Mouse models of uveal melanoma: Strengths, weaknesses, and future directions

Abstract Uveal melanoma is the most common primary malignancy of the eye, and a number of discoveries in the last decade have led to a more thorough molecular characterization of this cancer. However, the prognosis remains dismal for patients with metastases, and there is an urgent need to identify treatments that are effective for this stage of disease. Animal models are important tools for preclinical studies of uveal melanoma. A variety of models exist, and they have specific advantages, disadvantages, and applications. In this review article, these differences are explored in detail, and ideas for new models that might overcome current challenges are proposed.


| INTRODUC TI ON
Uveal melanoma is a rare (estimated incidence of 6 cases per million) and unique subtype of melanoma that arises in the uveal tract of the eye, most commonly in the choroid (Damato & Damato, 2012;McLaughlin et al., 2005). Local interventions, such as radiation therapy and enucleation, are effective at treating the primary tumor (Krantz, Dave, Komatsubara, Marr, & Carvajal, 2017). However, up to half of the patients will develop metastatic disease, predominantly to the liver (Rietschel et al., 2005). For these patients, liver-directed therapy and participation in clinical trials are recommended, but most die from their disease, and median survival is only 10.2 months (Khoja et al., 2019;Kujala, Makitie, & Kivela, 2003;National Comprehensive Cancer Network).
Despite this, great strides have been made in understanding the molecular features of uveal melanoma. In the past decade, the collective work from several groups has led to the identification of important recurrent mutations and overactive signaling pathways in this cancer. Early oncogenic driver mutations occur in a nearly mutually exclusive pattern in the guanine nucleotide-binding protein subunit alpha-q/11 signaling pathway (Field et al., 2018;Moore et al., 2016;Robertson et al., 2017). This includes constitutively active variants of GNAQ and GNA11, which are found in over 90% of cases (Van Raamsdonk et al., 2009. A smaller subset of tumors harbor activating mutations in the G protein-coupled receptor cysteinyl leukotriene receptor 2 (CYSLTR2) or phospholipase C beta 4 (PLCB4) (Johansson et al., 2016;Moore et al., 2016). There is a second node of nearly mutually exclusive mutations that classifies uveal melanomas and affects prognosis. Inactivating mutations are found in BRCA1-associated protein 1 (BAP1), while recurrent point mutations are observed in the eukaryotic translation initiation factor 1A X-linked (EIF1AX) or a splicing factor such as SF3B1 (Field et al., 2018;Harbour et al., 2010Harbour et al., , 2013Martin et al., 2013).
These discoveries were largely enabled by the analysis of patient tumor specimens and have greatly advanced our understanding of the molecular underpinnings of uveal melanoma tumorigenesis and their prognostic significance. Various animal models have likewise been indispensable in elucidating the biology and potential therapeutic vulnerabilities of this cancer (Cao & Jager, 2015;Yang, Cao, & Grossniklaus, 2015). In the past several years, there have been many promising preclinical studies that have used these models to identify novel treatment strategies, several of which are now in the early stages of clinical trials (Vivet-Noguer, Tarin, Roman-Roman, & Alsafadi, 2019;Yang, Manson, Marr, & Carvajal, 2018).
In this review article, we discuss the strengths and weaknesses of existing animal models of uveal melanoma, with an emphasis on mouse models. We also identify unmet needs that will require future model development and refinement. The goal of any animal model of uveal melanoma should be to faithfully recapitulate the processes of tumor initiation, growth, metastasis, and response to therapy as observed in patients with this disease.

| ANIMAL MODEL S OF U VE AL MEL ANOMA
Though the focus of this review is mouse models of uveal melanoma, other species certainly have their advantages. Rabbits (Oryctolagus cuniculus), for example, have large eyes that facilitate the implantation of tumor cells and subsequent monitoring using techniques such as fundoscopy, ultrasound, and magnetic resonance imaging (Bontzos & Detorakis, 2017;Gao, Tang, Liu, Yang, & Liu, 2018). The zebrafish (Danio rerio) is a model organism that has been used more widely in many scientific fields in recent years (Meyers, 2018). Both xenograft (Fornabaio et al., 2018;van der Ent et al., 2014) and transgenic (Mouti, Dee, Coupland, & Hurlstone, 2016;Perez, Henle, Amsterdam, Hagen, & Lees, 2018) zebrafish models of uveal melanoma have been developed. These models are excellent for high-throughput pharmacologic screening and in vivo microscopy. The genetic models have yielded valuable insights into uveal melanoma signaling, such as the establishment of the importance of YAP activation in the initiation of this cancer. However, tumorigenesis in these models required mutation of p53, and metastasis was difficult to assess because of the induction of multiple primary tumors (Mouti et al., 2016;Perez et al., 2018).
Mice (Mus musculus) are the most widely used laboratory animal in the study of uveal melanoma (Cao & Jager, 2015). Their fecundity, gestation time, and size make them the most cost-effective mammalian model (Zuberi & Lutz, 2016). Furthermore, genetic manipulation of mice has produced various strains that are used in many uveal melanoma models. The primary goals of this review are to compare the different types of mouse models of uveal melanoma and propose directions for further development.

| INO CUL ATI ON S ITE S
The majority of murine models of uveal melanoma require the inoculation of cells or tumors into mice. Some uveal melanoma cell lines can be grown subcutaneously, which is convenient for measuring growth and response to therapy. However, others grow poorly subcutaneously but flourish in the tissue from which they were derived (Ozaki et al., 2016). In these cases, orthotopic models are preferable and may better model the human disease. Models of primary uveal melanoma in which the route of inoculation results in growth in the iris, ciliary body, or choroid F I G U R E 1 Routes of injection for orthotopic models of primary uveal melanoma. The needle trajectories for the three most commonly used types of injections are depicted (anterior chamber in orange, suprachoroidal in green, and intravitreal in blue). All three result in growth of cells in the uveal tract and therefore produce orthotopic models of uveal melanoma are considered orthotopic (Figure 1). Inoculation of cells into the anterior chamber of the eye was one of the first techniques developed and reliably produces tumors in the iris that are capable of metastasis (Niederkorn, 1984). In 2000, a suprachoroidal injection technique was described in which cells are deposited into the posterior compartment (not to be confused with posterior chamber) of the eye .
In this approach, the needle is inserted through the limbus and into the choroid. Injected cells occupy the suprachoroidal space and likely spill into the subretinal space and vitreous. This technique is advantageous because it rapidly produces tumors in the choroid and ciliary body, the sites at which uveal melanoma most commonly occurs in patients. Furthermore, it reduces extraocular growth as compared to transconjunctival inoculations and consistently produces distant metastases (Tables 1 and 2a,b). A third type of orthotopic model is intravitreal injection. Although uveal melanoma does not arise in the vitreous humor, this environment is supportive of tumor growth and injected cells mimic human disease by invading and involving the uveal tract (Kilian et al., 2016;Yoo et al., 2016). All three of the above inoculation methods are amenable to combination with enucleation, which allows for longer follow-up and the study of metastatic outgrowth.
The eye is bypassed in some models in order to more quickly and reliably produce large tumors in visceral organs, especially the liver.

| SYNG ENEIC CUTANEOUS MEL ANOMA MOUS E MODEL S FOR S IMUL ATING U VE AL MEL ANOMA
The syngeneic cutaneous melanoma mouse model has been used for decades in uveal melanoma research. In this system, cutaneous melanoma cells are implanted in mice of the same genetic background as the mice from which the line was derived. Although the cell lines used are not uveal in origin, this system allows for the investigation of intraocular growth and metastasis of melanoma cells, as many of these lines metastasize to the liver (Table 1). This mimics the behavior of uveal melanoma in humans and allows for the study of the full metastatic process, including local invasion, intravasation, survival in the blood, extravasation, and growth in distant organs. The ability to examine the interaction between tumor and host cells as the cancer progresses in an immunocompetent animal is arguably the greatest strength of this model. Additionally, recipient mice may be genetically altered in order to study specific contributions of the host in melanoma progression (Lattier, Yang, Crawford, & Grossniklaus, 2013;Stei, Loeffler, Kurts, et al., 2016).
The most widely used syngeneic model is the inoculation of C57BL/6 mice with the B16LS9 cell line, a derivative of the B16 cutaneous melanoma line that was enriched for hepatic metastatic propensity through serial in vivo passaging (Rusciano, Lorenzoni, & Burger, 1994).
This cell line metastasizes to the liver from the eye, and its use has led to valuable insights into the behavior of metastatic melanoma.
For instance, this model was used to show that natural killer cells and  (2001) Tail vein Liver Ma and Niederkorn (1995) OMM1

| XENOG R AF T MOUS E MODEL S OF U VE AL MEL ANOMA
Xenograft models are another widely used approach. As the name implies, cells or tumors from a foreign source are grafted into mice.
Most commonly, human uveal melanoma cell lines are used. The primary advantage of these models is that the cells are derived from pa- Another exciting recent development has been the generation of PDX models from hepatic uveal melanoma metastases (Kageyama et al., 2017). In these models, tumor specimens obtained after surgery or biopsy were surgically implanted into the livers of NOD SCID gamma mice. The authors achieved an 83% engraftment rate and found that the histology, genetics, and proteomics of the implanted tumors resembled corresponding features of patient metastases.
Tumors could also be monitored by CT imaging. PDX models such as these hold promise for preclinical evaluation of experimental therapeutic compounds and the realization of personalized medicine.
Like all models, xenografts have disadvantages. The chief among these is the necessity of using immunocompromised mice. This can partially be avoided by taking advantage of the immune-privileged nature of the anterior chamber of the eye (Niederkorn, 2012).
However, this approach can only be used to study the primary tumor, and the majority of grafts spontaneously regress (Sutmuller et al., 2000). In this new era of immunotherapy, the inability to study the interplay between the tumor and host immune system, especially in sites of metastasis, is a major limitation. In uveal melanoma, this is somewhat tempered by the low response rate of patients to PD-1 and/or CTLA4 inhibition (Algazi et al., 2016;Carvajal et al., 2017).
However, other immunomodulatory pathways and cell types have been implicated in this cancer and are being actively investigated (Dougall, Kurtulus, Smyth, & Anderson, 2017;Robertson et al., 2017;Yang et al., 2016). Mice with humanized immune systems would be ideal recipients for xenograft models of all tumor types. Efforts to create such mice are ongoing but are complicated by, among other issues, graft-versus-host disease and interspecies differences in cytokine specificity (Allen et al., 2019;Wege, 2018). Other criticisms of xenografts, particularly PDX models, include their high cost, low engraftment rate, and low throughput (Siolas & Hannon, 2013). These are valid concerns, and the actual utility of these models in informing the treatment of patients with uveal melanoma will become more apparent in coming years.
Another approach to avoiding artifacts induced by two-dimensional cell culturing is the use of three-dimensional (3D) culture systems. Such "tumor organoid" models now exist for several cancers, including those arising in the colon, breast, and pancreas (Drost & Clevers, 2018;Yang, Sun, Liu, & Mao, 2018). 3D cultures derived from patient tumor specimens can be grafted into mice (patient-derived organoid xenografts) and faithfully match the molecular phenotypes and even treatment responses of the source tumors (Sachs et al., 2018;Vlachogiannis et al., 2018). Some even allow for the study of the tumor microenvironment, as they incorporate stromal cells such as cancer-associated fibroblasts and lymphocytes (Neal et al., 2018). In the uveal melanoma literature, there have been a few reports of 3D cultures in which cells form tumorspheres (Angi, Versluis, & Kalirai, 2015;Lapadula et al., 2019;Valyi-Nagy et al., 2018). Further work is needed to determine the feasibility of generating such cultures from patient tumors and whether these 3D cell models better reflect the biology of their parental tumors. If so, they may serve as superior tools for both in vitro assays and xenograft models.

| G ENE TI C ALLY ENG INEERED MOUS E MODEL S (G EMMS) OF U VE AL MEL ANOMA
The third class of mouse models of uveal melanoma encompasses mice that have been genetically engineered to produce tumors. The primary advantage of these models is that they make it possible to study autochthonous tumorigenesis in an immunocompetent host.
Older models include transgenic mice in which pigment cell-specific promoters of genes such as Tyrosinase drive expression of the SV40 large T antigen or HRAS, although some of these tumors originate from the retinal pigment epithelium rather than the uvea (Kramer, Powell, Wilson, Salvatore, & Grossniklaus, 1998;Syed et al., 1998;Tolleson et al., 2005). In the Tg(Grm1) model, the Dopachrome tautomerase (Dct) promoter controls expression of the metabotropic glutamate receptor to produce both uveal melanoma and cutaneous melanoma (Schiffner et al., 2014). RET-driven GEMMs develop melanocytic neoplasms throughout the body, including in the uveal tract (Eyles et al., 2010;Kato et al., 1998). The major weakness of all of these models is that they are driven by molecular changes not observed in patients with uveal melanoma; this limits their clinical applicability.
In the years since the discovery of GNAQ and GNA11 as the main oncogenic drivers of uveal melanoma, three genetically engineered mouse models using these genes have been published (Table 4). In the first, a Tet-on system was used to induce GNAQ Q209L expression in mice deficient for p16 Ink4a and p19 Ink4b (Feng et al., 2014).
Although over half of the mice developed melanocytic cutaneous lesions by 9 months, there was no report of uveal melanoma. Despite In a different model, the expression of GNAQ Q209L in a lox-stoplox conditional knock-in allele inserted at the Rosa26 locus produced uveal melanoma in 3 months with 100% penetrance (Huang, Urtatiz, & Van Raamsdonk, 2015). Furthermore, it appears that cells from these tumors intravasate into blood vessels and metastasize to the lungs. Mice also developed dermal melanomas and melanocytic neoplasms at other sites, including the leptomeninges and inner ear.
This model uses Mitf-cre to initiate oncogene expression. Lastly, another model in which a similar conditional knock-in allele encoding GNA11 Q209L is activated by the inducible Tyrosinase-creER T2 produced a comparable phenotype, albeit at a later timepoint (Moore et al., 2018). When Bap1 deletion was combined with GNA11 Q209L expression, uveal melanomas were unexpectedly smaller. However, skin melanoma burden increased, as did cellular proliferation of these tumors. Comparative genomics from this model identified RasGRP3 as a critical signaling node upstream of MAPK pathway activation, a finding that had been independently reported by another group that used orthogonal methods (Chen et al., 2017).
These models have shed light on key features of uveal melanomagenesis. First, they demonstrate that GNAQ and GNA11 are potent oncogenes. The deletion of tumor suppressors was not required to form uveal melanoma; indeed, in the second model, the expression of the human GNAQ transgene was only 3.3% of that of the murine wild-type allele as measured by RT-PCR of primary melanocyte cultures from the affected mice (Huang et al., 2015). Second, they illuminate differences between the pigment cell-specific promoters Like other models, these GEMMs are not without their disadvantages. Disease progression is considerably slower than in syngeneic or xenograft models due to the time required for tumor initiation.
A problem specific to the GNAQ Q209L model is the microphthalmia caused by the Mitf-cre allele (Alizadeh, Fitch, Niswender, McKnight, & Barsh, 2008). Additionally, inserting the oncogenes in the Rosa26 locus is somewhat artificial. A model in which an activatable allele is targeted to the endogenous mouse Gnaq or Gna11 locus might better model physiologic expression of these genes. This approach has been successful in generating valuable Braf V600E GEMMs of cutaneous melanoma (Dankort et al., 2007;Mercer et al., 2005).
A serious obstacle encountered in the above uveal melanoma  tool that could be used in conjunction with both existing and new alleles to generate genetic mouse models that would enable the study of the entire disease process of uveal melanoma in vivo.
In addition to this AAV approach, the RCAS-TVA system might achieve similar results. This method has been used to generate numerous GEMMs of cutaneous melanoma (Cho et al., 2015;Kircher et al., 2019;VanBrocklin, Robinson, Lastwika, Khoury, & Holmen, 2010). RCAS subgroup A is an avian retrovirus capable of infecting cells that express the TVA receptor. Dopachrome tautomerase-TVA transgenic mice express this receptor in pigment-producing cells, and this strain could be crossed with one of the conditional knock-in alleles described above. Intraocular injection of RCAS virus that encodes for Cre would then activate oncogene expression in melanocytes of the eye. An advantage of this model over the AAV approach is targeted delivery to cells of interest such that there is no requirement for inclusion of a pigment cell-specific promoter within the virus. This provides more room for genes of interest, which can be linked with Cre within the same viral vector to enable delivery to the same cells. Additionally, high titers of RCAS are easily produced in vitro using the chicken fibroblast DF-1 cell line (Fisher et al., 1999), and there is no need for helper virus in these cells. Retroviruses also permit long-term expression of genes due to genome integration, though this requires that the cells are dividing.
Despite these advancements in genetic models of uveal melanoma, the inherent differences in tumor biology between mice and humans cannot be ignored. Putative metastases that have been observed in published genetic models occur in the lungs, not the liver (Huang et al., 2015;Moore et al., 2018). Additionally, the loss of Bap1 did not enhance the aggressiveness of uveal melanomas; in fact, the ocular phenotype was weaker, and there was no increase in size or incidence of lung lesions compared with mice expressing GNA11 Q209L alone (Moore et al., 2018). The basis for these differences is not understood and merits further investigation. It must also be acknowledged that the chromosomal abnormalities and epigenetic modifications observed in patients with uveal melanoma are nearly impossible to model in a mouse. Thus, while these GEMMs, as well as improved models, will continue to provide valuable insights into the progression of uveal melanoma in vivo, it is unlikely that any one model will fully recapitulate the human disease in all of its intricacies.

| CON CLUS IONS
In summary, although there is no perfect mouse model of uveal melanoma, currently available models have been instrumental in elucidating critical signaling pathways and testing new therapeutic strategies for this cancer. Each type of model has distinctive strengths and weaknesses. Syngeneic models are excellent for the investigation of tumor progression in an immunocompetent host but use cutaneous melanoma cell lines. Xenograft models allow for the study of human uveal melanoma cells and tumors in a living organism, but this does not include the immune response because recipient mice must be severely immunocompromised. Genetically engineered models allow for studies of autochthonous uveal melanoma formation and dissemination, yet tumors in these mice differ from those in patients in terms of molecular complexity and metastatic behavior.
Investigators should leverage the models best suited to address their specific scientific questions. Future model development should aim to overcome current limitations and further enable efforts to investigate uveal melanoma biology and develop therapies most likely to succeed in patients afflicted with this cancer.

ACK N OWLED G EM ENTS
The authors would like to acknowledge Diana Lim for her assistance with graphic design and Sheri Holmen and Amanda Truong for critically reviewing the manuscript.

CO N FLI C T O F I NTE R E S T S
The authors declare no conflicts of interest.