Whole-Exome Sequencing Validates a Preclinical Mouse Model for the Prevention and Treatment of Cutaneous Squamous Cell Carcinoma

Keratinocyte (basal and squamous cell) skin cancers are the most common types of human malignancy. Since the 1960s, the average rise in new cases has been 3% to 8% per year (1). This trend is projected to continue because of depletion of stratospheric ozone, increased exposure to solar radiation, and longer life expectancy. Indeed, cutaneous squamous cell carcinomas (cSCC) are rapidly increasing in incidence, causing significant morbidity and mortality (2). In the 10-year period from 2001 to 2011, Scotland has seen a greater than 50% increase in the incidence of cSCC, with now around 3,000 new cases diagnosed annually and associated escalating health care burden and cost (3). In the United States, there were estimated 400,000 new cases of cSCC diagnosed in 2012 alone (4). It has been suggested that cSCCs represent an under-recognized health issue, and that deaths from cSCC may be as common as deaths from renal and oropharyngeal carcinomas and melanoma in some parts of the United States (4). This is particularly relevant to patients with multiple cSCC tumors, who have markedly elevated risk of recurrence and metastasis (5).

cSCCs typically arise in areas of “field change” on the head and neck where cumulative ultraviolet radiation (UVR) damage from the sun has induced multiple preinvasive skin lesions. Indeed, UVR is now well recognized as the major factor in the etiology of skin cancer (6). Being the most ubiquitous carcinogen in our environment, UVR is a complete carcinogen (an initiator and a promoter). The solar UV spectrum has two physiologically relevant wavelength components, the shorter and more highly energetic UVB (280–315 nm), which damages the epidermis of the skin, and the longer and more penetrant UVA (315–400 nm), which reaches the underlying dermis. Exposure to solar UVR causes generation of reactive oxygen species, damage to DNA, lipids, and proteins, including DNA damage repair proteins, inflammation, and immunosuppression (1). Together, these deleterious processes contribute to skin photoaging and photocarcinogenesis.

In addition to being among the most common, cSCCs are among the most highly mutated human malignancies, with reported mutation rates of 33 to 50 per Mb of coding sequence (7, 8). Such extraordinarily high mutation burden makes the possibility for success of a single-target therapy unlikely and highlights the need for development of prevention strategies. To be able to test candidates and develop strategies for skin cancer prevention (and/or treatment) agents, it is essential to have a preclinical model that closely recapitulates the development of the human disease. One of the most commonly used skin carcinogenesis models is the mutant Harvey-Ras (Hras(Q61L))-driven papilloma formation induced by topical treatment with the carcinogen 7,12-dimethylbenz[a]anthracene (DMBA) as the initiator, followed by chronic applications of 12-O-tetradecanoylphorbol-13-acetate (TPA) as the promoter. Recently, two independent groups have conducted comprehensive analyses of the mutational landscape of the mouse tumors that form in this model, confirming the driver mutations in Ras (Hras, Kras, and Rras2) genes with a characteristic DMBA/TPA signature (i.e., A>T and G>T transversions), which occur in a mutually exclusive fashion in approximately 90% of the tumors (9, 10). Although mutations in other genes were also identified, most of them did not have the DMBA signature, suggesting that they occur at the later stages during tumor development (9). Whereas the DMBA/TPA model constitutes an excellent model for RAS-driven tumors, as demonstrated by the significantly overlapping genes between the mouse tumors that form in this model and human SCC from cervix (44%), head and neck (35%), esophagus (25%) and lung (18%; ref. 9), activating mutations in RAS genes are infrequent (<11%) in human cSCC (8). In contrast to human cSCC, the DMBA/TPA–induced mouse tumors carry an average of 5.2 mutations per Mb of coding sequence (10). Overall, tumors that arise in this model do not represent the extraordinarily high level and broad spectrum of mutations that are characteristic of human cSCC.

A second skin carcinogenesis model uses UVB as the carcinogen. However, this model does not reflect sun exposure accurately as the solar UVR that reaches the surface of the earth consists predominantly (∼95%) of UVA, with only a small (albeit more carcinogenic) component of UVB wavelengths. The presence of the UVA component is particularly relevant for modeling the development of cSCC in specific high-risk populations, including patients that are undergoing chronic life-long treatment with thiopurine immunosuppressive and anti-inflammatory drugs, such as solid organ transplants recipients and inflammatory bowel disease patients. This is because the combination of UVA and the thiopurine metabolites that incorporate into DNA of proliferating skin cells is highly mutagenic, causing damage to DNA and DNA damage repair proteins (11–13). Indeed, azathioprine treatment photosensitizes the human skin to UVA radiation (14), and the skin cancer risk in organ transplant recipients is approximately 100-fold greater compared with the general population (15).

To overcome these limitations and build upon the “high-risk” skin carcinogenesis model in SKH-1 hairless mice (16, 17), we have developed a model in which immunocompetent SKH-1 hairless mice are subjected chronically and intermittently to suberythemal doses of solar-simulated UVR twice a week for 15 weeks (18). Irradiation is then discontinued, and tumor development is monitored. Although there are no tumor-bearing mice at the end of the period of irradiation, essentially all animals develop tumors during the subsequent 15 to 20 weeks, in the absence of further exposure to UVR. By use of this model, we found that genetic or pharmacologic upregulation of transcription factor NF-E2 p45-related factor 2 (Nrf2) protects against tumor development (18, 19). Importantly, protection was observed in both immunocompetent mice as well as in animals receiving chronic immunosuppressive therapy with clinically relevant doses of the thiopurine drug azathioprine (18). Very recently, a similar model (with a slightly different irradiation schedule) was employed by Kim and colleagues (20) who found that Fyn, a member of the Src family of protein tyrosine kinases, acts as a redox sensor in a signal transduction cascade induced by exposure to solar-simulated UVR. Using whole-exome sequencing of genomic DNA isolated from microdissected tumor tissue, we now show that the cSCCs that arise in this mouse model accurately represent the mutational spectrum of human cSCC.

Together, these findings imply that the “high-risk” solar-simulated UVR skin carcinogenesis mouse model represents a valid model for human cSCC and justify its use for preclinical testing during the drug development process of topical and/or systemic agents for the prevention and treatment of human cSCC. Furthermore, such testing could be performed on cancer-preventive agents, which have UVR-absorbing effects (e.g., sunscreens) and are applied during the irradiation period, as well as on drug candidates, which act by non-UVR–filtering mechanisms and are applied during the tumor development period after cessation of the UV irradiation schedule.

The animal experiments were performed according to the rules and regulations described in the UK Animals (Scientific Procedures) Act 1986. All experimental animals were age-matched and female. The animal study plan was developed after ethical approval was granted (project license 60/5986) and was further approved by the Named Veterinary Surgeon and the Director of Biological Services of the University of Dundee (Dundee, Scotland, United Kingdom). SKH-1 hairless mice (initially obtained from Charles River Laboratories) were bred and maintained in the Medical School Resource Unit of the University of Dundee on a 12-hour light/dark cycle and 35% humidity. Throughout the study, the animals had free access to water and pelleted RM1 diet (SDS Ltd.).

Cutaneous carcinogenesis was initiated when the mice were 8 weeks old by subjecting the animals chronically twice a week (on Tuesdays and Fridays) for 15 weeks to solar-simulated UVR (composed of 2 J/cm² UVA and 90 mJ/cm² UVB). UVA340 lamps (Q-Lab) were used as the irradiation source. Irradiation from these lamps simulates the solar UVR from 365 nm to the solar cutoff of 295 nm, with a peak emission at 340 nm. The radiant dose was quantified with a UVB Daavlin Flex Control Integrating Dosimeter and was further confirmed by use of an external radiometer (X-96 Irradiance Meter; Daavlin) before and after each irradiation session. The mice were placed in clear, bedding-free cages and then exposed to UVR. To prevent excessive heating and discomfort to the animals, the irradiation unit (Daavlin) was equipped with an electrical fan. Tumors (defined as lesions >1 mm in diameter) that formed on the dorsal skin were measured and mapped once a week.

The experiment was terminated and the tumors were collected at 20 weeks after completion of the irradiation schedule (i.e., 35 weeks after the onset of irradiation). The tumor samples were snap-frozen in liquid N₂ and stored at −80°C. Normal nonirradiated ventral skin from the same animals was also obtained at the same time and stored under identical conditions. This normal skin served as a source of matched germline DNA. Laser capture microdissection was employed to enrich for tumor cell populations and prevent contamination by infiltrating inflammatory cells using Zeiss Palm Microbeam microscope (Zeiss). Depending on tumor size and purity (as estimated by the examination of H&E reference slides that were prepared in parallel), approximately 20 to 90 sections of 8-μm thickness were cut onto 1.0-mm PEN membrane slides. After staining with 0.05% acid fuchsin (Acros Organics) in distilled water and 0.05% toluidine blue O (Acros Organics) in 70% ethanol, the sections were microdissected. Tumor cells were collected into 180 μL ATL buffer (Qiagen). Genomic DNA was extracted from tumor cells and from matched normal skin samples (which were crushed into fine powder under liquid N₂) using the QIAamp DNA Micro Kit (Qiagen). Exome capture and sequencing was performed by Oxford Genome Technology at a sequencing depth of 100× for the tumor DNA, and 50× for the normal skin DNA.

Raw sequencing reads were aligned to the reference genome (UCSC mm9) using Bowtie2, a fast sensitive read alignment software (21). Single-nucleotide variants that differed between tumor and normal samples were identified using SomaticSniper 1.0.5. with somatic score filter set at 40 (22). Small insertions and deletions (INDEL) were called using Platypus, a haplotype-based variant caller for next-generation sequencing (23). Mutation calls were annotated using SNPEff (24). Copy number alterations were estimated from the exome sequencing data using the cn.MOPS (copy number estimation by a mixture of Poisson) algorithm (25). Matched normal samples processed in the same batch as the tumor samples were used as controls to identify sample-specific copy number alterations. The mapped reads for the exome sequencing have been deposited in the Sequence Read Archive database (project number: PRJNA352449). Mutational signature and statistical analyses were performed in R statistical environment (26).

The development of cSCC in humans is usually associated with chronic sun exposure in early life (27). We therefore subjected mice, beginning at 8 weeks of age, chronically, twice a week for 15 weeks, to solar-simulated UVR (comprised of 2 J/cm² UVA and 90 mJ/cm² UVB). At termination of the irradiation schedule, none of the animals had developed tumors. However, during the subsequent 20 weeks, tumor incidence reached 100% of animals. Visual inspection and mapping of the lesions that formed on the irradiated dorsal skin of each mouse revealed multiple tumors of variable sizes (Fig. 1), with tumor multiplicity being on average of 5 tumors per mouse (18, 19). Interestingly, many animals displayed a “field change,” also known as “field cancerization” (28), which is a typical representation of the clinical situation. This observation implies that, similar to humans, SKH-1 hairless mice develop multiple primary cSCCs in close proximity arising within histologically dysplastic epithelium following chronic intermittent exposure to solar-simulated UVR.

The histopathologic spectrum of the mouse sSCCs (mcSCC) that develop in this model resembles very closely the spectrum of human cSCCs (hcSCC; Fig. 2; Table 1). All lesions show clear evidence of background generally severe epidermal dysplasia, often with follicular extension. The lesions group into three main categories: (i) actinic keratosis with no definite invasion; (ii) actinic keratosis with probable early dermal invasion, and (iii) invasive SCC, all in the moderate- to well-differentiated range and some showing extension into the muscle layer. This spectrum of skin tumors is in stark contrast with the tumors that form in the DMBA/TPA skin carcinogenesis model, which are primarily papillomas.

We performed exome sequencing of 18 microdissected tumor samples (cases) derived from our SKH-1 hairless mice that had been chronically exposed to solar-simulated UVR as described in Materials and Methods. Tumor selection was based on size. The size and histopathologic characteristics of each individual tumor are described in detail in Table 1. Ventral nonirradiated skin from the same animals served as a source of matching normal control DNA. The sequencing revealed a median point mutation (SNP) rate of 155 mutations per Mb with a high of 279 and a low of 12 (Fig. 3A; Supplementary Table S1). The overwhelming majority of SNPs were identified as C.G>T.A transitions (78.6%, Fig. 3B), a characteristic UVR-induced mutational signature (29), confirming that UVR is the primary cause of mutations in the mcSCC. It is well established that exposure to UVR causes formation of cyclobutane dimers and pyrimidine–pyrimidone (6–4) photoproducts in DNA that initiate C.G>T.A transitions (30–32). Also consistent with the UVR-induced mutational signature, 81.6% of all C.G>T.A mutations in the mouse tumors occurred following a pyrimidine base (Fig. 3C), indicative of the formation of dipyrimidine cytosine-containing oncogenic photoproducts (29). In addition, the sequencing revealed a median short insertion and deletion mutation (INDELs) rate of 23.5 per case, including cases harboring zero INDELs to cases with as many as 370 INDELs (Supplementary Table S2).

Our results show a remarkably similar mutational signature of the mcSCC to the mutational signature previously described in hcSCC (8, 33). Pickering and colleagues reported that an average of 75% of the mutational events in human patient cases were C>T transitions and that 85% of these mutations were found at locations following a pyrimidine base (33). Similarly, South and colleagues observed that approximately 68% of all SNP mutations were C>T transitions in an independent cohort of hcSCC samples (8). Critically, direct comparison of these previously analyzed human hcSCC cases and our mcSCC tumors showed high overlap in terms of both frequency and type of SNP mutations uncovered (Fig. 3D).

The trinucleotide context of the mutation sites identified in the mcSCC show strong correlation with those in hcSCC reported by Pickering and colleagues, and by South and colleagues, with C>T mutation predominantly occurring around Y(C>T)N trinucleotide (Y represents pyrimidine and N represents A/T/G, or C; Fig. 3E; Supplementary Fig. S1). Moreover, similar comparison also revealed high correlation in terms of mutation signature between the mcSCC and a UVR-associated melanoma mouse model (Fig. 3F; Supplementary Fig. S1).

Pickering and colleagues reported recurring regions of copy number gain on chromosomes 7, 8q, 9q, 14, and 20. To compare the mcSCC to the reported copy number alteration in hcSCC, we estimated copy number alteration from our exome sequencing data. Recurring (in at least 3 of the 18 tumor samples) regions of copy number gains were detected on chromosome 2, 5, 7, 12, 14, 15, and 18 (Table 2, Supplementary Table S3). Interestingly, genes located within these regions significantly overlapped with those located in the regions of gains (chromosomes 14, 7, and 8q) reported in hcSCC (Table 2, enrichment P < 0.05, hypergeometric test). However, regions of loss (chromosomes 5, 6, 9, and 11) detected in the mcSCC did not significantly overlap with those in hcSCC.

Pathway analysis revealed mutations in genes encoding proteins that participate in multiple signaling pathways and cellular processes, such as the Jak–Stat signaling pathway, the neurotrophin signaling pathway, the adipocytokine signaling pathway, T-cell and B-cell receptor signaling pathway, chemokine signaling pathway, MAPK signaling pathway, Hedgehog signaling pathway, p53 signaling pathway, Notch signaling pathway, cytokine–cytokine receptor interactions, ECM–receptor interactions, endocytosis, phagocytosis, melanogenesis, focal adhesion, and metabolism. A complete list is shown in Supplementary Table S1.

Particularly striking was the occurrence of mutations in Trp53 and genes encoding multiple family members of the Notch signaling pathway. In hairless mice, early studies have linked mutations in Trp53 to exposures to UVB but not to UVA radiation (34). More recently, mutations in the TP53 gene as well as the genes that encode the NOTCH family of receptors were identified as driver mutations in hcSCC. Interestingly, whereas it has long been known that most if not all hcSCCs contain mutations in the TP53 gene, it was recently shown that a significant number of hcSCC tumors also harbor mutations in NOTCH receptor–encoding genes (8), with NOTCH1 or NOTCH2 loss-of-function mutations being present in approximately 75% of hcSCCs (35). We therefore conducted a detailed analysis on the mutations in Trp53 and Notch genes that were detected in our mcSCC tumor samples.

A total of 21 independent nonsynonymous mutations were found in the Trp53 gene, representing 15 of the 18 (83%) samples sequenced, whereas 6 of the 15 samples contained two or more mutations (Fig. 4A). The mutations were centered on 19 locations distributed across 5 of the 12 exons of the Trp53 gene. In addition, 20 of the 21 SNP mutations are located in the DNA-binding domain of the protein, suggesting possible loss-of-function consequences. The presence of all of these mutations in the mcSCC tumors was validated by Sanger sequencing (Supplementary Table S4).

Mutations in genes encoding the NOTCH family of receptors in hcSCC are usually inactivating and are typically found in the EGF repeat domain. To determine the frequency and location of mutations in these genes, we analyzed all four members of the mouse Notch family of receptors (Notch1-4) in the mcSCC tumors. Nonsynonymous SNP mutations in genes encoding the four Notch family members were found in 10 of the 18 mcSCC tumors sequenced (∼55%, Fig. 4B–E). As in the hcSCC tumors, 7 of the 16 mutations are located in the region encoding the EGF repeat domain of the mouse genes. A total of 6 mutations representing 5 different mcSCC tumors were found in Notch1 (Fig. 4B), 3 mutations representing 3 different mcSCC tumors were found in Notch2 (Fig. 4C), 5 mutations representing 4 different mcSCC tumors were found in Notch3 (Fig. 4D), and 2 mutations representing 2 different mcSCC tumors were found in Notch4 (Fig. 4E). We validated the presence of all of these mutations in the mcSCC tumors by Sanger sequencing (Supplementary Tables S5–S8).

As noted above, loss-of-function mutations in NOTCH1 or NOTCH2 occur in approximately 75% of hcSCCs (35) and are considered to be an early event in the development of hcSCC (8). Such inactivating mutations are consistent with the role of Notch signaling in skin, where in contrast to most other tissues, Notch1 negatively regulates keratinocyte self-renewal, promotes differentiation, and acts as a tumor suppressor through a number of cell-autonomous as well as non–cell-autonomous growth control mechanisms (36–41). The high frequency of mutations in Notch1-4 in our mcSCC tumors, which often form areas of field cancerization, is in close agreement with a recent report showing that in mice, loss of mesenchymal Notch signaling leads to field cancerization and multifocal epithelial tumors (42). It was suggested that this could be due to the role of Notch in inhibiting activator protein-1 (AP-1)–mediated transcription of secreted growth factors, proteases, and extracellular matrix proteins. The same study showed that Notch signaling is suppressed in stromal fields adjacent to cutaneous premalignant actinic keratosis lesions surgically excised from human patients. In addition, persistent suppression of Notch signaling, accompanied by enhanced activity of AP-1, is induced in the dermis of human skin explants and dermal fibroblasts upon exposure to UVA, but not UVB radiation (42), highlighting the importance of the UVA component of the solar UVR in the development of cSCC.

Taken together, data presented in this study show that the skin tumors that develop in this mouse model are similar to human cSCC tumors in terms of histopathology, mutational characteristics, and UVR-induced mutational signature. We conclude that by (i) using UVR as the carcinogen; (ii) incorporating both the UVA and the UVB components of the solar UVR at their natural ratio; and (iii) employing chronic but intermittent irradiation schedule, this mouse skin carcinogenesis model has high relevance to the human disease. We propose the use of the model for mechanistic studies aiming to further our understanding of cSCC development, as well as for testing potential prevention and treatment strategies for human cSCC.

No potential conflicts of interest were disclosed.

Conception and design: E.V. Knatko, C.A. Harwood, C.M. Proby, A.T. Dinkova-Kostova

Development of methodology: E.V. Knatko, K.J. Purdie, C.A. Harwood, A.T. Dinkova-Kostova

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E.V. Knatko, C.A. Harwood

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E.V. Knatko, B. Praslicka, A. Ooi, A.T. Dinkova-Kostova, A. Evans

Writing, review, and/or revision of the manuscript: E.V. Knatko, B. Praslicka, C.A. Harwood, C.M. Proby, A. Ooi, A.T. Dinkova-Kostova

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Higgins

Study supervision: C.M. Proby, A.T. Dinkova-Kostova

Other (facilitated study as a collaboration between University of Dundee and Queen Mary University of London): C.M. Proby

We thank Oxford Genome Technology for performing the whole-exome sequencing and Andrew Cassidy (University of Dundee) for Sanger sequencing.

This work was supported by Tenovus Scotland, Cancer Research UK (C20953/A10270 and C20953/A18644), and the British Skin Foundation (project number 7015; to A.T. Dinkova-Kostova).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.