Genetically Engineered Mouse Models of Melanoma

Abstract

Melanoma is a complex disease that exhibits highly heterogeneous etiological, histopathological, and genetic features, as well as therapeutic responses. Genetically engineered mouse (GEM) models provide powerful tools to unravel the molecular mechanisms critical for melanoma development and drug resistance. Here we expound briefly the basis of the mouse modeling design, the available technology for genetic engineering, and the aspects influencing the use of GEMs to model melanoma. Furthermore, we describe in detail the currently available GEM models of melanoma.

Introduction

Genetically engineered mouse (GEM) models have been successfully used for decades in modeling human cancer [1, 2]. Mice provide a relevant platform for prospective studies to investigate specific hypotheses and causative associations in human disease. Moreover, the continuous advances in genetic engineering allow more precise spatial and temporal control of genes, improving the capacity to recapitulate the events in carcinogenesis and disease progression [3]. However, many etiological, genetic, and physiological factors need to be addressed for appropriately modeling human disease in mice, especially for particular cancer types. In this review we describe the basic principles of modeling human cancer in mice and currently available GEM models of melanoma, highlighting the key factors to consider and the questions that still need to be addressed in the field.

Melanoma arises from malignant transformation of melanocytes, which are specialized pigment-producing cells derived from the developing neural crest [4]. In humans, melanocytes are predominantly located in the epidermal layer of the skin, but are also present in the hair follicle bulbs, eyes, inner ear, mucous membranes, central nervous system, and at lower densities in many internal organs [5]. Therefore, melanoma mostly develops in the skin but may also occur in other sites. The etiology, histopathology and genetic characteristics of melanomas are very diverse, resulting in a highly complex and heterogeneous disease. Clinically, melanomas are classified according to their growth pattern (e.g. superficial spreading, lentigo or nodular melanoma), anatomical site (e.g. cutaneous, mucosal, or acral melanoma) or pigmentation status (e.g. melanotic and amelanotic melanoma) [6]. A melanoma may exhibit more than one of these phenotypes; for example, amelanotic cutaneous melanoma. Moreover, recent advances in genomic analysis make it possible to categorize the molecular subtypes of melanoma, and their association with pathological features have provided new opportunities for diagnosis [7]. Here we focus on cutaneous melanoma, since it is the most frequent type in humans.

Cutaneous malignant melanoma is the deadliest type of skin cancer, and its incidence has been continuously increasing over the last few decades [8]. Melanoma can originate from a benign lesion, called a nevus, with irregular morphology, pigmentation and growth pattern. In the early stage of progression, transformed melanocytes expand horizontally in the epidermis leading to radial growth phase (RGP) melanomas. As it progresses to the vertical growth phase (VGP), melanoma cells invade the underlying dermis. In late stages metastatic melanoma cells reach the vasculature and lymph nodes, and subsequently disseminate to distal sites [9]. Metastatic melanoma is inherently resistant to most of the conventional chemotherapeutic agents and patient survival rates are typically low. Therefore, the identification of the key elements for melanoma development and their interactions is critical to generate better prevention and therapeutic strategies.

In this regard, modeling is required to decipher a system exhibiting high complexity, such as cancer, and delineate the essential contributors to the disease. In general, the modeling process comprises five steps: 1) defining the question to assess and the system used; 2) identifying the critical variables involved; 3) testing the model; 4) comparing the generated and expected outcomes especially with respect to human data; and 5) adjusting the system to improve the model and repeating the test until the new hypothesis is validated [10]. Once the scope of the question aimed to investigate (e.g. etiology, pathology, or therapeutic responses) is established, the relevant “driving factors” should be hypothesized and implemented into the model system. The model is then subjected to specific maneuver (e.g. UV irradiation, activation of oncogene, or therapeutic treatment) for testing the hypothesis. The outcome should be compared with data from human sources (e.g. epidemiological, pathological, or genomic data) to evaluate the relevance of the hypothesis. The unexpected or unpredictable results will provide information to decide the new combination of driving factors for the next cycle of modeling. In this sense, GEMs are a powerful tool to model human cancer, and has played an especially critical role in our current understanding of melanomagenesis and progression. The embryonic development of melanocyte precursors is well-conserved between human and mouse, making the latter suitable to model human melanocytic pathology. Moreover, current genetic engineering technologies enable flexible control of multiple genetic alleles independently, allowing construction of a model for testing multifactorial hypothesis. GEM models have been used to address three main aspects of melanoma biology: (i) the initiating events and driver mutations required for malignant transformation, (ii) the identification of biomarkers for diagnosis and (iii) the development and improvement of therapies (Fig. 1).

Fig. 1.

Timeline of expression of melanocyte-specific genes during mouse embryogenesis. The promoters of these genes are good candidates to be used to drive expression of transgenes in melanocyte-specific manner in mouse models of melanoma and melanocyte biology. The black bar depicts embryonic (E) developmental days of gestation (from E9.5 to E18.5). Mitf, Microphthalmia Associated Transcription Factor; Dct, Dopachrome tautomerase; Mlana, Melanoma antigen recognized by T-cells; Pmel, Premelanosome protein; Trp1, Tyrosinase-related protein 1; Tyr, Tyrosinase. (Adapted from ref. 33).

Factors to consider in melanoma GEM studies

Model building requires a platform that reflects the system studied and allows maneuvering of the variables involved in the hypothesis. In other words, relevance to the “real” system determines if the model is appropriate for a particular question. In this regard, understanding skin physiology, especially the development of melanocytes and their interaction with the microenvironment, the similarities and differences between mouse and human, as well as the environmental factors that influence melanoma formation, are crucial for modeling the disease. The technology of genetic engineering should be used to achieve the relevance by appropriate spatiotemporal control of the carcinogenic events. Here we discuss the basic factors to consider in modeling melanoma.

Skin morphology

Despite the conserved mechanisms of melanocyte development during embryogenesis, mouse and human skin have distinct anatomical and functional features which may influence the initiation and progression of melanoma. The mammalian skin is comprised of the epidermis, formed by several layers of keratinocytes, and the dermis, composed of connective tissue, hair follicles, sweat glands, blood vessels and nerves. In general, human skin is thicker with large interfollicular areas, and melanocytes are distributed along the basal layer of the epidermis as well as in the hair follicles. In contrast, mouse skin only contains 2–3 layers of keratinocytes and is covered by a dense coat of hair. Mouse melanocytes are located primarily in the bulb regions at the bottom of the hair follicles. Melanomas rarely develop spontaneously in mice. Those generated in GEMs are mainly dermal and share limited histological similarities with human melanoma, which have predominantly an epidermal or dermo-epidermal junctional component. The expression of a hepatocyte growth factor/scatter factor (HGF/SF) transgene constitutively in mice, resulting in chronic activation of the MET tyrosine kinase receptor, has been shown to modify melanocyte distribution, increasing their presence at the dermo-epidermal junction. This enhances their susceptibility to transformation by ultraviolet radiation (UVR) and the melanomas that originate in this GEM share more histopathological features with their human counterpart [11, 12].

Melanocytic lineage

A critical factor for the generation of tissue-specific GEM models is the promoter used to target a specific cell type. This is particularly relevant for melanoma since expression of oncogenes in non-melanocytic cells would trigger the development of undesirable non-melanoma malignancies. During embryogenesis and subsequent differentiation, melanocytes express specific genes required for their survival and maturation. Many of these genes are associated with melanocyte differentiation and pigmentation pathways, such as SRY Box 10 (SOX10), Paired Box 3 (PAX3), Tyrosinase (TYR), Dopachrome Tautomerase (DCT) or Microphthalmia-Associated Transcription Factor (MITF). During embryonic development, SOX10 and PAX3 induce expression of MITF in neural crest cells, initiating their differentiation to melanoblasts [13]. Subsequently, MITF, together with other factors, activates a series of genes in a sequential manner to control differentiation of melanoblasts and maturation of melanocytes (Fig. 2). Noticeably, since melanocytes derive from neural crest progenitors, some of these markers are shared by cells of neural origin [14]. For example, the most commonly used tissue-specific promoter in melanoma GEM models is derived from the Tyr gene, which is also expressed in Schwann cells and spinal ganglia [15]. Moreover, melanocytes are not a homogeneous population. It has been reported that epidermal and follicular melanocytes in mice may have different origins [16]. Melanomas with diverse pathological properties could arise depending on the developmental origin and differentiation status of melanocytes. Therefore, an extensive molecular and histopathological characterization of the tumors generated in GEMs is necessary for the most informative interpretation of these models.

Fig. 2.

The iDct-GFP mouse model. This tet-inducible (Tet-On) model expresses green fluorescent protein (GFP) in the melanoblast/melanocyte compartment during embryonic stages (a) and adult skin (b). In this mouse, the rtTA is driven by the Dct promoter, and GFP is under the control of TRE promoter. The expression of GFP is activated in the embryos by feeding the pregnant dam with doxycycline-fortified diet, and in adult mice by a single doxycycline injection 24h prior to imaging. (a) At embryonic day 11.5 (E11.5) time point, the GFP is expressed in the neural crest (NC), retinal pigment epithelium (RPE), and telencephalon (T) of an iDct-GFP embryo. (b) An iDct-GFP; HGF/SF-Tg mouse that was UVB-irradiated at neonatal day 3.5 and imaged via Xenogen IVIS system at 4 weeks of age. The enhanced number of epidermal melanocytes and nevi can be readily visualized on shaven dorsal skin, ears, eyes, paws, and tail (see text and ref. 43 for details).

Carcinogenic agents

Another important factor to consider in GEM models is the use of carcinogenic agents to initiate or enhance development of melanomas. UVR as the main environmental risk factor for melanoma has been used extensively to study melanomagenesis in mice. Some studies have demonstrated that pigmentation (melanin content and/or deposition) plays a key role in the ability of UVR from different wavelengths (UVA vs. UVB) to initiate melanoma in mice [11, 17]. Both human and mouse can produce two types of melanin in the skin, pheomelanin and eumelanin, and the relative frequency of each is responsible for skin and hair pigmentation [18]. Eumelanin is the black pigment that is primarily responsible for the photoprotective function. Pheomelanin is yellow-reddish in color, and is mainly present in individuals with red hair and freckled fair skin, which has been epidemiologically associated with a higher risk of developing melanoma. The synthesis of eumelanin is regulated in part by the melanocortin-1 receptor (MC1R), such that MC1R inactivating mutations decrease the eumelanin/pheomelanin ratio and increase spontaneous melanomagenesis in a mouse model of melanoma [19]. The proportion of eumelanin/pheomelanin also varies in different mouse strains. Consequently, the differences in melanin production and UVR absorbance between mouse and human should be considered when assessing translational applicability of data obtained from GEMs.

The carcinogen 7,12-dimethylbenz[a]anthracene (DMBA), used commonly to induce epithelial tumors in the skin, can accelerate melanoma development in some GEM models [20]. Further analysis of these models will allow identification of distinct pathways targeted by this carcinogenic agent and determine whether they correlate with human etiology or represent particular types of melanomas. In order to evaluate the relevance of driver oncogenes versus carcinogenic agents in melanoma initiation and progression or the relative role played by each, different regimens of oncogene activation and carcinogenic treatment should be analyzed.

Age

The implementation of sophisticated genetic engineering technologies allows induction of genetic modifications at specific timings in the life of the mouse. The timing chosen to introduce the genetic alteration would significantly impact melanoma development since melanocytes at different developmental stages may have differential susceptibility to transformation. For example, epidemiological studies have shown that sun exposure during childhood increases the risk of developing melanoma as adult. The hypothesis that melanocytes in young children’s skin are more prone to transformation since they are not fully differentiated, maintaining properties of precursor cells (melanoblasts), has been supported by animal studies. For example, HGF/SF-Tg mice, which harbor epidermal melanocytes, develop melanomas only if they are subjected to UVB irradiation at neonatal age (see below) [12].

Skin microenvironment

There is also evidence suggesting that the skin microenvironment could contribute to melanomagenesis. For example, it has been demonstrated that skin inflammation triggered by UVR enhances angiogenesis and favors melanoma metastasis [21]. However, as mentioned earlier, mouse and human skin present significant structural differences including the distribution of immune cells and microbiome composition. These microenvironmental differences could have implications in melanoma development and should be considered when GEM models are analyzed. Below we have focused on GEM models targeting only melanocytic cells since the specific contribution of stromal cells to melanomagenesis has not been fully explored yet using GEMs.

The design of GEM melanoma models must include additional elements beyond the induction of the driver events under study. The spatiotemporal control of genetic modifications, the environmental factors involved, the unique features of mouse morphology and physiology, and the strain genetic background of the mice [22] need to be taken into consideration in order to avoid confounding and potentially misleading interpretations of the data, as well as to achieve maximum relevance to human disease.

Spatiotemporal control of gene expression in GEMs

Mouse modeling allows direct in vivo analysis of the consequences of genetic alteration(s) on tumor genesis, progression, and metastasis. However, in many cases, germline absence (homozygous knockout) of a putative tumor suppressor (or otherwise essential) gene or constitutive overexpression of an oncogene can result in embryonic lethality, uncontrollable tumorigenesis, and/or a tumorigenic tissue spectrum unrepresentative of human disease. For example, mice with germline homozygous knockout of the tumor suppressor PTEN (Pten^−/−) are embryonic lethal by day 9.5 of gestation, and heterozygous (Pten^+/−) mice are susceptible to multiple types of tumors, especially T-cell lymphoma [23]. Continuous improvements in the transgenic mouse technology have given rise to capabilities to express and/or deactivate genes of interest in a tissue-specific as well as time-dependent manner. These capabilities have allowed generation of sophisticated mouse models that conditionally overexpress and/or knock out multiple genes of interest simultaneously and on demand. These models have made remarkable contributions towards understanding of the roles of many genes, and specific mutations therein, in the etiology of melanoma.

The Cre-recombinase/LoxP system is the most commonly used technique to produce conditional knockout or conditional mutant mouse models [24, 25]. This system is based on expression of the Cre recombinase gene under the control of a tissue-specific promoter. Cre recombinase recognizes a 34bp DNA sequence called LoxP, and catalyzes recombination between two directly repeated LoxP sites at high efficiency. This recombination results in excision of the DNA sequence flanked by the two LoxP sites (floxed). This method can be used to either delete essential sequences to cause complete gene knockout of a tumor suppressor or expression of a constitutively activated mutant oncogene. This system was refined to introduce temporal control of Cre expression on demand [26]. In the Cre-ER version, Cre is fused with a mutant ligand-binding domain of the estrogen receptor [27]. Cre-ER is inactive and requires the ER ligand tamoxifen (or its metabolite 4-hydroxytamoxifen, 4-OHT) for activation. A further improvement to reduce the background and increase the sensitivity and efficiency of Cre-ER was achieved by introducing further mutations in the ER domain, denoted as Cre-ER^T2. The Cre-ER^T2 is the most effective means of achieving LoxP-dependent excision of target DNA sequences in both spatial and temporal manner [28].

Using the Cre-ER^T2/LoxP system, the McMahon and Bosenberg groups generated a mouse model that conditionally expresses the constitutively activated Braf^V600E mutant oncogene with simultaneous deletion of the PTEN tumor suppressor in a 4-OHT-dependent and melanocyte-specific manner [29]. The most common melanocyte-specific Cre-ER^T2 transgenic mouse currently in use is driven by the Tyr promoter [30, 31], which is also used in the McMahon/Bosenberg mouse melanoma model. While the Tyr::CreER^T2 mouse is appropriate for most applications requiring melanocyte-specific gene modulation, under some circumstances other promoter-driven Cre may be desirable. For example, The Dct and Mitf promoters are activated at an earlier time point (before embryonic (E) day 10.5) during melanocyte development than Trp1 and Tyr (Fig. 1) [32, 33]. In fact, Tyr is one of the last melanocytic genes to be expressed (E15.5) along the melanocyte developmental timeline [33]. However, the other promoters mentioned earlier may also drive extra-melanocytic expression of Cre, which may or may not be desirable. For example, Dct, Trp1, and Mlana promoters also drive expression in the retinal pigment epithelium (RPE) [34–37]. MITF-Cre is expressed in the ocular melanoblasts, but not in the RPE, and seems to be highly specific for the melanocytic lineage during embryogenesis and postnatal development [38]. Recently, the Mitf-Cre model has been used to drive expression of the GNAQ^Q209L transgene to produce uveal melanoma with 100% penetrance, whereas the Tyr::Cre-ER^T2 failed to do so [39].

While Cre-ER^T2/LoxP is a powerful system for spatiotemporal control of gene expression, the consequence is mostly irreversible. A reversible and inducible system is required to identify driver oncogenes or to label cells in a particular developmental stage. Tetracycline (Tet) system allows switching on and off (Tet-On and Tet-Off, respectively) the expression of a gene of interest in a tissue-specific manner [40]. The Tet system is composed of (1) a tetracycline transactivator (tTA; Tet-Off) or the reverse tetracycline transactivator (rtTA; Tet-On system) under a tissue-pecific promoter, and (2) the gene of interest driven by a tetracycline responsive element (TRE) promoter. Administration of doxycycline, a more stable analog of tetracycline, to mice harboring the two transgenes results in suppression (Tet-off) or induction (Tet-on) of the expression of the gene driven by the TRE promoter. The Tet-inducible systems can control gene expression at precise timing and on demand [40]. In fact, we have used this system to build a mouse model allowing us to label melanocytes and their precursors at different developmental stages (see section below). This system has also been used to construct a mouse model of melanoma to study oncogene addiction. There are multiple other excellent systems for spatiotemporal control of gene expression in GEMs, such as gene delivery by Rous Sarcoma Virus in mice expressing its corresponding receptor, TVA. We discuss these models in detail below.

Melanocyte Cell Labeling in Mice

In-depth study of the biology of any cell lineage is best performed while the cells reside in their natural morphological and physiological microenvironment. This task has been difficult to achieve for melanocytes especially, due to the fact that melanocytes make up only about one percent of the cellular milieu of the mammalian skin. The Dct-LacZ transgenic mouse model has been used to label melanocytes by expressing the beta-galactosidase gene under the control of the melanocyte-specific promoter of the Dct gene, and its detection via X-gal staining and colorimetric assay [41]. Although this mouse model has been instrumental in studying the biology of melanocytes, the label is neither inducible nor fluorescent, which limits its utility.

Tetracycline-inducible transgenic mouse modeling is a powerful system to direct tissue-specific expression of genes of interest [40]. We have previously generated a mouse model that expresses rtTA under the control of the Dct promoter (Dct-rtTA). We crossed this mouse with a transgenic mouse that harbors the green fluorescent protein (GFP) gene under the control of TRE promoter [42], resulting in targeted GFP labeling of the melanoblast/melanocyte compartment during embryonic and postnatal developmental stages as well as in adult skin (Fig. 2) [43, 44]. This compound mouse model (iDct-GFP) provides an opportunity to study melanocytes not only at the cellular level, but the fluorescent melanocytes can also be isolated by Fluorescence Activated Cell Sorting (FACS) and studied at the molecular level, which is otherwise extremely difficult because of their paucity in the skin. Another important utility of the iDct-GFP mouse model is that, when used in conjunction with a melanoma mouse model, it allows production of “green melanomas,” such that the melanoma cells can be FACS-purified without contamination of normal stromal and immune system-related cells, which routinely skew downstream molecular analysis of whole melanoma tissues. Furthermore, this model also allows isolation of normal melanocytes from non-tumor skin tissue to be used as normal controls, as whole skin is not an appropriate normal control for melanoma [43]. The Dct-rtTA mouse can also be used to direct melanocyte-specific expression of any TRE-driven transgene. The Tet-inducibility in this mouse is very tight with no “leakage” of transgene expression in the absence of doxycycline [43]. Another Dct-driven rtTA mouse has been reported, although the transgene expression may not have been as tight [45].

Fucci2a is a bicistronic version of the Fucci (Fluorescent Ubiquitination-based Cell Cycle Indicator) cell cycle reporter, incorporating bidirectional transgene driving mCherry-hCdt1(30/120) and mVenus-hGem(1/110) using a fragment of the mouse Rosa26 promoter (R26Fucci2aR). Cre-mediated activation of the transgene allows cell cycle-regulated expression of mCherry (red) in G1 phase, and mVenus expression (green) in S/G2/M phases of the cell cycle [46]. Mort et al. have used Tyr::Cre-mediated expression of Fucci2a in melanocyte-specific manner to quantify proportions of melanoblasts in the three cell cycle phases during embryonic skin samples [46]. The same group has devised another transgenic mouse model (Rosa26::mT/mG) that allows expression of tdTomato (red) in all cells under the control of Rosa26, but upon Tyr::Cre-mediated recombination and excision of tdTomato, EGFP (green) is expressed exclusively in melanocytes [47]. Both of these models are powerful tools to fluorescently label melanocytes for imaging and isolation purposes. However, the activation of the fluorescent label is not reversible, as it is in the case of Tet-inducible system.

Melanoma Mouse Models

Mouse models have been incredibly useful in delineating the roles of different oncogenes and tumor suppressor genes in the etiology of melanoma, and have also provided valuable tools as preclinical models (Fig. 3). Genetic manipulations have been combined with chemical and environmental carcinogens to address pressing questions regarding susceptibility and gene-environment interactions. Here we discuss the most important features of the mouse models of cutaneous melanoma, with particular emphasis on the gene(s) that have been manipulated to induce and/or enhance melanomagenesis (see Table 1).

Fig. 3.

The main pathways altered in GEMMs of cutaneous melanoma. The principal oncogenic modifications introduced in mouse models of melanoma involve the activation of RAS/RAF/MEK/ERK and phosphoinositide-3 kinase (PI3K) pathways, the most frequently altered pathways in human melanomas. Stimulation of receptor tyrosine kinases (RTK) by binding of growth factors (GF) leads to the activation of these pathways resulting in induction of proliferation and/or migration and invasion. Alternatively, RAS/RAF/MEK/ERK pathway can be induced by direct activation of RAF proteins by G protein-coupled receptors (GPCR). Deletion of the tumor suppressor phosphatase and tensin homolog deleted on chromosome 10 (PTEN) can also indirectly activate PI3K pathway. In addition, cell cycle progression can be induced by the oncogenic activation of cyclin-dependent kinase 4 (CDK4) or the inactivation of the tumor suppressors cyclin-dependent kinase inhibitor 2A (CDKN2A) or p53. The specific allelic modifications available in mouse models of melanoma are highlighted in red (also see Table 1).

Table 1.

Genetically engineered mouse models of cutaneous melanoma

Promoter	Genetic Modification	Carcinogen	Latency	Penetrance	Reference(s)
Tyr	SV40 transgenic	UVB	57 wk	80%	53
Tyr	HRasG12V, Ink4a/Arf−/−	None	5.5 mo	61%	57
Tyr	HRasG12V	DMBA	45 wk	100%	58
Tyr	HRasG12V	UVB	45 wk	20%	58
Tyr	HRasG12V, p53+/−	None	65 wk	12%	59
Tyr	HRasG12V, p53−/−	None	17 wk	26%	59
Tyr	HRasG12V, p16Ink4a−/−	None	~75 wk	40%	60
Tyr	HRasG12V, p19Arf−/−	None	~60 wk	~60%	60
Tyr	NRasQ61K	None	>1 yr	29%	61
Tyr	NRasQ61K, Ink4a/Arf+/−	None	1 yr	83%	61
Tyr	NRasQ61K, Ink4a/Arf−/−	None	6 mo	94%	61
Tyr::CreERT2	Nras61R/61R; p16Ink4alox/lox	None	26 wk	70%	63
Tyr::CreERT2	Nras61R/61R; p16Ink4alox/lox; Lkb1lox/lox	None	22 wk	85%	63
Tyr-rtTA	TRE-NrasQ61K; Cdkn2a−/−	None	15 wk	50%	64
Tyr::CreERT2	KrasG12D; p53lox/lox	None	39 wk	100%	65
Tyr::CreERT2	KrasG12D; Lkb1lox/lox	None	39 wk	100%	65
Tyr::CreERT2	KrasG12D; p53lox/lox; Lkb1lox/lox	None	11 wk	100%	65
Tyr::CreERT2	Kras12D/WT; p16Ink4alox/lox	None	36 wk	76%	63
Tyr	BrafV600E, Ink4a/Arf−/−	None	~14 wk	~95%	67
Tyr	BrafV600E, p53−/−	None	~21 wk	~80%	67
Tyr::CreERT2	BrafCA, p16Ink4a−/−	None	7 mo	80%	68
Tyr::CreERT2	BrafCA, Cdkn2alox/lox	None	14 wk	100%	69
Tyr::CreERT2	BrafCA, Cdkn2alox/lox	None	9 wk	100%	69
Lkb1lox/lox	Tyr::CreERT2; Ptenflox/floxCdkn2alox/lox	None	9 mo	100%	76
Tyr::CreERT2	BrafCA	UVB	5.3 mo	100%	71
Tyr::CreERT2	BrafCA, Trp53R172H	None	3.5 mo	100%	71
Tyr::CreERT2	BrafCA; Ptenflox/flox	None	5 wk	100%	29
Tyr::CreERT2	BrafCA, Ptenflox/flox; Bcat-STA	None	3 wk	100%	77
MT1	HGF/SF-Tg	None	15–21 mo	22%	84
MT1	HGF/SF-Tg	UVA/UVB	25 wk	25–31%	11, 12
MT1	HGF/SF-Tg	UVB	18 wk	56%	11, 12
MT1	HGF/SF-Tg, Ink4a/Arf+/−	UVB	25 wk	66%	86
MT1	HGF/SF-Tg, Ink4a/Arf−/−	UVB	7 wk	75%	86
MT1	HGF/SF-Tg	DMBA-TPA	30 wk	100%	20
MT1	HGF/SF-Tg, Cdk4R24C/R24C	DMBA-TPA	12 wk	100%	20
MT1	HGF/SF-Tg, Cdk4R24C/+	DMBA-TPA	14–16 wk	100%	20
	Cdk4R24C/R24C	DMBA-TPA	20 wk	100%	90
MT1	HGF/SF-Tg, Cdk4R24C/R24C	UVB	33 wk	100%	91
MT1	Ret	None	18 wk	65%	92
Dct	Grm1	None	87 wk	100%	95
Dct-rtTA	GnaqQ209L, Ink4a/Arf−/−	None	32 wk	50%	100
Dct-TVA	RCAS-NrasQ61R-IRES-Cre, Ink4a/Arfflox/flox	None	8 wk	63%	103
Dct-TVA	BrafCA; Cdkn2alox/lox; Ptenlox/lox	None	8 wk	100%	104

wk, weeks; mo, months; yr, year. See text for more details.

Gene name: Tyr, Tyrosinase; CreER^T2, tamoxifen-dependent Cre recombinase 2; MT1 (Mt1), metallothionein 1; Dct, dopachrome tautomerase; rtTA, reverse tetracycline-controlled trans activator; TVA, avian leukosis virus receptor; SV40, simian vacuolating virus 40 T antigen; HRas (Hras), Harvey rat sarcoma virus oncogene; NRas (Nras), neuroblastoma ras oncogene; Ink4a/Arf (Cdkn2a), cyclin-dependent kinase inhibitor 2A; p16^Ink4, Cdkn2a isoform p16; p19^Arf, Cdkn2a isoform p19; BRAF (Braf), Braf transforming gene; Braf^CA, Cre-activated Braf^V600E allele. HGF/SF (Hgf), hepatocyte growth factor; Cdk4, Cyclin D kinase 4; Ret, ret proto-oncogene; Grm1, metabotropic glutamate receptor 1; Gnaq, guanine nucleotide-binding protein G_q subunit alpha. RCAS is a proviral vector and IRES refers to internal ribosome entry sequence.

CDKN2A

One of the first clues to melanoma susceptibility was obtained through linkage analysis in familial human melanoma, with identification of the cyclin-dependent kinase inhibitor 2A (CDKN2A) locus as frequently mutated in melanoma [48]. The CDKN2A locus codes for two physically overlapping tumor suppressors via alternative promoters and reading frames, p16INK4A and p14ARF, both of which are negative regulators of the cell cycle. p16INK4A inhibits the cyclin D-CDK4/6 interactions and prevents phosphorylation of RB, which blocks the G1 to S phase progression [49]. p14ARF (p19Arf in mouse) promotes p53-dependent cell cycle arrest by facilitating the degradation of HDM2 [50, 51]. Thus, in effect, loss of function mutations at the CDKN2A locus can simultaneously inhibit the p53- and RB-mediated cell cycle regulation, leading to aberrant cellular proliferation. Bradl et al. used melanocyte-specific transgenic expression of the SV40 T-antigen under the control of the Tyr promoter to generate a mouse model of melanoma (Tyr-SV40E) [52]. Expression of SV40 T-antigen inhibits both the p53 and RB pathways, thus mimicking loss of the CDKN2A locus. This mouse model produced metastatic ocular melanomas with complete penetrance in high transgene expresser lines within 4 weeks, with infrequent superficial dermal melanomas (<10%) [52]. In contrast, moderate transgene expresser lines did not produce fatal ocular melanoma until 6 months of age. Klein-Szanto et al. UVB-irradiated a moderate expresser Tyr-SV40E transgenic line at 4 days of age, for 2 to 4 consecutive days, with grafting at 20 weeks of age on to low transgene expresser recipient mice to allow prolonged observation [53]. These mice produced melanomas with up to 80% penetrance by 57 weeks, commensurate with the number of UVB exposures [53]. Mice with germline knockout of the Cdkn2a locus, which eliminated both p16Ink4a and p19Arf, were found to be susceptible to multiple tumor types, mostly sarcomas and lymphomas, but no melanomas were produced [54]. However, the Cdkn2a-null background synergizes with multiple other genetic modulations to enhance melanomagenesis in mouse models (see below and Table 1).

RAS

Upregulation of the RAS/RAF/MEK/ERK pathway has emerged as an obligate event in the etiology of a large majority of melanomas (Fig. 3). Activating mutations in BRAF (51–63%) and NRAS (21–28%) are mutually exclusive, and together account for as many as 75–80% of melanomas [55]. Mutations in HRAS and KRAS are much rarer, featured in only ~2% of melanomas. The causal role of RAS oncogene mutations in melanomagenesis has been tested via several mouse models (Table 1).

In 1995, Broom-Powell et al. generated a transgenic mouse model that overexpressed the mutant human HRAS^G12V under the control of the Tyr promoter (TP-Ras). This model produced hyperpigmentation and hyperproliferation of melanocytes, but failed to transform them to melanoma [56]. In 1997, Chin et al. reported that melanocyte-specific expression (under Tyr promoter) of HRAS^G12V in the homozygous Ink4a/Arf-null background leads to generation of spontaneous cutaneous melanomas with 61% penetrance [57]. Fifty percent of the mice succumbed to tumors by 5.5 months of age, but no metastases were observed. No cutaneous melanomagenesis was noticed in these mice in the Ink4a/Arf^+/− background. In 1999, Powell et al. used their TP-Ras model to show that topical application of DMBA induced metastatic melanoma at 86–100% penetrance depending on the duration of treatment [58]. 12-O-tetradecanoylphorbol-13-acetate (TPA) failed to produce melanomas, but produced papillomas at low percentage. Intriguingly, a DMBA-TPA treatment protocol for 38 weeks produced melanomas in only 44% of mice. UVB irradiation produced melanomas in 20% of the mice [58]. In 2001, Bardeesy et al. showed that Tyr-HRas^G12V cooperates with loss of p53 to promote melanoma development [59]. Tyr-HRas^G12V, p53^+/− and Tyr-HRas^G12V, p53^−/− mice produced non-metastatic cutaneous melanomas at 12% and 26% penetrance, with average latencies of 65 and 17 weeks, respectively. Sharpless et al. used the Tyr-HRas^G12V model to dissect the individual tumor suppressor contribution of the p16Ink4a and p19Arf, and showed that homozygous loss of either gene was sufficient to drive melanomagenesis, with p19Arf loss having a greater effect [60].

Human melanomas have been shown to carry two activating mutations in the NRAS oncogene; 84% of these mutations are localized to codon 61 (NRAS^Q61R), while only 7% to codon 12 (NRAS^G12D). Ackerman et al. generated a mouse model with transgenic expression of the constitutively active mutant NRas^Q61K under the Tyr promoter [61]. These mice produced cutaneous melanomas at a low penetrance of 29% and a long latency of >1 year. However, when crossed into the Ink4a/Arf-null background, melanomagenesis was significantly accelerated; Tyr-NRas^Q61K, Ink4a/Arf^+/− and Tyr-NRas^Q61K, Ink4a/Arf^−/− mice produced melanomas at 83% (latency 1 year) and 94% penetrance (latency 6 months), respectively, and 36% of them aggressively metastasized to lungs and liver [61]. Pederson et al. have reported an inducible Tyr::CreER^T2-driven NRas^G12D transgenic model [62]. Upon tamoxifen treatment at 2 months of age, these mice exhibited melanocytic proliferation and formation of human blue nevi-like lesions, but failed to produce cutaneous melanomas. When crossed with non-inducible Tyr::CreA mice, so that the NRas^G12D transgene was activated in developing embryonic melanoblasts, leptomeningeal melanoma formed at a median of 4 months in homozygous mice and at 12.5 months in heterozygous mice [62].

Burd et al. have tested the relative contribution of the activated NRAS oncogenes to produce melanoma [63]. Using the Tyr::CreER^T2 system, they conditionally knocked in the Nras^61R and Nras^12D mutations with simultaneous deletion of the p16Ink4a^lox/lox locus. With p16Ink4a deleted, Nras^12D/12D mice failed to generate melanoma efficiently (only 3% penetrance), but Nras^61R/61R mice produced melanomas in 70% of the mice with a median latency of 26 weeks. Loss of the liver kinase B1 (Lkb1) gene further exacerbated melanomagenesis in cooperation with Nras^61R/61R to 85% penetrance with latency of 22 weeks [63].

Kwong et al. used the Tet-inducible system to construct a Tyr-rtTA;TRE-Nras^Q61K;Cdkn2a^−/− mouse model in which melanoma developed upon induction of Nras^Q61K in 50% of the mice by 15 weeks [64]. However, the tumor regressed completely when doxycycline was withdrawn, which provided evidence of NRAS oncogene addiction.

Activating mutations in the KRAS oncogene are found very infrequently in human melanoma. Its weak oncogenicity in melanomagenesis is exemplified by the fact that the Tyr::CreER^T2-activated Kras^G12D mutation does not induce melanoma in mice [65]. However, when the Kras^G12D mutation was combined with loss of p53 or Lkb1 via the Cre/Lox system, melanomagenesis occurred in 100% of the mice with a mean latency of 39 weeks. Simultaneous loss of both p53 and Lkb1 cooperated to decrease the latency to only 11 weeks [65]. The Kras^G12D mutation also led to melanomagenesis in the p16Ink4a-null background with 76% penetrance and median latency of 36 weeks [63].

BRAF

BRAF^V600E is the single most common mutation found in human melanomas. This mutation results in constitutive activation of BRAF and its downstream pathway members. Tyr::Cre-mediated expression of BRAF^V600E in developing melanoblasts was shown to have lethal developmental defects and death at or just before birth [66]. Activation of expression of BRAF^V600E in adult skin was reported to produce benign melanocytic lesions with features of senescence, but no melanomas [29, 67]; however, one group did report melanoma production with long latency [68]. Ink4a/Arf loss was shown to cooperate with BRAF^V600E to significantly increase the penetrance and reduce the latency of melanomagenesis [67, 68]. Goel et al. expressed the mutant BRAF^V600E under the direct control of the Tyr promoter/enhancer region, which produced hyperpigmentation, benign melanocytic hyperplasia, and depending on the level of transgene expression, dermal melanomagenesis with very long latency [67]. Aptly, loss of the Cdkn2a or p53 locus dramatically increased incidence of melanomagenesis and decreased lifespan [67].

Dhomen et al. used Tyr::CreER^T2 to activate BRAF^V600E (Braf^CA), which produced skin hyperpigmentation, formation of nevi, and hypopigmented dermal tumors in 54% of the mice within 12 months. When a p16Ink4a-specific knockout (but intact p19Arf) background was introduced, 80% of the mice produced melanomas within 12 months, and the mean latency dropped to 7 months [68]. Knocking out the whole Cdkn2a locus (with both p16 and p19 deleted) increased the penetrance to 100% with a drastically reduced latency of <100 days [69]. Furthermore, Tyr::CreER^T2-mediated deletion of Lkb1 in Braf^CA mice abrogated oncogene-induced senescence, but failed to fully transform melanocytes; however, additional loss of Cdkn2a rapidly led to melanomagenesis with median latency of <60 days [69]. Other genetic drivers of melanoma have been identified by the screening results of Sleeping Beauty Insertional Mutagenesis in a Braf^V600E melanoma mouse model [70].

Using their previously generated Tyr::CreER^T2, Braf^V600E model [68], Viros et al. have shown that these mice are susceptible to accelerated melanomagenesis from UV irradiation at a low weekly dose (160 mJ/cm²) for up to 6 months [71]. All of these GEM mice developed melanomas within 7 months at a median latency of 5.3 months, as compared to 70% of mice at a median latency of 12.6 months without UV irradiation. Tumor multiplicity was also enhanced from an average of 0.9 to 3.5 tumors per mouse. These tumors were significantly associated with UV signature C-to-T transition mutations in Trp53 tumor suppressor gene. When one of these mutations (Trp53^R172H) was introduced through lox-stop-lox system (Trp53^{+/LSL−R172H}) along with the Braf^V600E, all the mice developed increased number of melanomas within 3.5 months [71]. These results validate a role for UV-induced mutations of Trp53 in melanomagenesis.

PTEN

PTEN was identified to be a frequently mutated or deleted tumor suppressor gene in more than 40% of melanoma cell lines and in 38% and 58% of primary and metastatic melanomas, respectively [72, 73]. PTEN is a dual (lipid/protein) phosphatase that has been implicated in the regulation of embryonic development, cell adhesion, migration, apoptosis, stem cell growth and differentiation. The primary tumor suppressor functions of PTEN are via downregulation of AKT activity and upregulation of the pro-apoptotic machinery, the combined effects of which are cellular survival, aberrant cell growth, cell spreading and migration [74]. The critical role of PTEN as tumor suppressor in melanoma was validated by Sleeping Beauty insertional mutagenesis in an oncogenic BRAF-induced mouse model of melanoma, which identified ceRNAs required for maintaining PTEN expression and tumor suppression [75]. Interestingly, even in the absence of an oncogenic stimulus, Tyr::CreERT2-mediated simultaneous deletion of Pten and Cdkn2a loci gave rise to melanomas with complete penetrance within 9 months [76].

Dankort et al. used the Tyr::CreER^T2/Braf^CA system to simultaneously activate BRAF^V600E and delete PTEN (Pten^flox/flox) in cutaneous melanocytes by topical treatment with 4-OHT [29]. While the activation of BRAF^V600E alone only produced benign melanocytic lesions, additional homozygous loss of Pten caused rapid initiation of melanomagenesis and progression to malignancy and metastasis to lymph nodes and distal organs. This phenotype was 100% penetrant within 4–5 weeks [29]. Addition of a stabilized (degradation resistant) version of β-Catenin (Bcat-STA) oncogene significantly reduced the latency to only 3 weeks with metastasis to lung, bowel, and spleen [77].

HGF/SF

The HGF-MET signaling pathway has been convincingly implicated in human melanoma metastasis and resistance to BRAF^V600E inhibition [78–80]. Upregulated expression and activating mutations in the HGF receptor, MET, have been associated with aggressive human melanoma [81, 82]. The HGF/SF-Tg mice express HGF/SF under the control of the widely expressed MT1 gene promoter and feature constitutive activation of the RAS/RAF/MEK/ERK pathway [83]. These mice generate a variety of tumors of mesenchymal and epithelial origin, including cutaneous melanomas, albeit at low frequency (22%) and a long latency with an average age of onset of 15 to 21 months [84]. When adult HGF/SF-Tg mice were UV irradiated three times weekly, non-melanoma tumorigenesis was accelerated but no melanomas formed during the same timeframe [85]. In 2001, Noonan et al. reported that a single exposure of the HGF/SF-Tg mice to UV radiation during the neonatal phase (postnatal 3.5 d), at a dose lower than the total dose administered to the adult mice previously (9.58 kJ/m² vs. 19.16 kJ/m² of UVB/UVA from FS40 sunlamps), was sufficient to induce melanoma [12]. This melanomagenesis was observed with the relatively higher penetrance of 25–31% with a shortened median time to onset of 25 weeks [11, 12]. Filtered wideband UVB exposure through a xenon arc lamp (14 kJ/m²) accelerated melanomagenesis to 56% of the mice with a median onset of 18 weeks [11]. When the HGF/SF-Tg mice were bred into the Ink4a/Arf-null background, melanomagenesis was further enhanced with a median onset of only 50 days in HGF/SF-Tg, Ink4a/Arf^−/− mice (75% penetrance), as compared to 152 days and 238 days in HGF/SF-Tg, Ink4a/Arf^+/− and HGF/SF-Tg, Ink4a/Arf^+/+ mice, respectively [86]. Tormo et al. have reported that neonatal DMBA treatment at day 4 followed by TPA application to the HGF/SF-Tg mice gives rise to melanomas with 100% penetrance by 30 weeks of age [20].

Notably, HGF/SF-Tg mice possess extra-follicular melanocytes that survive in the epidermis and at the epidermal/dermal junction; as a consequence, UV-induced melanomas develop in a “humanized” microenvironment and progress in a manner that is highly reminiscent of human melanomas, i.e. junctional activity with pagetoid spreading [12, 84]. As in the patient population, UV-inducible HGF/SF-Tg melanocytic lesions experience various stages of radial and vertical progression; e.g. nevoid, superficial spreading, nodular, and metastatic.

CDK4

Loss of p16INK4A in melanoma leads to increased cyclin D1-CDK4/6 signaling, which promotes cell cycle via phosphorylation and deactivation of RB1 [49]. CDK4 is mutated or amplified in 5% of human melanomas (at a higher frequency in BRAF, NRAS and NF1 triple-wildtype patients) [87]. An activating mutation in CDK4 (R24C), which leads to abolition of p16INK4A-CDK4 interaction and constitutive activation of CDK4, was identified in melanomas [88, 89]. Homozygous CDK^R24C/R24C knock-in mice produce squamous cell carcinomas in 15% of mice by 23 weeks of age, but no melanomas. DMBA-TPA treatment produces large melanomas by 20 weeks of age with complete penetrance; however, UVB treatment was ineffective [90]. Tormo et al. combined the CDK^R24C/R24C and HGF/SF-Tg mice to show that neonatal DMBA-TPA treatment formed dozens of rapidly growing melanomas, which required sacrifice of all mice by the 12^th week [20]. The HGF/SF-Tg, CDK^R24C/+ mice also produced aggressive melanomas by 14–16 weeks with slightly reduced multiplicity with 100% penetrance [20]. Gaffal et al. reported that HGF/SF-Tg, CDK^R24C/R24C mice exhibit greater susceptibility to UVB-induced melanomagenesis than the HGF/SF-Tg alone in the C57BL/6 background [91]. A single neonatal UVB dose (6 kJ/m²) increased the penetrance of melanomagenesis in the HGF/SF-Tg, CDK^R24C/R24C mice as compared to the HGF/SF-Tg mice, with a decreased median melanoma-free survival of ~33 weeks vs. ~65 weeks, respectively [91].

RET

Transgenic expression of the RET receptor tyrosine kinase gene under the control of the metallothionein (MT1) promoter gave rise to a melanoma model that resembled the human giant congenital melanocytic nevus, which exhibited slow step-wise progression towards metastatic melanoma in the course of more than 12 months [92]. The progression of the lesions was associated with concomitant increase in the Ret transgene expression and activation of MAPKs, c-Jun, and matrix metalloproteinases [92]. Furthermore, the endothelin receptor B (Ednrb) levels were observed to be lower in the melanomas as compared to the benign tumors. When the Ret mice were crossed with Ednrb^+/− mice, although melanomagenesis was delayed, melanomas that did form were more aggressive, metastasized to lung, and led to shorter lifespan [93]. Trp1 promoter-driven expression of Ret in transgenic mice was shown to be restricted to the RPE and led to microphthalmia and tumor formation in the eye [94].

GRM1

The GRM1 gene encodes the metabotropic glutamate receptor 1, and was found to be aberrantly overexpressed in a random genomic insertional mutant mouse line, which was predisposed to develop multiple melanomas [95]. The role of GRM1 in melanomagenesis was confirmed when transgenic expression of Grm1 under the control of Dct promoter was found to promote melanomagenesis. Analysis of human melanoma cell lines and biopsies showed that 40% were positive for GRM1 expression, whereas in normal primary melanocytes expression was undetectable [96]. GRM1 expression has also been shown in uveal melanomas, and this association has been validated by the susceptibility of the Dct-Grm1 transgenic mice to uveal melanocytic neoplasia [97].

GNAQ

Genes encoding the heterotrimeric Gaq family members, GNAQ and GNA11, have been identified to harbor activating mutations in a large majority of uveal melanomas (83%) but only 6% of cutaneous melanomas [98, 99]. Tet-inducible melanocytic expression of the GNAQ^Q209L mutant failed to generate cutaneous melanoma; however, when bred into the Ink4a/Arf^−/− background, >50% of the mice produced cutaneous melanomas with a median latency of ~32 weeks [100].

RCAS/TVA System

Direct retroviral vector-based delivery of genes/constructs of interest in a cell type-specific manner has allowed testing of tumorigenic potential of multiple genes in an accelerated manner. One such system utilizes the replication-competent avian sarcoma (RCAS) long terminal repeat and the tumor virus A (TVA) vector for transgene delivery [101]. Transgenic mice expressing the gene encoding the receptor for subgroup A avian leucosis virus (tva800) are susceptible to infection by RCAS in vivo, which delivers transgenes efficiently and stably to replicating cells [101]. Mice that express TVA under the control of Dct promoter have been generated [102]. VanBrocklin et al. crossed these mice with the Ink4a/Arf-floxed mice to generate Dct::TVA; Ink4a/Arf^lox/lox mice. When Cre recombinase and NRAS^Q61R were delivered utilizing the RCAS viral vectors, 36% of these mice generated melanomas within 12 weeks [103]. The penetrance was further increased to 63% when both NRAS^Q61R and Cre expression was linked through a single-vector delivery [103]. Recently, Cho et al. generated Dct::TVA; Braf^CA; Cdkn2a^lox/lox; Pten^lox/lox mice, which produced melanoma upon RCAS-Cre delivery in 100% of the mice with mean survival of 58 days [104]. Using this model, they identified AKT1 as a promoter of melanoma metastasis.

Concluding remarks and future directions

Over the last two decades, mouse models have played a vital role in advancing our understanding of the molecular underpinnings of the process of melanomagenesis. With ever-increasingly sophisticated molecular technologies being incorporated in mouse modeling and a relatively recent explosion of genomic data leading to identification of rarer driver genes, we are in fertile times to advance our understanding of melanoma genesis, progression and metastasis to unprecedented levels. Several major avenues remain underappreciated and underexplored. For example, the connection between aging and melanoma or the molecular subtypes and heterogeneity of melanoma remain understudied and can profoundly benefit from development of relevant mouse models. Recent advances in deep sequencing has led to the classification of melanomas according to their major driver oncogenes as mutants of BRAF, NRAS, or NF1, or “triple negative” for all three [7]. Future studies in GEM models will determine whether the molecular mechanisms of melanoma initiation are shared or distinct between these categories (Fig. 1). Moreover, the effects of environmental factors on melanoma development can be evaluated by GEM modeling. For example, studies have shown that UVB and UVA may induce melanoma in HGF/SF-Tg mice via different mechanisms [17], and our recent analyses revealed that they exhibit distinct exomic mutation patterns (unpublished data). Such results will help to address the connection between carcinogenic mechanisms and tumor heterogeneity, facilitating the analyses of genomic data from human melanoma.

GEM models can also be used in diagnostic and therapeutic studies. Clinical outcome of melanoma highly depends on the stage of the disease at the time of diagnosis. GEMs will help to distinguish between benign and malignant lesions, and those with high risk of progressing to metastatic dissemination, allowing discovery of diagnostic biomarkers. In addition, melanomas could arise from normal melanocytes without the preneoplastic stage. For example, up to 80% of melanomas don’t show histological signs of a pre-existing nevus. Studies in GEM models could also provide insight into this observation. Finally, GEMs provide a platform for the preclinical testing of drug efficacy and the identification of response-predictive biomarkers and the mechanisms of resistance. A good example is the Tyr::CreER^T2, BRAF^V600E, Pten^flox/flox mouse model, which has contributed extensively to a better understanding of the molecular mechanisms of resistance to BRAF^V600E inhibitors [105, 106]. The exciting development of immunotherapy in recent years has greatly increased the demand of syngeneic mouse models with competent immune system, which is the target of such therapies. Indeed, overall, GEMs and GEM-derived models are currently the only preclinical platform for evaluation and optimization of immunomodulatory therapies [107]. In this regard, engineering the mice to appropriately model a detailed subtypes of melanoma and its immune context becomes more important than ever. Advances in genetic engineering technologies, especially CRISPR, are expected to promote the precision in spatiotemporal control of gene expression and further improve mouse models of melanoma.

Condensed Abstract.

A general overview of the genetically-engineered mouse modeling and a comprehensive review of the currently available mouse models of melanoma have been provided.