samples were stained with 10 µg/ ml propidium iodide for 30 min at room temperature and viewed the with a LEICA DM IL LED fluorescence microscope. The CometScore 2.0 software was used for data analysis.
2D cell cycle analysis and flow cytometry DT40 cells were treated with cisplatin, carboplatin or oxaliplatin for 1 h and allowed to recover for 16 or 24 h. Recovered treated cells were labelled with bromodeoxyuridine (BrdU), fixed and prepared for cell cycle analysis as described (68). The samples were analysed using flow cytometry on an Attune NxT flow cytometer (Thermo Fisher Scientific).

Supplementary data
Supplementary data are available at Mutagenesis Online.

Data access
Raw sequence data have been deposited with the European Nucleotide Archive under study accession number PRJEB39852.