With the advent of next-generation sequencing technologies, mutation detection in whole genomes has become possible. Currently, thousands of cancer exomes and genomes have been sequenced in large consortium-based efforts (Bailey et al., 2018; Priestley et al., 2019b; Campbell et al., 2020). The data of these large-scale genome projects have been made available to the biomedical research community and efforts to increase the number of included patients are ongoing. These collections of somatic mutations are providing an unprecedented amount of information about the activity of mutagenic processes before and after carcinogenesis. As cancer is the result of a clonal expansion originating from a single founder cell, all the mutations present in that ancestral cell will be shared by all cells in the tumor (Nik-Zainal et al., 2012b). Depending on the depth of sequencing and the clonal makeup of the tumor, subclonal mutations from the most predominant subclones can be detected as well (Figure 1A) (Miller et al., 2014; Roth et al., 2014). Thus, mutagenic exposure during the lifetime of the ancestral cell or during early carcinogenesis will be captured in the tumor genome (Nik-Zainal et al., 2012b). Mutagenic processes that are active later after tumor initiation and stochastically present in a single or a small subset of tumor cells are not detectable with traditional cancer genome sequencing methods. Sequencing of single cancer cells is required to accurately detect each subclonal mutation (Wang et al., 2014; Roerink et al., 2018). This approach gives a detailed insight in the mutagenic processes active during tumor progression at individual cell resolution. Clonal expansions of cells also take place in vivo during tumor development, such as in the case of tumor relapses, metastases or intratumoral selection pressures (e.g., treatment) favoring the outgrowth of a specific subclone. During these steps, low frequency subclonal mutations in the original tumor can become clonal and detectable by sequencing (Figure 1A) (Priestley et al., 2019a; Li et al., 2019). Both exome and whole genome sequencing can be employed to study cancer genomes. However, whole genome sequencing provides a higher resolution, as it enables the detection of a large number of passenger mutations (Helleday et al., 2014). In addition, whole genome sequencing enables detailed characterization of broad (karyotype), local ploidy changes (loss-off-heterozygosity), and accurate estimates of cancer purity, which is ideally suited for studying mutational spectra and clonal compositions.

Different mutagenic exposures during the lifetime of a cell can be disentangled by identifying recurrent mutational patterns, or “mutational signatures” across cancer genomes. These mutational signatures are defined by individual mutation classes, such as single base substitutions (SBS), double base substitutions (DBS), or short insertions and deletions (indels). For the most common mutation class, SBS, different mutagenic processes can induce specific base changes, such as UV resulting in C > T mutations and smoking in C > A mutations. These base changes are by convention always reported from the perspective of the pyrimidine base, such as C > A, or T > G, because the strand containing the mutagenic damage cannot directly be deduced. In addition to the type of substitution, the causative processes can display a preference for the direct 5’ and 3’ bases flanking the mutated base. Therefore, single base substitutions are usually depicted as trinucleotide changes, in which the middle base is mutated, resulting in 96 different possibilities (Alexandrov L. B. et al., 2013). When various mutagenic processes are differentially active in a set of tumor genomes, they can be extracted from mutation data using dimension reduction algorithms (Alexandrov L. B. et al., 2013). These algorithms, such as non-negative matrix factorization, reduce the mutation spectra of a multitude of individual cancer genomes into a limited set of recurrent 96-trinucleotide signatures, and the contribution of these signatures in each of the cancer genomes (Figure 1C). The higher the diversity in mutation spectra in the set of assessed genomes, for example, by analyzing across various cancer types with different exposures, and the larger the cohort of samples, the more distinct signatures can be extracted. Indeed, by analyzing increasing numbers of cancer genomes, more and more signatures have been defined in the last decade (Nik-Zainal et al., 2012a; Alexandrov L. L. B. et al., 2013; Alexandrov et al., 2020) (Figure 1B). Each of these signatures is regarded to reflect a specific mutational process (Helleday et al., 2014) and new signatures are still being discovered (Pleguezuelos-Manzano et al., 2020; de Kanter et al., 2021). For many signatures the biological cause remains unknown, for other signatures an underlying molecular association has been proposed and for a subset the underlying causative process has been experimentally confirmed. These molecular causes range from endogenous processes that are active in all cells of the body, to exposure to specific chemotherapeutic agents (Pich et al., 2019; Alexandrov et al., 2020). While the first mutational signatures were composed of single-base substitution patterns (SBS), signatures have now also been defined for double base substitutions (DBS), and indels (ID), and are catalogued as part of the COSMIC database (Alexandrov et al., 2020). Future developments in signature extraction are focusing on the integration of additional genomic characteristics, such as specific genomic regions (Vöhringer et al., 2021), tissue-specific signatures (Degasperi et al., 2020) as well as signatures from structural variants (Li et al., 2020). While most of known signatures have been discovered in cancer genomes, the activity of the underlying mutagenic processes is often not limited to tumor cells alone and can be operative in normal cells.

As a tumor genome can serve as an historical archive, it will contain mutations that were acquired before the onset of tumorigenesis in a precancerous normal cell (Helleday et al., 2014). Most cancers, however, are characterized by a mutator phenotype (Loeb et al., 2003), which may be caused by excessive proliferation, loss of specific DNA-repair components, treatment or distorted cellular metabolism. Therefore, it is difficult to exactly determine which mutations were acquired before the malignant transformation (Stratton et al., 2009). Sequencing the genomes of normal cells can be used to identify which mutagenic processes are already active in normal non-malignant cells. However, detection of somatic mutations in bulk normal tissue is difficult due to the stochastic nature of mutation accumulation and the polyclonal architecture of most tissues. In addition, the amount of DNA of a single cell is not sufficient for standard sequencing technologies and needs to be amplified. To achieve this, three methods have been frequently employed. First, in vitro expansion of single stem/progenitor cells into clonal cultures has been used to obtain sufficient DNA of the parental cell (Jager et al., 2018; Rosendahl Huber et al., 2019). This method has been used to determine the mutation accumulation in hematopoietic, intestinal, colorectal, liver, skin, muscle, kidney, and lung cells (Welch et al., 2012; Behjati et al., 2014; Blokzijl et al., 2016; Franco et al., 2018; Lee-Six et al., 2018; Osorio et al., 2018; Franco et al., 2019; Yoshida et al., 2020). While in vitro expansion is a highly accurate approach to amplify genomic DNA, as cells are well equipped to copy their own DNA faithfully, and circumvents the need for specialized library preparation methods, it is limited to cells with sufficient replicative potential, such as stem cells. A second approach is based on the microdissection of naturally occurring clonal structures that exist within normal tissues, such as skin, esophagus, colonic crypts, bladder, and endometrium (Martincorena et al., 2015; Martincorena et al., 2018; Brunner et al., 2019; Lee-Six et al., 2019; Yokoyama et al., 2019; Lawson et al., 2020; Moore et al., 2020; Olafsson et al., 2020; Yoshida et al., 2020; Sanders et al., 2021). The downside of this approach is that it relies on the presence of clonal structures or expansions within the tissue of choice, which are not always present, such as in the brain. In addition, only mutations that are present in all cells of the clonal population are captured, whereas more recent mutagenic events are missed. To determine the somatic mutations present in these cells, whole genome amplification (WGA) using strand displacement polymerases has been employed (Lodato et al., 2015; Gawad et al., 2016; Vijg et al., 2017; Lodato et al., 2018). However, WGA-based methods are notorious for introducing amplification biases, resulting in overrepresentation of random loci, and allelic dropouts (Hou et al., 2012). These biases make WGA-based methods less suited for assessing mutations in samples with a relative low mutation burden, such as normal cells. Nonetheless, new promising methods to directly study somatic mutations in single cells have been developed that address amplification-induced artifacts, such as primary template-directed amplification (PTA) (Gonzalez-Pena et al., 2021). PTA relies on the introduction of exonuclease-resistant amplification terminators, resulting in a more uniform amplification of the genome. In addition, duplex-sequencing, such as Nanorate sequencing (Nanoseq), enables the sequencing of two complimentary DNA strands. By comparing the sequences, errors that arise during sequencing can be corrected for, as these are not shared by the two strands. This approach has been employed to detect somatic mutations in cells with no replicative potential, such as differentiated cells (Abascal et al., 2021).

Using the methods described above, it has been determined that the genomes of healthy cells gradually accumulate mutations during life (Blokzijl et al., 2016; Lee-Six et al., 2018; Osorio et al., 2018; Machado et al., 2021). This rate varies between ∼40 mutations per year in liver, small intestine, and colon, while hematopoietic stem cells and muscle stem cells are on the lower end with about 13–15 novel mutations each year (Franco et al., 2018; Lee-Six et al., 2018; Osorio et al., 2018). A more pronounced variation has been observed for the mutation spectra and signatures across different organs, suggesting a tissue-specific activity of mutagenic processes. Of these, some processes are active in a clock-like manner in most tissues causing mutation accumulation at a rate that is constant within a tissue (Alexandrov et al., 2015). One of these processes is reflected by signature SBS1, which is induced by spontaneous deamination of methylated cytosines, present in a CpG dinucleotide context, into thymine residues. SBS1 seems more predominant in genomes of fast-dividing cells, such as colon and intestine as well as tumors originating from these tissues (Alexandrov et al., 2015; Blokzijl et al., 2016; Lee-Six et al., 2019). Another clock-like signature, SBS5, is observed in practically all tissues (Alexandrov et al., 2015). Finally, hematopoietic stem and progenitor cells display a unique aging-related signature, termed HSPC signature (Maura et al., 2019; Brandsma et al., 2021). The cause for these latter two signatures remains unknown, but their continuous action in normal cells suggests a role for endogenous mutagenic sources, which are constantly present. In addition to these clock-like processes, the mutational consequences of a variety of environmental mutagenic processes can sometimes be observed in normal cells. For example, in colonic crypts, approximately half of all studied individuals displayed a specific mutational signature, which is characterized by T > N single base substitutions in an ANNT context (of which the underlined base is mutated) and deletions of a single thymidine in short T-homopolymers preceded by adenine (Lee-Six et al., 2019). This signature is caused by exposure to colibactin, a mutagenic toxin produced by pks + E. coli (Pleguezuelos-Manzano et al., 2020). These colibactin-induced mutations are shared in phylogenetically related crypts, indicating that these mutations have been induced early during life (Lee-Six et al., 2019). The mutational consequences of pks + E. coli can also in rare cases be observed in the genomes of bladder, neuroendocrine and head, and neck cancers. The presence of this mutational signature in these cancers is likely indicative for a history of colibactin exposure, which potentially increased the risk for developing cancer as a direct result of enhanced mutation accumulation (Boot et al., 2020; Pleguezuelos-Manzano et al., 2020). Another example of variable damage induced by environmental exposure is in the skin, where UV-induced mutations can be observed in both melanoma and healthy melanocytes (Tang et al., 2020). In skin not exposed to sunlight, melanocytes harbor a low number of UV-induced mutations. Interestingly, compared to melanocytes present in skin continuously exposed to UV-light, melanocytes from intermittently UV-exposed skin contain a higher number of UV-induced mutations (Tang et al., 2020). This observation suggests that the number of genotoxin-induced mutations does not necessarily has to correlate with the dose of exposure. In addition to melanocytes, the mutational signatures attributed to UV-damage have also been observed in skin-residing lymphocytes (Machado et al., 2021) and sporadically in T-cell lymphoma (Jones et al., 2021), pointing to a past exposure to UV-radiation.

Exposure to genotoxins in cigarette smoke can induce mutations in a patchy manner, affecting some cells but not others, as was recently demonstrated by the sequencing of bronchial epithelial cells in smokers, non-smokers, and ex-smokers (Yoshida et al., 2020). As expected, the non-smokers accumulated aging-related mutations at a constant rate with little variance amongst each assessed cell within the same donor. In contrast, this variance was increased in smokers as well as ex-smokers. Most bronchial epithelial cells displayed a several-fold increase in mutation load, which could be attributed to the tobacco smoke-associated signature SBS4. However, some cells in the lungs of smokers and ex-smokers had no additional mutation load, implying that some bronchial epithelial cells were not or less exposed to the mutagen. Interestingly, these cells with a near-normal mutation load were present in higher numbers in ex-smokers, which may explain why after years of quitting smoking the risk of developing lung cancer is reduced (Yoshida et al., 2020).

The mutagenic properties of environmental components can also be experimentally determined using cell culture assays. In these experiments, primary cells, cell lines or organoids are exposed in vitro to mutagens, followed by clonal expansion and sequencing analysis (Jager et al., 2018). During culture, cells experience both background mutagenesis, which seems to be driven by oxidative stress (Kuijk et al., 2020), and mutation accumulation as a result of the genotoxic exposure. Such in vitro experiments can confirm associations between a mutational signature and hypothesized mutagenic exposure (Kucab et al., 2019). Recently, a landmark study has examined the mutational impact of 79 different environmental genotoxins, providing a resource to causally link mutational signatures to specific mutagenic exposures (Kucab et al., 2019). One such carcinogen is aflatoxin B1, which is produced by strains of the fungus Aspergillus that grows on contaminated food. This carcinogen is a known mutagen, which induces SBS24 mutations as determined by signature extraction using cancer genomes (Alexandrov L. L. B. et al., 2013) and confirmed by in vitro mutagenicity assays (Huang et al., 2017; Kucab et al., 2019). Another carcinogenic compound, aristocholic acid, causes signature SBS22, and has been implicated with bladder and liver cancer (Poon et al., 2015; Ng et al., 2017). Also this association between signature and environmental genotoxin has indeed been confirmed in vitro (Nik-Zainal et al., 2015). While aflatoxin B1 and aristocholic acid directly alkylate the DNA, several indirect mutagenic processes have been identified. External exposures can result in unsuccessful apoptosis (Ichim et al., 2015; Hawkins and Miles, 2021) or altered expression of DNA replication and repair enzymes, elevating mutational loads (Russo et al., 2019; Cipponi et al., 2020). As the mutagenicity of these processes has been determined using colony outgrowth assays or reporter gene assays, mutational signatures from these processes are lacking, and could be established in the future using WGS-based methods.

One of the major conformation-changing processes that occurs in DNA is its transcription into RNA. This process requires a temporary separation of the two DNA strands to synthesize a complementary RNA molecule. When RNA polymerase II (RNA Pol II) encounters a blocking DNA lesion, it cannot proceed and will stall (Vermeulen and Fousteri, 2013). To continue transcription, RNA Pol II recruits transcription-coupled nucleotide excision repair (TC-NER), which initiates the repair of the blocking lesion. As this process is dependent on transcription, TC-NER can only take place in actively transcribed regions of the genome. Importantly, only the transcribed (template) strand will repaired be by TC-NER. This preferential repair of the transcribed strand results in a characteristic mutational strand asymmetry in expressed regions (Green et al., 2003; Haradhvala et al., 2016) (Figure 2A). The bias of the mutational strand asymmetry (i.e., which base is on the transcribed strand and thus protected) can indicate which nucleotide of the mutated base pair was originally damaged. For instance, the mutational signature induced by smoking, SBS4, displays a depletion of C > A mutations when the guanine of that mutated base pair was on the transcribed strand. This bias indicates that damaged guanine residues underlie the signature as these are preferentially repaired on the transcribed strand. Indeed, mutagenic agents in tobacco smoke, such as benzo(a)pyrene, are known to react with guanine and thereby damaging the DNA. Thus, the presence of a transcriptional strand bias in a specific signature can give clues into the DNA damage that cause specific signatures.

Recent studies implicate that the replicative state of a cell is an important determinant for incorporating chemotherapy-induced DNA damage (Pich et al., 2021). According to this model, the turnover of damaged nucleotides is so quickly, that only cells in a replicative state during chemotherapy exposure will incorporate damaged nucleotides during DNA-replication. After DNA replication, damaged nucleotides are converted in mutations. This mechanism might be important in more mutagenic exposures where the damaged nucleotides are concerned, such as in thiopurine treatment, which also can induce a mutational signature (Li et al., 2019). In the future it might be feasible to take this mechanism into account during chemotherapy treatment to spare healthy replicating cells.

When DNA-lesions encounter a replication fork, these lesions can cause mispairing (Boiteux and Jinks-Robertson, 2013). The mutagenicity of such lesions can depend on the activity of different polymerases on the leading and lagging strands (Seplyarskiy et al., 2018). DNA replication can result in asymmetric distribution of mutations between Watson and Crick strands after short mutagenic exposures. During DNA replication, mispairing opposite damaged bases on the parental DNA strand can result in mutations in the newly synthesized strand of DNA in a strand-specific manner (Figure 2B). This strand lesion segregation was first observed in mice receiving the highly mutagenic agent diethylnitrosamine (DEN), which induces liver cancer (Aitken et al., 2020). As the exposure to DEN was a single short burst, a clear separation of mutations in a Watson-Crick strand asymmetry could be observed in the genomes of the liver tumors. It is supposed that such a mechanism is involved during every DNA replication when damaged bases result in mispairing during DNA synthesis. However, a clear strand-specific separation is generally not visible in the genomes of tumor and normal cells, as exposure to a mutagenic source across multiple cell divisions will result in mutation accumulation on alternating Watson and Crick strands (Aitken et al., 2020). This prolonged exposure will result in DNA damage on both Watson and Crick strands, and therefore in most cancer genomes this mutational asymmetry is not observed. However, strand lesion segregation is observed in renal, liver and biliary tract-tumors containing aristocholic acid or aflatoxin exposure (Aitken et al., 2020). This segregation therefore indicates that the exposure to these environmental genotoxins was limited to a single replication round.

Ionizing radiation particles have the potential to induce DNA double strand breaks (DSBs) in the genomes of affected cells. These DSBs, when repaired incorrectly, can result in short insertions or deletions, and larger chromosomal aberrations. Indeed, in radiation exposed cells, mice and tumors indel mutations and larger structural variants displaying the hallmarks of erroneous DSB repair can become visible (Behjati et al., 2016; Rose Li et al., 2020; Kocakavuk et al., 2021). While there is no single hallmark of radiation-induced cancers, and mutation induction may vary on the specific type of radiation (Rose Li et al., 2020), detection of radiation-induced mutations post-treatment could be used to determine sensitivity to radiation in the future (Kocakavuk et al., 2021).

Different chromatin states, gene expression, timing of DNA replication during the S-phase and differential DNA repair across the genome can result in regional differences in the density and types of mutations (Makova and Hardison, 2015; Supek and Lehner, 2015). These regional mutation densities are so specific that the distribution of mutations across cancer genomes can be used to predict the tissue type of origin of tumors (Jiao et al., 2020). In addition, the three-dimensional conformation of the genome influences regional mutation rates at boundaries between different genomic topologically associating domains (Akdemir et al., 2020). In addition, steric hindrance of DNA-binding molecules can impair the repair of DNA damage at specific DNA loci. In melanomas, it has become apparent that transcription factor-binding sites are enriched for mutations, as nucleotide excision repair is not able to access these sites resulting in mutations enriched at transcription-factor binding motifs (Perera et al., 2016; Sabarinathan et al., 2016).

Some mutagenic processes display a clear preference for a specific context beyond the direct 5’ and 3’ flanking bases. Assessing an additional base on each side of the mutation results in a 1,536 different pentanucleotide categories. While it is technically possible to delineate signatures using pentanucleotide changes, spreading mutations across a much larger number of categories results in much sparser data, complicating analyses (Alexandrov L. B. et al., 2013). Moreover, nucleotides even further away from the mutated base may be specifically enriched at mutated bases, suggesting a mechanistic cause. Such an extended context has been demonstrated for mutations induced by the cytidine deaminase enzyme APOBEC. These mutations are enriched at DNA hairpin sites, indicating that these secondary structures provide an optimal substrate for APOBEC enzymes (Buisson et al., 2019; Langenbucher et al., 2021). In addition, the mutations induced by colibactin-producing pks ⁺ E. coli, which are mainly T > N SBS mutations display a striking enrichment for adenines at the -3 position. As colibactin is known to preferentially bind to adenine (Wernke et al., 2020), the broader sequence context suggests that colibactin causes a cross-link between the -3 positioned adenine and the adenine opposite the mutated base (Boot et al., 2020; Pleguezuelos-Manzano et al., 2020) (Figure 2C).

Environmental genotoxins have the potential to induce many somatic mutations, often recognizable by the mutational signatures they leave behind in the genomes of cells (Kucab et al., 2019). However, the presence of such a mutational signature does not necessarily mean that the exposure contributed to carcinogenesis. As mutations in specific driver genes are required for carcinogenesis, evidence of driver gene induction by environmental mutagens could be used to obtain additional evidence to link environmental genotoxins to the initiation of cancer. The specific type and context characteristics of driver mutations can be used to attribute cancer driver mutations to specific mutational signatures (Poulos et al., 2018; Temko et al., 2018). In addition, the topography of mutations can provide further evidence for the involvement of specific genotoxins in causing the oncogenic mutations driving cancer. In the colorectal cancer-driving APC gene, 5.3% of the mutations display the extended motif characteristic for colibactin-induced mutations (SBS88/ID18), implying a causative role for pks ⁺ E. coli in the induction of colorectal cancer (Pleguezuelos-Manzano et al., 2020; Terlouw et al., 2020). However, the analysis of all APC mutations in a large colorectal cancer cohort indicates that most mutations are C > T substitutions at CpG sites, pointing to a major role for the endogenous clock-like deamination of methylated cytosines (SBS1) in the induction of cancer driving mutations (Blokzijl et al., 2016). However, even if a relatively large fraction of driver mutations is induced by endogenous processes, a tumor harbors typically between 2 and 10 cancer driver mutations (Martincorena et al., 2017). Additional driver mutations can be induced by environmental genotoxins on top of the driver mutations caused by endogenous processes, which may be sufficient to induce full malignancy. In this scenario, environmental carcinogens may be responsible for the induction of a larger fraction of cancers than suggested by the fraction of driver mutations linked to these carcinogens (Tomasetti et al., 2017; Volkova et al., 2020).

Despite the well-established role of driver mutations in cancer, recent sequencing studies have shown that normal tissues can harbor many cancer driver mutations without being malignant. In 83% of skin naevi (common moles), oncogenic BRAF V600E mutations have been found (Pollock et al., 2003). While melanoma can originate from a naevus, most naevi never progress to melanoma. Moreover, in the skin of eyelids of elderly individuals between 55 and 73 years of age, 18–32% of small clonal populations of skin cells can contain mutations in classical cancer driver genes, such as NOTCH1 and TP53. These mutations might underlie the clonal expansion of these driver-containing cells, but the tissue is phenotypically normal. Deep-sequencing of esophageal epithelium revealed a similar presence of clonal expansions, which contained cancer driver mutations in the same genes as found in normal skin (Martincorena et al., 2015; Martincorena et al., 2018; Yokoyama et al., 2019). While some of these expansions can be detected in young individuals, their frequency and size are increased in older individuals. Despite the presence of these driver mutations in up to 30% of the cells, the tissue still functions normally and appears nonmalignant. A similar clonal expansion of cells has been observed in the hematopoietic system, named clonal hematopoiesis (CH). The incidence of CH is higher in elderly individuals and was initially detected by somatic mutations in DNMT3A, ASXL1, and TET2 (Genovese et al., 2014; Jaiswal et al., 2014). These genes are frequently mutated in leukemia and are considered to have a leukemic driving potential (Martínez-Jiménez et al., 2020). Later studies discovered CH without driver mutations is highly prevalent (Zink et al., 2017), raising the question whether CH is a pre-cancerous state. These observations raise the question whether these clonal expansions represent pre-cancer states, or if the enhanced clonality of a tissue is a characteristic of normal aging (Brash, 2015; Colom et al., 2021).