Long-Term Latency of Highly Mutated Cells in Normal Mouse Skin Is Reversed by Exposure to Tumor Promoters and Chronic Tissue Damage Open Access

Cancer is clearly a genetic disease, as demonstrated by the complexity of cancer genomes that can carry hundreds of mutations in oncogenes and tumor suppressor genes, collectively known as cancer driver genes. These mutations are known to be caused by exposure to environmental mutagens or may result from spontaneous errors in DNA replication (1). Although the causative role of many mutations in driving cancer growth has been well established by functional studies, recent deep sequencing analysis of completely normal human tissues has uncovered an abundance of cells carrying many of the same mutations that are found in human cancers. Affected tissues include skin (2), esophagus (3, 4), colon (5), liver (6), endometrium (7), and lung (8). These provocative observations have raised fundamental questions about the mechanisms that cause clonal expansion of mutated cells in normal tissues and why they rarely progress to actively growing cancers (9).

These recent observations on human tissues, while surprising, could have been predicted based on seminal studies carried out using mouse cancer models in the 1940s. Mottram, Berenblum, and others showed that environmental chemicals linked to cancer fall into two broad categories capable of either initiating cancer or promoting preinitiated cancer cells to grow and form visible lesions [reviewed in ref. 10]. Importantly, Berenblum and Shubik showed that initiation (at that time presumed to be a mutational event) was essentially permanent and did not cause any tumors to develop over long periods of time (11). The promotion phase, induced by chronic exposure to agents that cause inflammation and tissue damage, was proposed to be the rate-limiting causative step in cancer development. These observations were replicated in mouse models decades later (12), but the nature of the event(s) induced by chemical initiation, including the number of mutations induced by the mutagen and the demonstration of their persistence over time in normal tissue, was never verified using molecular approaches. The importance of this question of mutation persistence after chemical initiation has recently been emphasized by studies using genetically engineered mouse models (GEMM). Although previous studies using GEMMs to initiate carcinogenesis by activation of strong Ras mutations showed a strong dependence on wounding or promoter treatment to stimulate cancer growth (13, 14), others have shown that normal cells in GEMMs can suppress the outgrowth of their Ras mutant counterparts, even under conditions of wound healing (15, 16). These disparate results emphasize the importance of examining the nature and persistence of initiated cells in models of carcinogenesis that do not require genetic manipulation and are dependent on exposure to environmental factors similar to those implicated in cancer development in humans.

Notably, recent whole-genome sequencing (WGS) of mouse tumors induced by exposure to known or suspected human carcinogens suggested that most do not exert their effects through a mutagenic mechanism and are therefore more likely to act through promotional mechanisms (17). Similarly, sequencing of hundreds of human esophageal carcinomas from countries with varying incidence rates failed to reveal a mutational signature that may indicate the presence of a specific causative mutagen responsible for increased susceptibility (18). These observations highlight the possibility that a fundamental step in cancer development is exposure to nonmutagenic environmental or lifestyle factors that stimulate the clonal selection of preexisting mutant cells that otherwise may lie latent in normal tissues.

In view of the potential importance of these findings for cancer prevention, we replicated the original studies of Berenblum and Shubik, showing that although initiation is permanent and necessary for cancer development, promotion is, in fact, the rate-limiting step. We carried out WGS of tumors induced by the standard model (initiation followed by promotion after 1 week) or by the delay model, in which promotion is postponed for many months following the initiation event. We also extended the original model in several ways by administering the initiating agent during embryonic development, thus exposing the initiated cells to many rounds of cell division during fetal and adult growth, and examining the potential impact of other promoting factors on mutational burden, including obesity, high-fat diet, and tissue wounding. In addition, we carried out WGS of tumors initiated genetically by Cre-mediated oncogene activation in stem cells, followed by exposure to tumor promoters or tissue wounding. Taken together, our data allow us to conclude that a single mutation in the appropriate cancer driver gene is sufficient for initiation and that cells carrying tens of thousands of carcinogen-induced mutations survive for long periods of time in normal tissue and are not outcompeted by normal cells or removed by the immune system. Repeated cycles of normal cell division during fetal development and growth to the adult stage do not result in activation of promotion, and exposure to agents that cause inflammation and tissue damage is required. Whole-genome sequence analysis of this diverse panel of tumors also confirmed that promotional factors such as 12-O-tetradecanoylphorbol-13-acetate (TPA) or wounding do not cause any significant changes in single-nucleotide variant (SNV) mutation burden, which are likely to contribute to their roles in causing clonal selection and outgrowth. Our data suggest that normal cells carrying oncogenic mutations are very stable and lie latent in tissues until exposed to tissue damage, leading to clonal selection, and support proposals to base cancer prevention strategies on the inhibition of mechanisms of promotion.

We first sought to determine the number of mutations that are directly induced in normal mouse tissue by a single initiating dose of the mutagen 7,12-dimethylbenz(a)anthracene (DMBA) and assess their persistence over time and in response to various factors known to promote carcinogenesis in this model. We carried out mutational signature analysis by WGS of 107 squamous papillomas or carcinomas from four different tumor models as shown in Fig. 1A–D. In three of these models, we used DMBA to initiate tumorigenesis by causing a specific A>T mutation at codon 61 of the Hras gene (19). In contrast to previous studies of long-term chronic carcinogen exposure (17), the application of a single dose of the carcinogen allows us to be confident of the timing of the initiating mutation. Any additional mutations induced by subsequent treatment of the animals with the tumor promoter TPA, as a consequence of dietary factors or obesity (model B) or cell growth during fetal development (model D), should not show the specific DMBA signature, which is dependent on the formation of bulky adducts with adenosine residues (20). In the fourth model, Hras or Kras alleles were activated genetically using Cre recombinase under the control of Lgr6 or Lgr5 stem cell promoters (model C), in the complete absence of any mutagenic initiator (21). Promotion was carried out by exposure either to TPA or to full-thickness wounding, in order to determine whether TPA induces specific mutations different from those occurring spontaneously during normal wound healing.

The total genome-wide mutation burden seen in tumors from these models varies over three orders of magnitude with total SNV counts ranging from 20 to 48,425 and the majority of mutations being SNVs mapping to intergenic or intronic regions (Fig. 1E–H). The lowest mutation burden is seen in skin tumors initiated genetically by the activation of Hras or Kras using Cre recombinase under the control of the promoters of the stem cell marker genes Lgr5 or Lgr6 (21). As previously demonstrated for genetically induced tumors of the lung (22), skin tumors initiated genetically had a very low overall mutation load (average 242.6 as compared with an average of 6,963 in DMBA-induced tumors; Fig. 1E and F; Supplementary Table S1).

As compared with WGS analysis of 188 tumors induced by chronic exposure to 20 different environmental carcinogens, the total mutation burden in DMBA-initiated tumors is more variable (Supplementary Fig. S1A and S1B; ref. 17). The mutation load in the single low-dose DMBA samples is comparable with or exceeds the total mutation burden induced by chronic exposure to the most highly mutagenic environmental carcinogens tested, for example, trichloropropane or vinylidene chloride. However, some DMBA samples had a very low mutation burden comparable with tumors induced by most other agents for which no mutation signatures were detected.

We next carried out a detailed analysis of the mutational signatures found in skin tumors from the various models (Fig. 1) using SigProfiler (see “Methods”; ref. 23). In total, we extracted four de novo signatures, which decomposed into nine known human COSMIC single base substitution (SBS) mutational signatures (23) in addition to a mutational signature for DMBA (Fig. 2A–C). The dominant de novo mutational signature in all samples initiated using DMBA was initially decomposed into a combination of human SBS22 and SBS25. The former is associated with aristolochic acid exposure in several human solid tumor types (24). Both DMBA and aristolochic acid form adducts with adenosine residues in DNA, resulting in A>T transversion mutations (24, 25).

To identify specific differences between the DMBA signature and SBS22/SBS25, we examined the pentanucleotide sequence rather than the trinucleotide sequence context of these mutations (Supplementary Fig. S2A–S2C; see “Methods”). This analysis showed that these signatures can be distinguished at the pentanucleotide level, and we therefore identified SBS. DMBA as a specific signature of exposure to DMBA (Fig. 2C). This mutational signature identifies mutations induced at the specific time point of DMBA exposure, in contrast to mutations attributed to other endogenous processes such as proliferation or inflammation, which are much more likely to occur during subsequent tumor promotion or progression. Interestingly, this analysis also detected a double-base substitution signature (DBS.DMBA) that was only detected in samples initiated using this carcinogen (Fig. 1; Supplementary Fig. S3A and S3B). We did not observe a clear correlation between any of the insertion–deletion signatures identified with DMBA exposure (Supplementary Fig. S3C and S3D).

Several other mutational signatures were identified in mouse tumors, including the clock-like signatures SBS1, SBS5, and SBS40, as well as additional signatures that have been attributed to the activity of reactive oxygen species (ROS) or chronic inflammation (SBS18, SBS36, SBS17a, and SBS17b; refs. 26, 27). However, no additional novel signatures were identified that could be attributed to the interventions shown in Fig. 1. To investigate the possibility that some mutational signatures could be indirectly caused by DMBA, for example, as a response to stress induced by carcinogen exposure, we carried out an analysis of the relationship between the total number of DMBA-induced mutations in each sample and the numbers of mutations associated with other signatures. Surprisingly, this analysis showed a very strong correlation between the total number of SBS. DMBA mutations and those attributed to the “clock-like” signature SBS5 (Fig. 2D, Spearman’s r = 0.74; P < 2.0e–19). However, the same observation does not extend to the clock signatures SBS1 and SBS40 (Spearman’s rho = 0.07 and 0.19, respectively).

As SBS5 is almost ubiquitous in human cancers, this association cannot be specifically due to the misrepair of DNA adducts induced by the carcinogen DMBA but may be a consequence of more generalized genomic instability that is proportional to the total levels of damage induced by DMBA, resulting in the activation of fragile sites in the genome. Notably, carcinogen-induced DNA damage can lead to genomic changes or aberrant transcripts in the fragile site gene Fhit, and germline deletion of this gene in mice leads to the development of tumors with high levels of the SBS5 mutational signature (28). An association between total DNA damage, as monitored by mutational burden from WGS analysis, and the levels of SBS5 was also seen in human tumors from The Cancer Genome Atlas and in mouse tumors induced by chronic exposure to a range of carcinogenic chemicals (Supplementary Table S2; ref. 17). We conclude that in addition to a specific mutational signature attributable to DMBA-DNA adduct formation and misrepair, a more general DNA damage response may result in additional mutations with the SBS5 mutational signature.

An important endogenous source of mutations in human tumor genomes is ROS, a broad category of highly reactive oxygen free radicals, including peroxides, superoxide, hydroxyl radical, singlet oxygen, and alpha-oxygen (29). ROS signatures SBS18 and SBS36 show strong cosine similarity and both demonstrate a strong predominance of G-to-T transversions. We noted that on iterative runs of analysis, depending on which samples are included, de novo observed mutational signatures can be decomposed into either SBS18 or SBS36 and thus seem to be computationally interchangeable. We therefore designated mutations assigned to either SBS18 or SBS36 as SBS.ROS. Correlation analysis across all samples for ROS signatures SBS17a, SBS17b, SBS18, and SBS36 shows that as expected, SBS17a and SBS17b are strongly correlated (r = 0.87, P < 1.33e–33). More importantly, we identified a strong correlation between the sum of mutations assigned to SBS17a and SBS17b with the sum of mutations assigned to SBS.ROS (Spearman’s correlation r = 0.66, P < 1.86e–14; Fig. 2E; Supplementary Fig. S4A). This suggests that the induction of DNA damage by ROS stimulates several independent processes leading to DNA repair and residual DNA changes, resulting in the induction of distinct ROS-associated mutational signatures.

Figure 1E and F show that SBS.ROS, as well as SBS17a and SBS17b, was present at variable levels in many samples investigated, mostly in noncoding regions (Fig. 1G). In the embryonic treatment cohort, every tumor sample exhibited a significant mutation contribution from SBS.ROS, SBS17a, and/or SBS17b (P < 0.0001, Fisher’s exact test; Supplementary Fig. S4B and S4C). We hypothesize that in utero exposure to systemic DMBA at a stage of development when the skin of the pups is highly proliferative may result in different patterns of DNA damage. It is possible that the induction of a DNA damage response in skin during active proliferation, in contrast to 8-week-old adult mouse skin, which is largely quiescent, results in a greater degree of oxidative damage, but further studies would be required to confirm this association. We also observed SBS.ROS in 50% of DMBA/TPA–induced tumors subsequently treated with cisplatin, consistent with the known potential of cisplatin to induce a cytotoxic oxidative stress response (Fig. 1H; Supplementary Fig. S5A–S5C).

Finally, we considered the possibility that a high DMBA mutation load could potentially confound the detection of ROS signatures, particularly as DMBA may also induce G>T mutations that mask SBS18 and SBS36. We repeated the signature extraction after the removal of 90% of T>A mutations, scaled by the relative trinucleotide frequencies observed in the DMBA signature (see “Methods”). We found that the likelihood of detecting SBS.ROS mutations in tumors from embryonic cohort mice was not affected by the removal of DMBA mutations although SBS.ROS mutations were more consistently detected in the adult DMBA/TPA treatment groups using this method (Supplementary Fig. S5D).

Mice in which Hras or Kras mutations were initiated genetically rather than by chemical mutagens [Fig. 1 (Model C)] also developed aggressive tumors, but only after chronic exposure to TPA or in response to full-thickness skin wounding (Supplementary Fig. S6A). The combination of activated Kras and wounding was particularly potent, resulting in large skin tumors, predominantly at the sites where wounds were clipped, that required the mice to be euthanized after 6 to 8 weeks. No obvious differences were found by WGS analysis of tumors resulting from Hras or Kras mutations or that were related to the cell of origin (Lgr5- or Lgr6-positive stem cells; Fig. 1F; Supplementary Fig. S6B). No tumors developed in these animals by activation of Ras mutations with no further treatment for at least 5 to 6 months, as previously shown after the delivery of mutant Ras by retroviral oncogene delivery to the back skin of mice (30). Notably, genetically initiated tumors did not carry any novel mutational signatures that could be attributed to the activity of either TPA or wounding, and all carried the endogenous signatures SBS1, SBS5, and/or SBS40, with variable contributions from the ROS signatures SBS.ROS, SBS17a, or SBS17b (Fig. 1F; Supplementary Fig. S6). In addition, very few large-scale translocations were observed in these samples (Supplementary Figs. S6C and S7).

Analysis of mutational burden and signatures in tumors induced using model B, in which initiation was followed by a long delay before starting tumor promotion using TPA, demonstrated that initiated cells persist long term in the skin without causing any obvious pathology, as previously suggested by early studies on the rate-limiting role of promotion in multistage carcinogenesis(11). Importantly, our data show that DMBA-initiated cells carrying thousands of DMBA-signature mutations, as well as those carrying a single genetically induced Ras mutation, were persistent in the skin but remained sensitive to the effects of tumor promoters.

We performed WGS of squamous cell carcinomas (SCC) associated with genetically determined high or low body mass index (BMI) (31) or diet-induced obesity (32, 33). We have previously shown that interspecific backcross mice [(SPRET/Ei × FVB) × FVB] show a wide distribution in body weight and weight/length ratio (henceforth high and low BMI groups; Fig. 3A). Mice of the high BMI group, in particular, male mice of this cohort, develop earlier onset and a higher incidence of skin tumors compared with low BMI mice of the same backcross population (Fig. 3B; ref. 31). To complement this model of genetically induced high or low BMI, we also investigated a model of environmentally induced obesity by sequencing tumors induced in mice that were fed a standard, high-fat, or calorically restricted diet (32, 33). Although we detected a wide range of mutation burdens in these tumors, no significant differences in mutation load or mutation signatures were observed (Fig. 3C) although there was a trend for higher mutation loads in the tumors from the low compared with the high BMI group (Student T test; P = 0.11).

As no significant differences were observed in total mutation burden or mutational signatures attributed to DMBA or ROS among animals in either the genetically determined or dietary obesity cohorts, we speculated that obesity or a high-fat diet may cause changes in gene expression, resulting in increased tumor growth capacity. To test this possibility, we carried out microarray analysis of gene expression in squamous cell skin tumors in the high (n = 36) versus low (n = 41) BMI groups. We found significant enrichment for pathways such as antigen processing, MHC peptide antigen presentation, and positive regulation of T cell–mediated cytotoxicity in low BMI carcinomas (Fig. 3D). In contrast, in the high BMI tumors, we observed enriched expression of genes in the T-cell anergy pathway and regulation of T-cell tolerance induction (see Supplementary Dataset File for a list of differentially enriched genes). Future single-cell transcriptome studies will be necessary to tease apart whether the expression identified here is due to intrinsic tumor cell expression or tumor-infiltrating immune cells.

We observed a broad range of mutations that mapped to driver genes in mouse SCCs from this cohort, which have been previously reported as recurrently mutated in human cancers (Fig. 4A; Supplementary Fig. S8A–S8D; refs. 34, 35). To examine the contribution of specific mutational processes to driver mutations, we examined the probability that a given driver mutation could be attributed to a specific mutational signature (Fig. 4B; see “Methods”). We observed that driver mutations in a variety of genes from chemically induced tumors have incurred DMBA-associated A>T mutations at the same time as Hras during initiation; notably, no such driver mutations in previously reported human oncogenes were detected in tumor samples that were not initiated by DMBA (Supplementary Fig. S6). The A>T mutations were seen in genes encoding membrane proteins Fat1-4, tumor suppressors involved in cell–cell communication, and Hippo/YAP signaling (36), as well as Card11, which is involved in Nfkb signaling and immune responses (37). These genes, as well as many of the other genes in Fig. 4A, including epigenetic regulators Kmt2a-d, are commonly mutated in human squamous tumors of the skin, esophagus, or head and neck (38, 39).

Although most recurrent driver mutations display the DMBA mutational signature, some tumors had driver mutations that were associated with other detected signatures, including SBS5, SBS1, SBS18, or SBS36 (Fig. 4B). Some of these mutations may be attributed to endogenous processes activated at the time of initiation (e.g., SBS5; see “Discussion” above) or, alternatively, by ROS generation, leading to mutations with SBS18 or SBS36 signatures during the promotion or progression phases of carcinogenesis. Gene mutations associated with these other processes were found in Trp53, Notch1, Tert, and Tgfbr2, all of which are also commonly mutated in human squamous tumors (40).

We next compared the driver mutations observed in the initiation/promotion models of carcinogenesis in the skin to the solid tumors induced in different tissues by chronic low-dose exposure to suspected human carcinogens we previously examined (17). In contrast to the uniform patterns of initiating mutations seen in the two-step DMBA/TPA model, which were clearly attributable to the mutational signature induced by DMBA, tumors induced in the forestomach by trichloropropane or in the kidney by vinylidene chloride did not have a high frequency of driver mutations specifically attributable to the mutational signatures directly caused by exposure to these agents (17). Closer inspection of these data however showed that although the specific mutations were not obviously attributable to the direct mutagenic action of each carcinogen, there was clearly carcinogen-specific selection of driver mutations by different exposures to the various nonmutagenic carcinogens (Supplementary Fig. S9; see also “Discussion”).

We have demonstrated here that the number of somatic mutations induced in normal mouse skin by a single treatment with the mutagen DMBA, either during embryonic development or in adults, varies over two to three orders of magnitude but is clearly insufficient for tumor formation in the absence of a promoting stimulus or tissue damage by TPA. This situation mirrors the observations of abundant mutations in normal human tissues that do not cause any obvious pathologic changes (2–4). In the mouse, cells carrying thousands of SBS.DMBA signature mutations persist over long periods of time in normal tissue but only give rise to actively growing lesions after exposure to 1 to 2 months of treatment (41) with the promoting agent (Fig. 1). DMBA-initiated cells in normal skin clearly have enough mutations, and in the correct combinations, for eventual papilloma development but have an absolute requirement for the promotion stage.

Our analysis of whole-genome sequences of squamous tumors initiated by activation of mutant Ras indicates that the contribution of promotion processes, including treatment with TPA, wounding and chronic inflammation, obesity, or a high-fat diet, to total mutation burden is minimal. Normal cell proliferation, for example, as seen during fetal development, neonatal growth, and adult homeostasis, does not seem to be sufficient for the completion of the promotion phase, as mice exposed to DMBA initiation in utero do not develop papillomas even after the large number of cell divisions required for fetal and adult growth. Our data support the conclusion that promotion requires a regenerative type of cell division, such as that seen in the adult skin during wound healing or growth in response to repetitive tissue damage by TPA.

Although the majority of point mutations found by WGS analysis of skin tumors initiated with DMBA are attributable to the direct action of the initiating mutagen DMBA, some rare mutations were found in established driver genes such as the Trp53 tumor suppressor gene or in transforming growth factor beta receptor 2 (Tgfbr2) that did not carry the SBS.DMBA signature but rather had mutations more likely attributable to clock-like signatures or SBS.ROS (Fig. 4B). Trp53 mutations are known to occur in the late stages of tumor development in this model (42), and transforming growth factor beta (TGFβ) signaling is also important for late-stage tumor invasion (43). Notably, all of the tumors that had mutations in Tgfbr2 were from mice subjected to caloric restriction or that had a low BMI in the genetic BMI model (P < 0.0001, Fisher’s exact test; Fig. 1C). TGFβ signaling plays an important role in insulin signaling (44); thus, it is possible that metabolic restriction may have resulted in the selection of cells carrying mutations in this pathway (45). However, further studies, including both genetic and functional approaches, would be necessary to establish a clear link between TGFβ signaling and tumor development in obese mice. We conclude that although the overall contribution of promotional processes to mutation burden is very low, some of the rare mutations that arise during promotion may be positively selected and could facilitate later stages of malignant progression.

The lack of a major contribution to SNV mutational burden does not preclude an important role for other forms of DNA damage, for example, the induction of aneuploidy or gross chromosome changes, that could be induced by tissue damage or chemical tumor promoters. Early studies of the effects of TPA on cells in culture identified a “clastogenic” effect of TPA treatment, resulting in chromosome aberrations (46, 47). In vivo, chronic TPA treatment resulted in the appearance of papillomas showing aneuploidy of chromosomes 7 and 6 in premalignant papilloma cells (48), and tumors induced by DMBA/TPA, rather than repeated DMBA treatment, showed a higher level of chromosomal alterations (49). Recent single-cell analysis of multiple stages of progression using this skin model system has further identified a series of aneuploidy events in benign papillomas promoted by TPA treatment (50). These included the trisomies of chromosome 7 or 6 seen previously but also other genomic alterations in chromosomes 10 and 1 that seem to be involved in different stages of the evolution of papilloma cells toward malignancy (50). Further studies will be required to determine whether such chromosomal events are in some way directly induced by chemical promoter treatment or are a consequence of the rapid cell proliferation induced by promoters or during normal wound healing.

Genetic activation of mutant Kras or Hras expression in the skin did not lead to tumor formation over several months, but a single full-thickness wound led to the induction of rapidly growing lesions within a few weeks. These tumors almost completely lacked point mutations in known cancer drivers other than the engineered mutations in Ras genes, as previously noted for other engineered mouse cancer models (22). It is possible that direct induction of mutations in Lgr5⁺ or Lgr6⁺ stem cells may abrogate the requirement for many other mutations or alternatively that simultaneous activation of Ras mutations in multiple adjacent cells in the same compartment may allow initiated cells to escape competitive inhibition by surrounding normal cells (51). The strong selection of Ras-mutated cells after a wounding stimulus is in agreement with historical chemical carcinogenesis studies (52), as well as with genetic models demonstrating clonal selection of cells expressing mutant Ras genes by tissue damage or tumor promoter treatment (13). These results however differ from those derived from some alternative models, in which normal cells can outcompete their mutant counterparts after wounding (15, 16). These distinct outcomes may be due to variation in the mutant Ras alleles or gene promoters used to target different cell populations in the tissue of interest. For example, Krt19 has been used to target mutant Ras to the bulge region of the hair follicle (16), but these cells do not act as a common target for mutation by mutagen treatment, leading to RAS activation and cancer initiation in the dorsal skin (Kandyba and colleagues, submitted). Responsiveness to tumor induction by mutant Ras can also vary between different types of skin (53). Differences in tissue biology and experimental design may therefore result in an altered balance between the numbers of normal and mutant cells, altering the competitive landscape to favor the selection of different clones.

Our data, together with the results of previous work on WGS of mouse tumor models (17), emphasize the potential importance of environmental promoting factors as agents of selection for preexisting cells carrying driver mutations. Examples of promoter-induced selection of different driver mutations have previously been reported for skin (54) and liver tumors (55). In the previous study by Riva and colleagues (17), spontaneously arising lung tumors had the highest frequency of Braf^V637E (equivalent to human V600E) mutations, but this driver mutation is very rarely found in tumors promoted by a variety of different nonmutagenic promoting factors (P < 0.012; Fisher’s exact test). Another common spontaneous liver mutation was Hras^G13R (in 5/12 spontaneous tumors), but tumors induced by exogenous agents only rarely (4.9%) had the HrasG13R mutation, with most displaying mutations in alternative locations in Hras or Ctnnb1 (Supplementary Fig. S9). In the lung, only 4 of 14 spontaneous tumors had mutations in Kras, whereas 59.5% of all carcinogen-promoted lung tumors showed mutations in this strong cancer driver gene. We conclude that different types of exogenous factors damage normal tissues in distinct ways and stimulate tissue regeneration through alternative pathways that select from a wide array of preexisting spontaneous or carcinogen-induced mutations. The recent observations of abundant potential driver mutations in completely normal human tissues support the hypothesis that environmental promoting factors play a major role in human cancer etiology. Strong support for this possibility was recently published by Hill and colleagues (56), who demonstrated that air pollution, which is known to be associated with lung cancer in nonsmokers, induces tumors with a low mutational burden and frequent mutations in the EGFR cancer driver gene. Further studies from our laboratory (Kandyba and colleagues, submitted) have identified long-lived stem cells expressing Lgr6 in the upper hair follicle of mouse skin, rather than classical bulge stem cells, as the major cells of origin of chemically induced skin tumors that can remain latent over long periods but still retain the capacity to form tumors after exposure to tumor promoters. These results confirm the observations by Berenblum and colleagues (11) that initiated cells can remain in the skin permanently and that the most important and rate-limiting step in carcinogenesis is repeated exposure to promoting factors rather than initiation.

FVB/NJ (RRID: IMSR_JAX:001800) mice were purchased from the Jackson Laboratory. The standard model is based on a well-established two-step carcinogenesis model in which adult wild-type mice are exposed to topical DMBA at 8 weeks of age, followed by twice weekly administration of topical TPA for an additional 20 weeks. Mice are subsequently observed for tumor formation and sacrificed once tumor(s) are observed.

This experiment has been described previously in detail by Halliwill and colleagues (31). Briefly, male Mus spretus (SPRET/Ei; RRID: MGI:3510421) and female Mus musculus (FVB/N; RRID: MGI:2160001) mice obtained from The Jackson Laboratory were crossed to generate F1 hybrids. Female F1 mice were then crossed to male FVB/N mice to generate backcrossed mice, referred to as FVBBX mice. Tail tips were taken at 7 weeks and snap frozen in liquid nitrogen for RNA and DNA. For mice not undergoing tumor induction, tail, dorsal skin, and other tissues were harvested after sacrifice by asphyxiation and cervical dislocation at 7 weeks.

Chemical carcinogenesis was initiated using the two-stage DMBA/TPA protocol. In brief, seven-week-old mice were shaved and treated with 25 μg DMBA in acetone over an approximately 2-in. squared region on the center of the back. For the following 20 weeks, 200 μL of 10−4 mol/L TPA in acetone was applied twice per week to the DMBA-treated area. Mice were shaved 2 days prior to the first TPA treatment for that week. TPA treatments were continued for 20 weeks.

This experiment has been described previously in detail (33). Briefly, ICR (RRID: MGI:2160272) female mice (3–4 weeks of age, Harlan Teklad) were placed on the 10 kcal% fat control diet at 7 weeks of age and initiated with 25 nmol of DMBA (Eastman Kodak Co.). Four weeks after initiation, mice were randomized and placed on the four diets (n = 30 per group). The three groups consisted of control (10 kcal%), high-fat diet (70 kcal%), and caloric restriction (15% or 30%). Four weeks later, mice received twice-weekly topical treatments of 3.4 nmol of TPA (Alexis Biochemicals) for 50 weeks. Mice were weighed before randomization and then every 2 weeks for the duration of the experiments. Tumor incidence (percentage of mice with papillomas) and tumor multiplicity (average number of tumors per mouse) were determined weekly until multiplicity reached a plateau (29 weeks). Carcinoma incidence and average carcinomas per mouse were determined weekly from initial detection until 50 weeks after tumor initiation. SCCs were confirmed by histopathology.

FVB/NJ (RRID: IMSR_JAX:001800) mice were purchased from the Jackson Laboratory. For wounding experiments or TPA treatment studies, 8- to 10-week-old mice of the following strains were used: LGR5-EGFP-CreERT2+/-/LSL-KrasG12D, LGR6-EGFP-CreERT2^+/−/LSL-Kras^G12D, LGR5-EGFP-CreERT2+/-/LSLHrasG12V, and LGR6-EGFP-CreERT2^+/−/LSL-Hras^G12V. For wounding studies, mice were administered one topical dose of 4-hydroxytamoxifen (1 mg in 200 μL of 100% ethanol; Sigma-Aldrich) on the back skin to induce Cre expression, followed by full-thickness skin wounding 7 days later. For TPA-induced skin tumor promotion, 8- to 10-week-old mice of the same strains were initiated with a single dose of 4-hydroxytamoxifen (1 mg in 200 μL of 100% ethanol) on the dorsal skin, followed by biweekly promotion with TPA (200 µL of a 10–4  mol/L solution in acetone) for 20 weeks. Papilloma number was recorded up to 20 weeks. Mice were sacrificed when carcinomas developed and reached a size of >2 cm in longest diameter, papillomas were removed from the skin, and all internal tumors were resected.

Embryonic Cohort: FVB/NJ (RRID: IMSR_JAX:001800) mice were purchased from the Jackson Laboratory. Animals were paired at 8 to 10 weeks. Females with positive vaginal smears (spermatozoa) were given four doses of DMBA (15 mg/kg) by oral gavage on days 14 to 17 of pregnancy. Litters were divided in two groups. The first group at 8 weeks was treated by biweekly promotion with TPA (200 μL of a 10⁻⁴ mol/L solution in acetone) for 20 weeks, and the second group was monitored for 20 weeks.

Delay cohort: FVB/NJ (RRID: IMSR_JAX:001800) 8-to-10-week-old mice were purchased from the Jackson Laboratory. One group received a single dose of DMBA (25 μg per mouse in 200 μL acetone applied to shaved dorsal back skin), followed by a delay period of 25 weeks before commencing treatment with TPA (200 μL of 10⁻⁴ mol/L solution in acetone) twice weekly for 20 weeks. The second group received a single dose of DMBA (25 μg per mouse in 200 μL acetone applied to shaved dorsal back skin) followed 1 week after tumor initiation by promotion with TPA (200 μL of 10−4 mol/L solution in acetone) twice weekly for 20 weeks.

FVB/NJ (RRID: IMSR_JAX:001800) mice were purchased from the Jackson Laboratory. Cisplatin [cis-Diammineplatinum(II), D3371] was purchased from TCI and was dissolved in sterile saline solution at 0.8 mg/ml. A single dose of DMBA was applied to the back skin of the mice at 8 weeks. Mice were then treated with TPA twice a week for 10 weeks, followed by tumor formation. Approximately 4 weeks later, mice were injected IP with three doses of 100 μL cisplatin or saline solution (cisplatin-ctrl) administered at 4-day intervals (day 1, 5, and 9); cisplatin was dosed at 5 μL/g of body weight at the time of injection.

With the exception of the dietary cohort, animals were housed in standard conditions in accordance with the rules and protocols stipulated by the UCSF Institutional Animal Care and Use Committee, and all mouse experiments were approved by the UCF Laboratory Animal Resource Center. Mice were euthanized when carcinoma reached 2 cm in diameter or met the humane or experimental endpoint, such as weight loss, according to the rules and protocols stipulated by the UCSF Institutional Animal Care and Use Committee. The dietary cohort animal protocols were independently reviewed and approved and were housed in standard conditions in accordance with the rules and protocols stipulated by Texas A&M University.

All tumors were reviewed by a board-certified pathologist. DNA was extracted using Qiagen kits and standard procedures, followed by library preparation, and quality assurance steps are as described previously (17, 57).

We performed WGS of 93 tumor samples with matched normal tail tissue from the same animal for all of the cohorts with the exception of the tumors in the dietary cohort (N = 14), in which case 10 normal tail tissues from animals of the same syngeneic background were sequenced as controls. The Illumina HiSeq X Ten platform generated 151 base pair paired-end reads. Sequencing reads were aligned to the reference mouse genome (GRCm38) using BWA-MEM (BWA-MEM2; RRID:SCR_022192; ref. 58). Sequence coverage was 31.75- to 47.79-fold (median 38.78) after duplicate removal.

The variant calling algorithms of the Cancer Genome Project, Wellcome Sanger Institute, were used with default settings: cgpCaVEMan24 (RRID:SCR_017089) for base substitutions (59), cgpPindel25 (RRID:SCR_000560) for indels (60), and BRASS (RRID:SCR_017091) for structural variants. We performed additional postprocessing steps to eliminate false positive calls, due to technology-specific artifacts and germline variants. For all samples, we removed base substitutions with a median alignment score of mutation-reporting reads < 140. To perform signature analysis, we filtered out variants present in multiple tumors for samples in the dietary cohort in order to decrease the likelihood of contaminating SNPs because these samples did not have animal-specific matched normals. This filter was not used for the driver gene analysis.

We used SigProfilerMatrixGenerator (RRID:SCR_023122; ref. 23) to categorize mutations into classes and to plot mutational spectra. De novo substitution, doublet/dinucleotide, and insertion-deletion (INDEL) signatures were extracted using SigProfilerExtractor (https://github.com/AlexandrovLab/SigProfilerExtractor; RRID: SCR_023121), which is based on a nonnegative matrix factorization method. Before comparing the extracted mouse mutational signature to COSMIC database signatures, we performed normalization, multiplying signatures by the human mutational opportunity (hg19) and dividing them by the mouse mutational opportunity (mm10). After de novo signature extraction, we used SigProfilerExtractor to decompose these signatures into known human signatures in the COSMIC mutation database.

The mutational catalogue of a cancer genome, defined over an alphabet of mutation types $Ξ$ with $k$ letters, can be mathematically expressed as a set $S={s1, s2,…,sn}$⁠. To perform a relative reduction of $p$ for a given set $Q$ of $l$ mutation types defined over the same alphabet $Ξ$⁠, where $Q$ is a proper subset of $S$⁠, the number of somatic mutations of each type, that is, $qi∈Q$⁠, was reduced by applying Poisson resampling with parameters $λ=p*qi$⁠. After performing the reduction, samples were subjected to a set of routine mutational signature analyses.

We intersected CaVEMan filtered calls against the homologues of a previously published list of 369 driver genes in human cancers (36). We selected mutations that altered the coding sequence: missense, nonsense, splice site mutations, and start lost or stop lost substitutions. Custom analyses and visualizations were performed using the R library Oncoplot (RRID:SCR_001905).

RNA was generated from frozen tumor samples from high- and low-BMI male animals. Tail tissue was frozen in liquid nitrogen, ground with a chilled mortar and pestle, and suspended in TRIzol. TRIzol RNA extraction was then performed, followed by purification using a Qiagen kit. RIN values were calculated using a Bioanalyzer, and samples with high quality (greater than six) were used for analysis.

Samples were hybridized to an Affymetrix mouse gene ST 1.1 array. Gene expression values were calculated in R using the oligo package (RRID:SCR_015729;1.26.6; ref. 61). Expression values were normalized at the transcript level via the robust multi-array average (RMA) algorithm using a custom set of probe annotations. This custom set was designed to account for the high quantity of sequence dissimilarities between FVB and SPRET mice by removing any probe with a known SNP between these two strains from the annotation (31). Expression levels were then normalized across plates via the COMBAT method (RRID:SCR_010974; ref. 62).

We used the Carmen Software (RRID:SCR_002795), which has been described previously (63), to compare the expression levels of genes independently in males, females, and a combined dataset. Because the phenotype of interest, namely the association between obesity and increased risk of carcinogenesis and tumor progression, was a strongly sexually dimorphic trait observed in males, we opted to show expression analysis results for the analysis in male mice. We performed this analysis for normal (back skin) and tumor tissues to identify the genes with the strongest expression correlation with high or low BMI.

All other analyses and plotting were performed using R (RRID:SCR_001905), utilizing custom scripts.

The raw sequencing data are available for download from the European Nucleotide Archive (https://www.ebi.ac.uk/ena/browser/home) under accession nos. ERP024409, ERP024410, ERP107157, ERP110302, and ERP118253. The code used in this study is available upon request.

Y.R. Li reports grants from the NIH during the conduct of the study as well as grants from Varian, the Slive Foundation, and Blue Earth Diagnostics outside the submitted work. R. Delrosario reports grants from the NIH during the conduct of the study as well as grants from the NIH outside the submitted work. L. Riva reports employment with Nerviano Medical Sciences and has contributed to the manuscript when she was employed by the Wellcome Sanger Institute. E.N. Bergstrom reports personal fees from io9 outside the submitted work and has a patent for US-20240410019-A1 licensed to the Regents of the University of California. L.B. Alexandrov reports personal fees from Inocras and personal fees and other support from io9 outside the submitted work; has patents for 63/289,601, 63/269,033, 63/366,392, 63/412,835, and PCT/US2023/010679 pending and a patent for 10,776,718 issued; reports his spouse is an employee of Hologic, Inc.; and is a cofounder, CSO, scientific advisory member, and consultant for io9; has equity; and receives income. The terms of this arrangement have been reviewed and approved by the University of California, San Diego, in accordance with its conflict of interest policies. No disclosures were reported by the other authors.

Y.R.Li: design and implementation of data analysis, Figure preparation and writing-original draft, writing-reviews; E. Kandyba, R. Delrosario: design, generation and analysis of animal models and tumors; Q. Tran, D. Wu, M.Q. Reeves: tumor analysis, nucleic acid isolation and transcriptional profiling; K. Halliwill: data analysis of backcross mice, genotyping and analysis of obesity samples; N. Bayani, O.K. Mirzoeva: resources, supervision, administration; L. Riva, S.M.A Islam, E.N. Bergstrom, K. Achanta: data analysis; J. DiGiovanni: provision of dietary cohorts, study design; L.B. Alexandrov: data analysis supervision and methodology; A. Balmain: overall study design and supervision, funding acquisition, writing original draft, review and editing.

The authors would like to acknowledge David Adams, Jon Teague, and Laura Humphreys as well as other members of the Sanger Institute for discussions and administrative and technical support as well as the UCSF sequencing and genomics core services. This work was delivered as part of the MUTOGRAPHS (C98/A24032) and PROMINENT (CGCATF-2021/100008) teams, supported by the Cancer Grand Challenges partnership funded by Cancer Research UK, the NCI (1OT2CA278668-01), and the Scientific Foundation of the Spanish Cancer Association. A. Balmain was supported by NCI grants RO1CA184510, UO1 CA176287, R35CA210018, and the Barbara Bass Bakar Professorship of Cancer Genetics. E. Kandyba is supported by the NCI R35CA210018 and the NCI R50 (R50CA251479). Y.R. Li was supported by the NCI F32 postdoctoral fellowship (F32CA232635), NIH/NCI Paul Calabresi K12 Career Development Award (5K12CA001727; PI: Joanne Mortimer), and the NIH Director’s Common Fund DP5 Early Independence Award (1DP5OD033424).