Uveal melanoma is the second most common form of melanoma and the most common primary intraocular malignancy. Until recently, very little was known about the genetics of this aggressive cancer. Mutations in oncogenes and tumor suppressors that are common in other cancers are conspicuously absent in uveal melanoma. In recent years, however, uveal melanoma has begun to yield its secrets, and a fascinating picture is emerging of how it develops and progresses. Mutations in the Gq alpha subunits, encoded by GNAQ and GNA11, appear to be early or perhaps initiating events that require further mutations for malignant transformation. On the other hand, mutations in the BRCA1-associated protein-1 (BAP1) appear to occur later and demarcate a molecular brink beyond which metastasis becomes highly likely. BAP1 mutations can also occur in the germline, leading to a distinctive cancer predisposition syndrome. These mutations appear to be key events that provide the potential for targeted therapy. This article will review the genetic findings in uveal melanoma over the past two decades and suggest important areas for future work.
Uveal melanoma (UM) is the most common primary malignancy of the eye, with an incidence of about 1200–1500 new cases per year in the United States, and it accounts for about 5% of all melanomas (Egan et al., 1988; Ramaiya and Harbour, 2007). UMs can arise anywhere in the uveal tract, comprising the iris, ciliary body, and choroid, and they often involve more than one of these structures. About 5% of UMs are isolated to the iris, and these are often considered separately because of their less aggressive clinical behavior and distinct genetic alterations. Risk factors for UM include light skin color, red or blonde hair, blue eyes, and cutaneous freckles and nevi (Gallagher et al., 1985; Seddon et al., 1990; Tucker et al., 1985; Van Hees et al., 1994). There is a slight male preponderance (Singh et al., 2005a). Some studies have shown an association between UM and increased sun exposure, sunlamp use, and southern latitude (Seddon et al., 1990; Tucker et al., 1985). Unlike cutaneous melanoma, however, the rates of UM have not increased over recent decades (Singh et al., 2011). Thus, the role of ultraviolet light exposure in UM is less clear than for cutaneous melanoma.
UM is similar to other forms of melanoma in its cellular morphology, expression of melanocytic lineage markers, propensity for metastatic spread, and resistance to therapy (Ramaiya and Harbour, 2007). However, UM differs in important ways from other types of melanoma, owing at least in part to its anatomic location within the uveal layer of the eye. UMs are not located within an epithelium, so downregulation of E-cadherin, epithelial-to-mesenchymal transition, and basement membrane invasion are not important steps in UM progression (Onken et al., 2006a). Likewise, the lymphatic structures in the eye are too small for the passage of cells (Yucel et al., 2009), so regional lymphatic spread of UM is extremely rare. Instead, UMs metastasize by hematogenous dissemination. The most common sites of involvement include liver (93%), lung (24%), and bone (16%), with the overwhelming majority presenting initially in the liver (Diener-West et al., 2005). The cause of this marked tropism for the liver remains unknown. Clinical features associated with poor prognosis include larger tumor diameter, ciliary body involvement, and advanced patient age (Augsburger and Gamel, 1990). Histopathologic prognostic factors include epithelioid cell type, inflammatory infiltration, and extracellular matrix patterning (De La Cruz et al., 1990; Folberg et al., 1992; Gamel et al., 1978; Makitie et al., 2001). The mortality rate at 15 years of diagnosis of the primary tumor is about 50% (Kujala et al., 2003), and median survival after detection of metastatic disease is about 9 months (Kath et al., 1993).
Most UMs exhibit a relatively low degree of genomic instability and aneuploidy compared with many other cancer types. One study found that UMs exhibited less than half the genomic instability of breast cancers (Papadopoulos et al., 2002). In another study of 52 UMs, only one tumor showed microsatellite instability (Cross et al., 2003). In two independent studies that examined a total of 180 primary UMs, about two-thirds were diploid, only one-third demonstrated aneuploidy (which was usually limited to a few specific chromosomal changes), and only 2% were tetraploid (Coleman et al., 1995; Karlsson et al., 1995). Thus, the recurring chromosomal abnormalities in UM discussed below are likely to be specific to tumor progression rather than random events.
Loss of part or all of chromosome 1p occurs in about a quarter of UMs (Figure 1) and more often occurs in the context of monosomy 3 (Hausler et al., 2005). 1p loss is one of the few chromosomal abnormalities that provides prognostic information that is independent of chromosome 3 status, with its presence portending decreased disease-free survival (Kilic et al., 2005). Microsatellite analysis of 70 UMs identified the smallest common region of loss on 1p to about 55 Mb at 1p31 (Hausler et al., 2005). No mutations in this region have been identified, but there are several potential candidates such as the Notch pathway members HES2 and HES5, and the p53 homologue TP73 (Kilic et al., 2008).
Gain of 6p and loss of 6q occur in about a quarter to a third of UMs (Figure 1). Both abnormalities are often present in the same tumor, suggesting the formation of an isochromosome 6p (Aalto et al., 2001). 6p gain has received much more attention than has 6q loss, but it is unclear which chromosomal arm is pathogenetically more significant or whether both are important. Overall, 6p gain is associated with a better prognosis than monosomy 3, which has led some investigators to speculate that 6p gain is somehow ‘protective’ against metastasis (Damato et al., 2009; White et al., 1998). However, it seems more likely that 6p gain is associated with better prognosis simply because it tends to occur in the absence of monosomy 3 (Ehlers et al., 2008; Parrella et al., 1999; Prescher et al., 1995). This relative mutual exclusivity of 6p gain and monosomy 3 may represent alternative evolutionary pathways that are available during tumor progression, the former being less likely to eventuate in metastasis than the latter (Landreville et al., 2008; Parrella et al., 1999). Using conventional karyotyping, CGH and FISH, the common region of 6p gain has been narrowed to 6pter–6p21 and of 6q loss to 6q16.1–6q22 (Bott et al., 2011; Speicher et al., 1994). However, no pathogenic mutations in these regions have yet been reported.
8p loss occurs in about a quarter and 8q gain in almost 40% of UMs (Figure 1). 8q gain is statistically associated with metastasis (Sisley et al., 1997) and has attracted a great deal more attention than 8p loss, yet its significance remains elusive. The smallest region of common gain on 8q has been narrowed to the large region 8q23–24→qter (Prescher et al., 1995; Speicher et al., 1994), which contains many potential oncogenes such as MYC, which is amplified in about 30% of UMs (Parrella et al., 2001). Other potential oncogenes in this region include DDEF1 and NBS1 (now referred to as ASAP1 and NBN, respectively), which are overexpressed in UMs with poor prognosis (Ehlers and Harbour, 2005; Ehlers et al., 2005). However, a pathogenetic significance for any of these observations has not been established, and no specific oncogenic mutations on 8q have been reported in UM. Using MLPA, which interrogates a limited number of loci (Van Dijk et al., 2005), 8q status purportedly yielded prognostic information that was independent of chromosome 3 status (Damato et al., 2009). However, this finding has not been corroborated by other investigators using higher-resolution genome-wide techniques such as array CGH (Kilic et al., 2005; Onken et al., 2008a). It is perhaps notable that 8q gain is a poor prognostic factor mainly when it occurs in the context of 8p loss (Bernstein et al., 2005; Onken et al., 2008a), suggesting the formation of an isochromosome 8q (Prescher et al., 1995). This accounts for about a quarter of UMs with 8q gain and occurs almost exclusively in poor prognosis monosomy 3 tumors (Figure 2). Our group found that 8p loss was a more important prognostic factor than 8q gain (Onken et al., 2008a), and we identified LZTS1, located within a minimal deleted region on 8p, as a potential metastasis suppressor gene (Onken et al., 2008a). Thus, it may be that 8p loss rather than 8q gain is more significant, both prognostically and pathogenetically. Further work is needed to determine the role of chromosome 8 abnormalities in UM progression.
We approached this problem using exome capture followed by next-generation sequencing (Harbour et al., 2010). Initially, two UMs were interrogated, both of which were known to be monosomic for chromosome 3 and to have given rise to metastasis. BRCA1-associated protein-1 (BAP1), located at chromosome 3p21.1, was the only gene on chromosome 3 that was mutated in both tumors. Using Sanger sequencing, we went on to find inactivating mutations in BAP1 in 27 of 57 (47%) UMs. These mutations occurred almost exclusively in metastasizing tumors that had also lost the other copy of chromosome 3, consistent with the ‘two-hit’ model for recessive cancer genes.
BAP1 mutations had previously been identified in a small number of breast and lung cancer cell lines (Jensen et al., 1998) and more recently in malignant pleural mesotheliomas (Bott et al., 2011; Testa et al., 2011), cutaneous melanoma (Wiesner et al., 2011), and possibly other cancers such as meningioma (Abdel-Rahman et al., 2011). BAP1 was identified through a screen for proteins that interact with BRCA1 and has been shown to cooperate with BRCA1 in tumor suppression in cultured cells (Jensen et al., 1998). BAP1 has also been shown to be involved in cell cycle regulation through interaction with host cell factor-1 (Machida et al., 2009), which functions as a transcriptional coactivator with E2F proteins during cell division (Tyagi et al., 2007). More recently, it was shown that calypso, the Drosophila homologue of BAP1, as well as human Bap1 protein, removes monoubiquitin moieties from histone H2A in a manner dependent on interaction with ASX (ASXL1 in humans) (Scheuermann et al., 2010). This activity regulates Hox gene expression, suggesting that BAP1 plays a role in transcriptional regulation during development. While the relative importance of these various interactions remains unclear, a crucial role for BAP1’s deubiquinating activity is strongly suggested by several lines of evidence: (i) the requirement for this activity for tumor suppression in cell culture experiments (Ventii et al., 2008) and (ii) most missense mutations directly target the deubiquitinating catalytic domain (Harbour et al., 2010).
Gene expression profiling
Cytogenetic alterations have provided important insights into the pathobiology of UM, but for use in clinical prognostication such markers are susceptible to sampling error owing to significant intratumoral heterogeneity (Damato et al., 2009). Thus, several groups have explored the use of gene expression profiling (GEP) as a potentially more robust prognostic method, as well as for its potential insights into UM pathobiology. In an early study using nylon filter arrays, several genes were found to be differentially expressed in 12 UM cell lines compared with 3 normal melanocyte cultures (Zuidervaart et al., 2003). Using high-density microarrays, another group found that UMs with disomy 3 exhibited a different GEP than those with monosomy 3 (Tschentscher et al., 2003).
Our group went a step further and showed that GEP could classify UMs into two prognostically significant groups using unsupervised clustering techniques without regard to cytogenetic status (Onken et al., 2004). Class 1 tumors had a low risk, and class 2 tumors had a high risk of metastasis. Notably, the prognostic accuracy of this GEP classification outperformed clinical, pathological, and cytogenetic prognostic indicators (Worley et al., 2007), and this has been confirmed by several independent groups (Petrausch et al., 2008; Van Gils et al., 2008). A likely reason for the superiority of GEP over cytogenetic methods for prognostication is that cytogenetic markers are often distributed heterogeneously throughout the tumor and are thus prone to sampling error. In contrast, GEP represents a functional ‘snapshot’ of the tumor’s microenvironment that is less variable across the tumor (Onken et al., 2010). We migrated the GEP to an assay comprising 12 discriminating genes and 3 control genes performed on a microfluidics platform that could be used on a routine clinical basis on very small samples from fine-needle biopsies (Onken et al., 2006b, 2010). The prognostic accuracy of this assay, and its superiority over chromosome 3 status for clinical prognostic testing, was recently validated in a prospective study involving ten centers across North America (Onken et al., in press).
Aside from its clinical value, gene expression profiling has provided important insights into the pathobiology of UM. The GEP of class 1 tumors closely resembles that of normal uveal melanocytes and low-grade uveal melanocytic tumors, whereas the GEP of class 2 tumors shows reduced expression of melanocytic genes and instead resembles the transcriptome of primitive neural/ectodermal cells (Chang et al., 2008; Onken et al., 2006a). Notably, depletion of BAP1 in cultured class 1 UM cells induced a change in cell morphology to class 2-like epithelioid phenotype and a shift in gene expression to resemble the class 2 GEP (Harbour et al., 2010). Taken together, these findings suggest that BAP1 may play a role in maintaining key aspects of melanocytic differentiation that, when lost, allow malignant progression.
MicroRNA expression profiling can cluster UMs into prognostically significant groups, with one study identifying let-7b and miR-199a as the most significant discriminators (Worley et al., 2008). In other studies, miR-34a inhibited UM cell proliferation and migration through downregulation of c-Met (McGarvey et al., 2008), and miR-137 was found to exhibit tumor suppressor activity through downregulation of MITF and CDK6 (Chen et al., 2011). The pathogenetic relevance of these microRNA alterations in vivo is yet to be determined.
The PI3K/AKT pathway is constitutively activated in a majority of UMs, and phosphorylated AKT correlates with poor prognosis in UM (Saraiva et al., 2005). In a study of nine UM cell lines, mutations in PTEN were not observed (Naus et al., 2000). However, in a much larger study of 75 primary UMs, LOH of the PTEN locus was found in 76% of tumors, and actual mutations within the PTEN coding region were found in 11% of tumors (Abdel-Rahman et al., 2006). PTEN inactivation was also found to be associated with increased aneuploidy and decreased survival in UM (Abdel-Rahman et al., 2006; Ehlers et al., 2008). Taken together, these findings implicate a role for PTEN in UM progression and warrant further work on this subject.
This curious absence of MAPK pathway mutations persisted until the recent discovery of mutations in GNAQ, which encodes the Gαq subunit, in almost half of UMs (Jensen and Rauscher, 1999). Mutant GNAQ was shown to activate the MAPK pathway, although it may have also important effects on other pathways such as the phosphatidylinositol–calcium second messenger system. Attention was drawn to this gene as a result of a forward genetic screen in mice that identified hypermorphic mutations in Gnaq or its paralog Gna11, which act through the melanocyte lineage factor Ednrb, as a cause of increased numbers of intradermal melanocytes (Van Raamsdonk et al., 2004). A subsequent study found that 83% of UMs contained mutations in either GNAQ or GNA11 affecting either Q209 or R183 in a mutually exclusive pattern (Van Raamsdonk et al., 2010). These mutations lead to constitutive activation of the Gαq and Gα11 subunits by abrogating their intrinsic GTPase activity required to return them to an inactive state.
GNAQ/11 mutations are found in benign uveal nevi and in the vast majority of UMs regardless of cytogenetic status, GEP class, or BAP1 status (Bauer et al., 2009; Jensen and Rauscher, 1999; Onken et al., 2008b). Further, these mutations are not sufficient for full malignant transformation to melanoma (Van Raamsdonk et al., 2009). This would seem to place GNAQ/11 mutations as early or perhaps initiating events in UM progression. On the other hand, BAP1 mutations are seen almost exclusively in metastasizing class 2 tumors with monosomy 3 (Harbour et al., 2010), suggesting that this mutation occurs relatively late in the primary tumor and may represent a rate-limiting step in metastasis. Either BAP1 mutation or loss of chromosome 3 can occur first, but both events appear to be necessary for the tumor to acquire the metastasizing class 2 phenotype (Harbour et al., 2010). These early and late mutational events allow a tentative outline of UM progression to be constructed (Figure 3).
Genetic comparison of uveal and cutaneous melanomas
Several groups have compared the cytogenetic alterations in uveal and cutaneous melanomas (Bastian et al., 1998; Curtin et al., 2005; Hoglund et al., 2004). Some of the most common chromosomal alterations in UM – 1p loss, 1q gain, 6p gain, 6q loss, 8p loss, 8q gain, and 11q loss – are also common in cutaneous melanoma (Figure 1). Monosomy 3, the most common change in UM, is also seen in cutaneous melanoma, although at a lower frequency. Likewise, loss on 9p and 10, which are very common in cutaneous melanoma, are also seen in UM, albeit not as often. Greater differences arise at the level of individual gene mutations. Activating mutations in BRAF and NRAS are common in some types of cutaneous melanoma, but are distinctively rare in UM (Cruz et al., 2003; Curtin et al., 2005; Davies et al., 2002; Mooy et al., 1991; Rimoldi et al., 2003; Soparker et al., 1993; Weber et al., 2003; Zuidervaart et al., 2005). Likewise, activating mutations in GNAQ/GNA11 occur in the vast majority of UMs, but are rare in cutaneous melanoma (Van Raamsdonk et al., 2010). Nevertheless, all of these mutations appear to have in common the constitutive activation of the MAPK pathway (Weber et al., 2003). BAP1 mutations, which are strongly linked to metastasis in UM (Harbour et al., 2010), also occur in cutaneous melanoma (Wiesner et al., 2011), but it is unclear whether these mutations play the same role in the latter as they do in the former. Taken together, there are clearly genetic differences between uveal and cutaneous melanoma, but there are also remarkable similarities, with many seemingly divergent features potentially having similar effects at the molecular and cellular level.
Familial uveal melanoma
Traditionally, it has been thought that familial UM is extremely rare (Singh et al., 1996b). A few studies have suggested a link between UM and breast cancer, possibly as a consequence of germline BRCA2 mutations (Iscovich et al., 2002; Scott et al., 2002; Sinilnikova et al., 1999). In one study, constitutional DNA samples were analyzed for BRCA2 mutations in 62 patients with UM who were selected primarily on the basis of a family history of breast cancer or UM harbored; three (4.8%) patients harbored BRCA2 sequence variants that were judged to be potentially deleterious (Sinilnikova et al., 1999). An Israeli study identified 4/143 (2.8%) patients with UM who carried a germline 6174delT BRCA2 mutation (Iscovich et al., 2002). However, as this alteration is prevalent in the Israeli population, the pathogenetic relevance of this finding is unclear. An Australian study found germline BRCA2 mutations in 2/71 (2.8%) patients with UM, but neither of these patients had a positive family history, thus leaving open the possibility that these were silent polymorphisms (Scott et al., 2002).
Given the rarity of familial UM in the literature, we were surprised to find that one of the patients with UM in our original study carried a germline BAP1 mutation that was reduced to homozygosity in the tumor by loss of the other copy of chromosome 3 (Harbour et al., 2010). We have identified another family from our ocular oncology center in which UM and cutaneous melanoma occurred in multiple family member in association with a germline BAP1 mutation (author’s unpublished data). The scope of tumors associated with this emerging BAP1 cancer predisposition syndrome has now been expanded to include malignant mesothelioma, cutaneous melanocytic tumors, and other cancers (Testa et al., 2011; Wiesner et al., 2011).
Thus, while familial UM is indeed uncommon, representing perhaps 2-5% of patients with new UM, it is not as rare as once believed. The reason that familial UM was not recognized more commonly in the past may be because of a reduced penetrance for UM in these families (Testa et al., 2011; Wiesner et al., 2011). This would be consistent with previous studies that found that familial UM rarely involved more than 2–3 family members and was more likely to occur in larger families (where reduced penetrance would more likely be recognized; Abdel-Rahman et al., 2011; Canning and Hungerford, 1988; Singh et al., 1996b; Young et al., 1994). A possible explanation for the reduced penetrance is that BAP1 inactivation appears to be a relatively late event in melanoma progression and requires an initiating event, such as an activating mutation GNAQ/11 in UM or BRAF in cutaneous melanoma, and loss of the other copy of BAP1 in order for the germline BAP1 mutation to become manifest. In support of this idea, a cutaneous melanocytic tumor associated with germline BAP1 mutation also harbored mutant BRAF, and only the portion of the tumor that had lost the other copy of BAP1 progressed to melanoma (Wiesner et al., 2011). Similarly, our reported case of UM associated with a germline BAP1 mutation harbored a GNAQ mutation and had lost the other copy of chromosome 3 (Harbour et al., 2010).
Implications for targeted therapy
While excellent local therapies exist for treating the primary ocular tumor, there are no consistently effective therapies for metastatic UM (Augsburger et al., 2009). The discovery of GNAQ/11 and BAP1 mutations in UM provides an unprecedented opportunity for targeted therapy of metastatic disease. Nevertheless, molecular targeting of these mutations will pose significant challenges.
For GNAQ/11 mutations, the therapeutic goal is to inhibit oncogenic downstream signaling resulting from these mutations. Direct inhibition of mutant Gαq or Gα11 may prove difficult, however, because these mutations abrogate the intrinsic GTPase activity that would normally allow these proteins to return to their GDP-bound, inactive state. This is similar to the longstanding problem targeting mutations in RAS family oncogenes, which also rely on intrinsic GTPase activity to terminate signaling (Diaz-Flores and Shannon, 2007). An alternative strategy is to inhibit downstream signaling molecules that are activated by GNAQ/11 mutations. One such target is MEK (Mitsiades et al., 2011), a key component of the MAPK mitogenic pathway that is activated by GNAQ/11 mutations (Jensen and Rauscher, 1999). Other potential targets that are currently being investigated include phospholipase C (PLC), which is activated by Gq, and protein kinase C (PKC), which is activated downstream of PLC (Patel et al., 2011).
Therapeutic targeting of BAP1 mutations poses different challenges. First, as BAP1 acts as a recessive cancer gene, the goal of therapy is to restore one or more functions of BAP1 that are lost when it is inactivated. This is technically more challenging than inhibiting an overactive oncogene. Second, it remains unclear which function(s) of BAP1 is responsible for its anticancer role. Nevertheless, it seems likely that this role is dependent on BAP1’s deubiquinating activity, as discussed in a previous section. Further, a major function of this deubiquinating activity appears to be the removal of monoubiquitin moieties from histone H2A, which alters local chromatin structure to regulate transcription. Thus, a compound that inhibits H2A monoubiquitination may at least in part offset the biochemical, and consequently cellular, effects of BAP1 loss. Histone deacetylase (HDAC) inhibitors represent one such class of compounds. Through inhibition of the Bmi1/Ring1 complex, which monoubiquitinates H2A (Bommi et al., 2010), HDAC inhibitors such as valproic acid, trichostatin A, LBH-589, and suberoylanilide hydroxamic acid can reverse the H2A hyperubiquitination that occurs in cultured UM cells depleted of BAP1, and this is accompanied by increased melanocytic differentiation, cell cycle exit, and a shift from class 1 to class 2 GEP (Landreville et al., 2012). This study also showed antitumor activity of valproic acid in vivo in an animal model of UM. Thus, HDAC inhibitors may have therapeutic potential in UM, and clinical trials are currently being planned.
In recent years, there has been tremendous progress in understanding the genetics of melanoma in general and UM in particular. Activating mutations in GNAQ/11 appear to represent a very early or initiating event, whereas inactivating mutations in BAP1 appear to demarcate a threshold in tumor progression beyond which metastasis and death await. The opportunities for targeted therapy afforded by the discovery of GNAQ/11 and BAP1 mutations are being explored. Therapies based on GNAQ/11 status are underway for metastatic UM. A greater understanding of the normal function of BAP1 in the melanocyte lineage and the adverse effects of BAP1 loss are needed to develop agents that specifically target this mutation. Further insights to guide therapy may derive from future research into the genetic events that occur after metastasis.
The author would like to acknowledge Michael D. Onken and Lori A. Worley for critical review of the manuscript, and Justis Ehlers for compiling the chromosomal data presented in Figure 2. Dr. Harbour is funded by grants from the National Cancer Institute (R01 CA125970 and R01 CA16187001), Melanoma Research Foundation, Melanoma Research Alliance, Research to Prevent Blindness Senior Investigator Award, Barnes-Jewish Hospital Foundation, Kling Family Foundation, and Tumori Foundation. Dr. Harbour and Washington University may receive royalties based on a license of related technology by the University to Castle Biosciences, Inc.