Dual Roles of RNF2 in Melanoma Progression

Epigenetic factors offer important new targets for cancer therapy given their crucial role in the regulation of major cancer-relevant transcriptional programs and their potential reversibility (1). Significant effort has been directed toward identifying key epigenetic regulators in certain cancer contexts and elucidating the specific mechanisms, cell biologic processes, and surrogate transcriptional networks governed by these factors. However, we have limited understanding of the roles of epigenetic regulators in melanoma progression.

Melanoma is an aggressive cancer with escalating incidence worldwide (2). Melanoma deaths stem primarily from widespread metastatic disease (2), though the genetic determinants and molecular mechanisms driving this disease remain poorly understood. Recent integrated genomic and functional screening efforts have identified proinvasive determinants of melanoma metastasis with potential prognostic significance (3). The list of 18 prognostic determinants that emerged from this screen was identified based on evidence of proinvasive and oncogenic capabilities in vitro and in vivo, in addition to genomic and expression alterations in human melanomas. On this list of 18, four were known epigenetic regulators: ASF1B (4), HMGB1 (5), RNF2 (6), and UCHL5 (7).

In this study, we focus on RNF2, a component of the polycomb repressor complex-1 (PRC1). RNF2 catalyzes monoubiquitination of lysine 119 of histone H2A (H2AK119ub; refs. 6, 8) and is overexpressed in gastrointestinal tumors, lymphomas, and pancreatic cancers (9, 10). However, it is not known whether RNF2 overexpression is relevant functionally and, if so, what mechanisms, biologic functions, or transcriptional networks are governed by RNF2 in a cancer context. Here, we elucidate the functional and biologic roles of RNF2 in melanoma.

RNF2 was previously identified as a candidate prometastasis oncogene (3). Here, we set out to validate its prometastatic and oncogenic activities and discern functional mechanisms. Metastatic function was assessed in multiple melanoma cell models, including two primary immortalized human melanocyte lines constitutively expressing TERT, p53^DD, CDK4^R24C, and either BRAF^V600E or NRAS^G12D mutant proteins (ref. 11; HMEL–BRAF^V600E and pMEL–NRAS^G12D) and two established human melanoma cell lines, WM115 and 1205Lu. Lentiviral transduction and overexpression of wild-type RNF2 (hereafter RNF2^WT; Supplementary Fig. S1A) promoted invasion in a Boyden chamber Matrigel invasion assay in HMEL–BRAF^V600E, WM115, and 1205Lu cells (Fig. 1A; Supplementary Fig. S1B). Similarly, RNF2^WT enhanced metastatic ability as measured by spontaneous distant metastasis (lung/liver/lymph node) in nude mice with tumor burden of 1.5 cm following intradermal injection of transduced WM115, 1205Lu, and pMEL–NRAS^G12D cells (Fig. 1B).

To complement this approach, loss-of-function studies in the highly invasive human melanoma cell lines 501Mel (harboring high levels of RNF2; Supplementary Fig. S1C) and engineered HMEL–BRAF^V600E melanocyte with stable shRNA targeting PTEN (HMEL–BRAF^V600E–shPTEN) showed significant reduction in invasive potential in vitro upon RNF2 knockdown with two independent shRNAs (Fig. 1C; Supplementary Fig. S1D and S1E). Because proinvasive properties are critical for seeding to distant organs during metastasis (12), we tested if RNF2 was required for seeding to distant organs. Indeed, RNF2 silencing in HMEL–BRAF^V600E–shPTEN cells reduced lung seeding potential (Fig. 1D; Supplementary Fig. S1F). Furthermore, in an immunocompetent C57BL/6 host, knockdown of RNF2 in highly invasive B16-F10 cells similarly reduced lung seeding (Fig. 1E; Supplementary Fig. S1G).

Next, to explore RNF2′s role as an oncogene, we assessed tumor formation following intradermal injection of RNF2^WT-overexpressing HMEL–BRAF^V600E and pMEL–NRAS^G12D melanocytes as well as WM115 and 1205Lu melanoma cells. RNF2^WT significantly increased tumorigenic potential compared with control (Fig. 1F–I; Supplementary Fig. S2A–S2D) in all four cell lines tested. Similar activity of RNF2^WT was observed in cell-based soft-agar colony formation assays, a surrogate for tumorigenesis (Fig. 1J). Reciprocally, shRNA-mediated knockdown of RNF2 in highly tumorigenic 501Mel and WM983B cells, which express high levels of RNF2 (Supplementary Fig. S1C), resulted in a significant reduction in tumor burden (Fig. 1K; Supplementary Fig. S2E–S2G). Consistently, proliferation defects were seen in 501Mel, HMEL–BRAF^V600E–shPTEN, and B16-F10 cells upon RNF2 knockdown (Supplementary Fig. S2H–S2J).

To substantiate the relevance of RNF2 in human melanoma, we verified that RNF2 expression correlates with disease progression at the mRNA and protein levels. Specifically, as summarized in Supplementary Fig. S3A, RNF2 mRNA expression was elevated in primary melanoma tissue compared with skin and nevi (13) and, in an independent cohort, was significantly higher in metastatic lesions when compared with localized primary tumors (Supplementary Fig. S3B). Correspondingly, tissue microarray (TMA) analysis verified progression-correlated expression across 480 cores derived from 170 patients (132 benign nevi cores from 36 patients), 196 primary melanoma cores derived from 59 patients, 60 lymph node metastasis cores derived from 29 patients, and 92 visceral metastasis cores derived from 46 patients (Fig. 2A; Supplementary Fig. S3C). Overall, RNF2 expression was low in normal skin cells, including melanocytes, and progressively increased from nevi to primary melanomas to lymph node metastases.

Leveraging the clinically annotated multidimensional dataset on melanoma generated by The Cancer Genome Atlas (TCGA) Network (14 2013-04-06), we investigated the relationship between RNF2 copy number and expression correlation with cumulative overall survival. Of the 268 samples with copy-number and expression data, we found copy-number gains of RNF2 in 42 samples (15.7%, defined by segmented copy-number value greater than 0.5), copy-number loss in 6 samples (2.2%, defined by copy-number value less than 0.5), and overexpression of RNF2 in 13 of 268 tumors (4.9%, defined by normalized expression z scores greater than 2). Overall, 44 tumors showed copy-number gain or overexpression of RNF2 with overlap of 11 samples (P = 2.5e–8, Fisher exact test), whereas 218 tumors showed neither copy-number change nor expression difference (hereafter referred to as “RNF2 normal”). Further, we found that amplification/overexpression of RNF2 significantly co-occurred with NRAS mutations (OR = 3.2; P = 0.00077) and was significantly mutually exclusive with BRAF mutations (OR = 0.37; P = 0.0046). Survival intervals from date of specimen submission to patients' death or last follow-up were available in 154 cases. Among these 154 cases, we found that, indeed, elevated RNF2 levels were associated with poorer overall survival (log-rank P value < 0.0039; Fig. 2B), confirming the prognostic significance of RNF2 in melanoma.

Given RNF2′s known transcriptional repressor and catalytic activities, we sought to determine whether RNF2′s catalytic activity is required for its proinvasion and protumorigenic phenotypes. Mutant forms of RNF2: RNF2^I53S and RNF2^R70C, shown previously to lack catalytic activity (15, 16), were engineered. We found that as expected, these mutants showed diminished invasion and metastasis activity compared with RNF2^WT (Fig. 1A and B; Supplementary Fig. S1A and S1B). However, to our surprise, both RNF2^I53S and RNF2^R70C mutants retained the capacity to enhance proliferation and anchorage-independent growth in vitro, and tumorigenicity in vivo, at levels comparable with RNF2^WT in all four melanoma/melanocytic cell models (Fig. 1F–J; Supplementary Figs. S2A–S2D and S3D). This observation suggested that RNF2′s protumorigenic potential does not require its catalytic activity. To verify this, we performed rescue experiments with vectors encoding the open reading frames of wild-type and catalytic mutants of RNF2 in WM983B cells wherein RNF2 was silenced with a 3′ untranslated region–directed shRNA. Consistent with the overexpression data in HMEL–BRAF^V600E cells, RNF2^WT and RNF2^I53S expression were similarly able to restore soft-agar colony formation ability (Fig. 2C; Supplementary Fig. S3E).

To address the possible confounding effect of endogenous RNF2 expression in the above study, we engineered a mouse line bearing a conditional RNF2 knockout allele with LoxP sites flanking exon 2 (Supplementary Fig. S3F), where Cre-mediated recombination results in the loss of RNF2 protein expression (Fig. 2D). The RNF2^L/L allele was introduced into an inducible melanoma model called inducible Braf Ink/Arf Pten (iBIP), which harbors the following alleles: Ink4a/Arf^−/−, Tyr-Cre^ERT2, Rosa26-LoxP-Stop-LoxP-Rtta, TetO-Braf^V600E, and Pten^L/L (17). The iBIP mouse model allows temporal and spatial control of tumor development and growth through melanocyte-specific, doxycycline-dependent Braf^V600E activation, restricted to the same cells as those undergoing 4-hydroxytamoxifen (OHT)–dependent Pten deletion in the Ink4a/Arf germline knockout background. Comparison of melanoma tumor burden following topical 4-OHT application in littermate iBIP;RNF2^+/+ and iBIP;RNF2^L/L mice showed that RNF2 deficiency was associated with significant reduction in tumor burden at 14 weeks and improved survival (Fig. 2E; Supplementary Fig. S3G).

Using this genetic system in which Rnf2 can be rendered homozygous null, we reassessed the differential requirement of RNF2 catalytic activity in cellular proliferation and invasion. Specifically, melanoma cells derived from iBIP; RNF2^L/L animals were transduced with lentivirus encoding RNF2^WT and RNF2^I53S (Fig. 2D) and assayed for proliferation and invasion along with the levels of H2AK119ub mark. Consistent with the studies above, RNF2 catalytic activity was dispensable for proliferation enhancement yet required for invasion (Fig. 2F–H). Taken together, these in vitro and in vivo functional assays suggested that, unlike its metastatic function, RNF2′s oncogenic potential is not dependent on its catalytic activity.

To explore the mechanistic basis of RNF2′s cancer-relevant activities, transcriptome profiling (Supplementary Fig. S4A) and ChIP-seq (chromatin immunoprecipitation followed by deep sequencing, performed using V5 antibody) occupancy profiling were performed in HMEL–BRAF^V600E melanocytes with enforced expression of RNF2^WT (hereafter HMEL–BRAF^V600E–RNF2^WT). These RNF2 ChIP-sequencing studies were also conducted in primary tumor cells derived from HMEL–BRAF^V600E–RNF2^WT melanocytes. ChIP-sequencing data analysis showed RNF2-occupied loci exhibited significantly higher enrichment of RNF2 compared with input (Supplementary Fig. S4B) and were evolutionarily conserved among 44 species (Supplementary Fig. S4C). Analyses of the distribution of RNF2 occupancy sites in relation to transcription start sites (TSS) revealed 3,465 genes in +/-5 Kb vicinity of the RNF2 occupied loci in HMEL–BRAF^V600E–RNF2^WT melanocytes (Supplementary Fig. S4D and S4E; Supplementary Table S1). Overlap of expression and occupancy datasets showed that 363 genes, whose promoters were occupied by RNF2 in HMEL–BRAF^V600E–RNF2^WT melanocytes, exhibited altered gene expression upon RNF2^WT overexpression compared with GFP in HMEL–BRAF^V600E cells (Fig. 3A; Supplementary Table S2). Although 47% of these genes with RNF2 occupancy were found to have decreased expression (compared with GFP) consistent with the classic repressive function of the RNF2–polycomb complex, it is worth noting that 53% of RNF2-occupied genes showed increased expression pointing to a likely role for RNF2 in transcriptional activation (see Discussion; Fig. 3A).

Pathway enrichment analysis of RNF2-occupied genes with increased expression showed enrichment in proliferation pathways, in addition to nucleotide synthesis and hypoxia pathways (Supplementary Fig. S4F, top 5 pathways shown), whereas RNF2-occupied genes with decreased expression are associated with regulation of transcription and nucleotide synthesis (Supplementary Fig. S4G, top 5 pathways shown). Among the RNF2-occupied genes exhibiting the most robust alterations in expression were those linked to TGFβ signaling (Fig. 3B), in line with the known role of TGFβ in invasion and metastasis (18). Thus, we next sought to determine whether RNF2 could modulate TGFβ pathway activation. First, we showed that, indeed, overexpression of RNF2^WT, but not RNF2^I53S, enhanced luciferase reporter activity driven by a generic TGFβ-responsive promoter in HEK293 cells (Fig. 3C; Supplementary Fig. S4H) and drove increased expression of TGFβ target genes (ID1, ID2, and ID3) in HMEL–BRAF^V600E melanocytes (Supplementary Fig. S4I). Consistent with a functional role of RNF2-driven TGFβ pathway activation in invasion, treatment of RNF2-overexpressing cells with an inhibitor of the TGFβ pathway (LY2157299; ref. 19) resulted in reduced invasion in Boyden chamber Matrigel invasion assays (Fig. 3D).

To identify candidate direct targets of RNF2 that govern TGFβ pathway activation, gene expression and promoter occupancy profiles were overlaid to define 363 genes (Fig. 3A), among which one of the most significantly changed genes was LTBP2 (Fig. 3E; Supplementary Table S2), which encodes a member of the latent TGFβ binding family of proteins that resides in the extracellular matrix and regulates bioavailability of TGFβ ligand (20) to positively or negatively influence TGFβ signaling (21). This finding gains added significance because LTBP2 is downregulated upon RNF2 overexpression and has been shown previously to inhibit the migration capacity of human melanoma cells (22). Thus, we next performed ChIP–qPCR to examine the LTBP2 promoter for occupancy by RNF2 in accordance with histone H2AK119ub modification. As shown in Fig. 3F, although the LTBP2 promoter was occupied by RNF2 in RNF2^WT, RNF2^R70C, and RNF2^I53S expressing HMEL–BRAF^V600E melanocytes, the H2AK119ub mark was observed only in RNF2^WT, not RNF2^R70C or RNF2^I53S, expressing cells. In other words, the catalytic-dead RNF2 was defective in catalyzing H2AK119ub, and RNF2 enzymatic activity is not required for RNF2 binding at the promoter of LTBP2. Consistent with RNF2 catalytic activity–dependent repression, quantitative RT-PCR confirmed downregulation of LTBP2 mRNA in RNF2^WT, but not RNF2^R70C or RNF2^I53S, expressing cells (Fig. 3G). This was also validated in the human melanoma cell lines 501Mel and WM983B, where we noted RNF2 occupancy in parental cells and loss of H2AK119ub signal in 501Mel and WM983B cells upon RNF2 knockdown (Fig. 3H and I). Consistently, activating histone acetylation marks were enriched on the LTBP2 promoter (Fig. 3H and I), and its mRNA expression was increased upon RNF2 knockdown (Fig. 3J). In addition, LTBP2-mediated modulation of TGFβ signaling is supported by the correlation of LTBP2 knockdown with upregulation of the TGFβ target genes ID1, ID2, and ID3 (Fig. 3K; Supplementary Fig. S4J), as well as with enhanced invasion activity in vitro (Fig. 3L). Finally, the functional epistatic link between RNF2 and LTBP2 is supported by the demonstration that LTBP2 overexpression partially inhibited the invasive activity of RNF2^WT-overexpressing melanocytes (Fig. 3M and N).

As noted above, many genes proximal to RNF2 occupancy sites in HMEL-RNF2^WT melanocytes showed increased expression (Fig. 3A). Indeed, the most significantly upregulated and occupied gene was CCND2, which encodes the cell-cycle regulator Cyclin D2 (Fig. 4A; Supplementary Table S2). ChIP–qPCR confirmed RNF2 occupancy at the CCND2 promoter in HMEL–BRAF^V600E cells overexpressing RNF2^WT, RNF2^R70C, and RNF2^I53S (Fig. 4B). In addition, CCND2 expression was induced by ectopic RNF2 (wild-type or catalytic dead) and remained high in HMEL–BRAF^V600E cells overexpressing RNF2^WT, RNF2^R70C, and RNF2^I53S (Fig. 4C), suggesting that transcriptional activation did not require catalytic activity or histone H2AK119 ubiquitination. Indeed, no enrichment of the H2AK119ub mark was detected on the CCND2 promoter (Fig. 4B), which instead possessed activating chromatin modifications, including H3K9ac, H3K27ac, H4TetraAc, and H3K4me3, in HMEL–BRAF^V600E cells overexpressing both wild-type and catalytic mutants of RNF2 (Fig. 4B). Accordingly, RNF2 knockdown caused repression of CCND2 expression (Fig. 4D) and removal of activation marks in WM983B and 501Mel cells (Fig. 4E and F).

To assess the potential role of RNF2-directed CCND2 upregulation in promoting increased proliferation and tumorigenesis, shRNA-mediated knockdown of CCND2 was performed in HMEL–BRAF^V600E–RNF2^WT-overexpressing cells (Fig. 4G). As shown in Fig. 4H–J, in vivo tumor formation (Fig. 4H), enhancement in two-dimensional proliferative capacity (Fig. 4I), and three-dimensional anchorage-independent growth (Fig. 4J) conferred by RNF2^WT overexpression were partially reversed upon CCND2 knockdown, suggesting that CCND2 contributes to pro-oncogenic activities of RNF2. Consistently, knockdown of CCND2 reduced the proliferative capacity of 501Mel and WM983B cells, which express high levels of RNF2 (Fig. 4K).

Next, we sought to understand how RNF2 might promote gene activation contrary to its known role in gene repression. We considered the possibility that the transcriptional fate of genes regulated by RNF2 might depend on the chromatin states that preexisted on their promoters before upregulation of RNF2. To identify these preexisting chromatin states of RNF2-regulated genes in melanocytes before RNF2 overexpression, we performed epigenomic analyses for 35 histone marks in the HMEL–BRAF^V600E cell system that was used in the RNF2 gain-of-function experiments (Rai and colleagues, unpublished data). There, we modeled histone modification profiles as 45 defined chromatin states using the ChromHMM modeling method (ref. 23; see Methods; Fig. 5A), which captures important biologic states such as poised or bivalent promoter/enhancer states (state 26 and state 6; ref. 24). Each of these chromatin states was annotated based on the enrichment of different histone marks as well as the enrichment of known genomic elements (Fig. 5A; Supplementary Fig. S5A–S5B; Supplementary Table S3). We overlapped RNF2 binding sites to these chromatin states and found that, although all RNF2 binding sites in the genome overlapped with a number of states, the sites that were associated with genes showing altered expression were limited to promoter and poised states (Fig. 5A). Interestingly, we noted that promoters of the genes upregulated by RNF2, including CCND2, were specifically enriched in state 26, whereas RNF2-downregulated promoters were markedly absent (Fig. 5A). These downregulated promoters displayed only active promoter states (states 1–3, 5). Therefore, we compared the cumulative presence of H3K27me3 marks on all upregulated and downregulated RNF2-bound promoters. As shown in Fig. 5B, H3K27me3 was significantly enriched at promoters that are upregulated by RNF2, compared with promoters of genes destined for repression by RNF2 that lack enrichment of this mark. Consistent with this, a UCSC genome browser view of the CCND2 promoter showed prominent peaks of H3K27me3 and H3K4me3, characteristic of state 26 (poised enhancer/promoter) and state 6 (poised promoter), around RNF2 binding sites immediately upstream of the TSS (Fig. 5C). In contrast, analysis of the LTBP2 promoter, which is repressed when RNF2 is expressed, showed enrichment of active promoter states (states 1 and 2) and active promoter marks (H3K4me3 and H3K9ac) as well as enhancer states (states 7, 8, and 9) and enhancer (H3K27ac) marks (Fig. 5D). These data suggest that genes activated by RNF2 may be marked, or poised, by the repression-associated mark H3K27me3 prior to RNF2-mediated activation and gain of histone acetylation marks.

Recently, MLL2, UTX, and p300 were identified as RNF2-associated proteins in mouse ES cells, which co-migrate on a sucrose gradient separately from RNF2-containing PRC1 components (25). This observation suggests that a fraction of RNF2 molecules may exist in an activating complex with MLL2, UTX, or p300. Therefore, we hypothesized that a subfraction of RNF2 may preferentially recruit activating factors to the H3K27me3-containing poised promoters. To investigate this, we first tested whether RNF2 overexpression led to H3K27me3 loss on activated promoters. Indeed, RNF2 overexpression led to loss of H3K27me3 occupancy as well as gain of histone acetylation marks (H3K9ac, H3K27ac, and H4TetraAc) on the CCND2 promoter (Figs. 6A and 4B). These histone modification events upon RNF2 overexpression were consistent with RNF2′s suggested interaction with UTX, an H3K27 demethylase, and p300, a histone acetyltransferase (25). Indeed, ChIP–qPCR showed that UTX and p300 were enriched on the CCND2 promoter after RNF2 overexpression (Fig. 6B). Consistent with these observations, we noted interactions between RNF2 and UTX by coimmunoprecipitation in HMEL–BRAF^V600E–RNF2^WT cells (Fig. 6C). Finally, we tested whether UTX and p300 recruitment by RNF2 had any impact on transcriptional activation of the CCND2 promoter. Downregulation of UTX or p300 individually by shRNAs significantly reduced CCND2 expression in RNF2-overexpressing cells but not in control cells (Fig. 6D). Together, these observations suggest that recruitment of UTX and p300 to the CCND2 promoter by RNF2 is critical for creating an activating chromatin environment as well as transcriptional activation.

To understand how RNF2 might act as both an activator and a repressor in the same cell, we asked whether a particular modified form of RNF2 is important for gene activation. It was previously shown that RNF2 is phosphorylated in a MEK-dependent manner, and this phosphorylation may be associated with histone acetylation events (26). Because RNF2 overexpression was studied in the context of activated MAPK signaling (due to BRAF^V600E mutation), which is known to activate MEK, we asked whether phosphorylation of RNF2 by MEK may be important for its role in gene activation in the context of melanoma. We first verified that MEK1 is indeed able to phosphorylate RNF2 using an in vitro kinase assay (Fig. 7A). Moreover, serine 41 (26) to alanine mutant derivative of RNF2 showed significantly reduced phosphorylation compared with wild-type, whereas S168A, and S208A mutant derivatives were phosphorylated to the same extent as wild-type RNF2. Further, treatment of RNF2^WT-overexpressing HMEL–BRAF^V600E cells with the MEK inhibitor trametinib led to a significant reduction in CCND2 gene activation by RNF2^WT, whereas LTBP2 expression remained unchanged and overexpression of RNF2^S41A failed to activate the CCND2 promoter (Fig. 7B). Consistently, RNF2 induced H3K27me3 demethylation, H3K27ac accumulation, and UTX/p300 recruitment at the CCND2 promoter, which were abrogated by MEK inhibition in RNF2^WT-overexpressing cells (Fig. 7C). In parallel, the S41A mutant was inefficient in promoting H3K27me3 demethylation and inducing H3K27ac and UTX/p300 recruitment (Fig. 7C). Consistently, MEK inhibition and the S41A mutant drastically reduced the interaction between RNF2 and UTX (Fig. 7D). Finally, we showed that MEK inhibition selectively reduces the increased proliferation conferred by RNF2 overexpression in HMEL–BRAF^V600E, WM115, and 1205Lu cells (Fig. 7E–G), suggesting a therapeutic strategy to suppress RNF2-mediated tumorigenesis.

Together, these data support a model wherein MEK-mediated RNF2 phosphorylation may induce its interaction with histone modifiers, such as UTX and p300, and their recruitment to poised H3K27me3 containing promoters. This recruitment, and subsequent loss of H3K27me3 with gain of activating histone marks, selectively creates an activating environment on gene promoters that exist in a poised state.

In this study, we elucidated distinct molecular mechanisms by which RNF2 regulates proliferation and invasion, highlighting the complex and multifaceted action of epigenetic regulators. Molecularly, the intersection of RNF2 chromatin binding and gene expression analyses identified RNF2-occupied repressed and active promoters. Biologically, a series of reinforcing functional assays utilizing both somatic and genetically engineered germline model systems demonstrated that RNF2′s catalytic activity is dispensable for CCND2 activation, which drives proliferation, but is required for suppression of LTBP2 and activation of TGFβ signaling for invasion and metastasis.

Although it has been suggested that RNF2 may promote gene repression by chromatin compaction independently of its catalytic activity (27), this is the first report of RNF2′s role in gene activation independent of its E3 ubiquitin ligase activity. In this regard, we found that approximately 53% of genes with RNF2-occupied promoters were upregulated in RNF2-overexpressing melanocytes. As with other transcription factors, an intriguing question is how RNF2 might act as both an activator and repressor in the same cell type. A subset of genes activated by RNF2 in this study have poised promoters strongly enriched in the H3K27me3 mark as well as showing weak enrichment of activating histone acetylation and methylation marks. We provide mechanistic insights that MEK1-dependent phosphorylation of RNF2 may promote its binding to activating chromatin modifiers, such as UTX and p300, which in turn remove H3K27me3 and acetylate the promoter, respectively, to open chromatin for gene activation. MEK-dependent phosphorylation of RNF2 provides precedence for a mechanism that signaling proteins may utilize the same molecule to effect gene-specific outcomes in a context-dependent manner. Finally, our data also suggest that MEK inhibitors could be used to block RNF2′s protumorigenic function and therefore could be potentially beneficial in the clinic to suppress growth of RNF2-amplified tumors.

TGFβ signaling has been shown to be critical for induction of proinvasion and migration genes, such as MMPs, N-cadherin, vimentin, and fibronectin. Here, we identified RNF2 as an important epigenetic regulator of TGFβ signaling. Promoter occupancy and expression analyses in this study revealed that RNF2 can directly bind to the LTBP2 promoter to create a repressive environment through H2AK119 ubiquitination and consequent gene silencing. Although LTBP proteins have been reported to both negatively and positively regulate TGFβ signaling (20), our study suggests that, in melanoma, LTBP2 acts as a negative regulator of TGFβ signaling in invasion. We noted that apart from LTBP2, mRNA expression of the EMT transcription factor ZEB2 was also increased in RNF2-expressing cells and may also contribute to the prometastatic phenotype conferred by RNF2. Moreover, we provide strong evidence for requirement of its E3-ubiquitin ligase activity in the promotion of invasive and metastatic properties by RNF2. This, against the backdrop of the well-known opposing effect of TGFβ signaling, raises the possibility that inhibition of RNF2 catalytic activity offers a new therapeutic intervention to target the metastatic activity of TGFβ in metastatic melanoma.

An important question is whether prometastatic and protumorigenic activities of RNF2 are completely independent of each other. Although we provide evidence that the proinvasive/metastatic function is dependent on RNF2′s catalytic activity and the protumorigenic role is independent of it, our data do not completely rule out the possibility that RNF2′s role in proliferation also contributes to its prometastatic phenotype.

Taken together, our findings provide strong evidence that epigenetic regulators, such as RNF2, directly and functionally control powerful gene networks that are vital in multiple cancer processes.

Cells were grown in standard tissue culture conditions (5% CO₂, 37°C). HMEL–BRAF^V600E cells were a kind gift of Dr. David Fisher. 1205Lu, WM115, 501Mel, and WM983B cells were obtained from either the ATCC or Coriell and maintained according to the manufacturer's instructions. Cell lines were authenticated by short tandem repeat profiling and tested every 2 months for Mycoplasma contamination. Cell proliferation assays were performed using an IncuCyte instrument (Essen Bioscience). The instrument captures bright field images every 2 hours and calculates cell density based on the area occupied by cells compared with total area. Soft-agar colony formation assay was performed as described earlier (3). Briefly, two layers of soft agar (bottom layer 0.8% and top layer 0.5%) mixed with DMEM growth medium and FBS were prepared. Two thousand cells were mixed in the top agar layer during plating, and colony formation was monitored. When the colonies reached appropriate size, the colonies were stained with p-iodonitrotetrazolium violet, pictures were taken, and the colonies were counted manually or with ImageJ software. Boyden chamber Matrigel invasion assay was performed as described earlier (3). Briefly, chambers were brought to room temperature and hydrated in serum-free media. One hundred thousand cells were seeded inside the chamber in serum-free media and assayed for the ability to move to the bottom of the chamber in response to 10% serum containing media present in the well after 24 to 48 hours.

Four-to-six-week-old NCR-NUDE mice were purchased from Taconic and injected intradermally with 1 million cells. Tumor volume was measured at designated time points. Mice were euthanized and tumors harvested when tumor size reached 1.5 cm. Mice were maintained in either the animal facility at the Harvard Center for Comparative Medicine or in the animal facility at The MD Anderson Cancer Center. All animal experiments were approved by an Institutional Animal Care and Use Committee review board.

TMA for melanoma progression has been previously described (28). RNF2 immunohistochemistry was performed using Prestige rabbit polyclonal antibody (Sigma). TMA slides were heated at 65°C for 1 hour, deparaffinized in xylene, and rehydrated. Antigen retrieval was performed by boiling at 115°C for 10 minutes and then at 95°C for 30 seconds. After cooling, slides were incubated in 3% H₂O₂ for 20 minutes, washed in PBS, and blocked in goat serum. Following incubation with primary antibody (1:200) overnight, slides were incubated in secondary antibody for 1 hour at 37°C. Slides were then washed and incubated in ABC elite reagent (Vector Labs) and developed using ImmuPACT (Novagen). Manual blinded scoring of the TMA core intensity was performed by two independent pathologists.

Chromatin immunoprecipitation was performed as described earlier (29). Library preparation was done using New England BioLabs reagents as described earlier (29). Sequencing was performed in HiSeq 2000 (Illumina). Data analysis was performed as described in Supplementary Methods.

RNA was isolated using the RNeasy Kit (Qiagen) per the manufacturer's instructions. cDNA was prepared using SuperScript III (Life Technologies) per the manufacturer's instructions. qPCR was performed using SybrGreenER (Invitrogen) and Stratagene instrument. Microarray experiments were performed in the MD Anderson Center for ncRNA Sequencing core facility. Microarray data were analyzed using LIMMA bioconductor package. Details of analysis are provided in Supplementary Methods. All genomic datasets are publicly available at the National Center for Biotechnology Information's Gene Expression Omnibus database (GSE51928, GSE51929, and GSE51930).

TCGA melanoma data (2013_04_06 stddata run) were retrieved from the Genome Data Analysis Center of the TCGA. Survival intervals from date of specimen submission to patients' death/last follow-up were available in 154 cases. Statistical significance of survival differences was estimated by Kaplan–Meier curves and log-rank test in R.

Proteins were made using RIPA buffer (Boston BioProducts) and complete mini protease inhibitor cocktail (Roche). Western blotting was performed by standard procedure using Invitrogen or Bio-Rad precast 4% to 12% gels. Antibodies used were anti-V5 (Invitrogen), anti-vinculin (Sigma), anti-H2AK119ub (Millipore), anti-RNF2 (Sigma), anti-ID1 (SCBT), anti-ID2 (SCBT), anti-ID3 (SCBT). Secondary antibodies used were from LICOR. Blots were developed using LICOR Odyssey imager.

Generation and characterization of iBIP mice and RNF2^L/L mice are described in Supplementary Methods.

We used ChromHMM (23) with default parameters to derive genome-wide chromatin state maps for all cell types, as described in our forthcoming study (Rai and colleagues, unpublished data). We binarized the input data with the ChromHMM's BinarizeBed method using a P value cutoff of 1e⁻⁴. We considered chromatin state models learned jointly on all chromatin marks at every increment of 5 states from 10 to 120 states. We chose a model with 45 states for our main analysis to balance interpretability and capturing informative state distinctions for the analysis here. In particular, the model with 45 states was the model with the minimum number of states that was able to separate bivalent and poised states from active states.

No potential conflicts of interest were disclosed.

Conception and design: K. Rai, J.W. Horner, L. Chin

Development of methodology: K. Rai, J.W. Horner, A.J. Lazar

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K. Rai, L.N. Kwong, E.Z. Keung, S. Sharma, N.S. Samant, J.B. Axelrad, A. Shah, D. Yang, A.J. Lazar

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K. Rai, K.C. Akdemir, P. Fiziev, C.-J. Wu, S. Sharma, D.R. Milton, J.W. Horner, A.J. Lazar, J. Ernst, L. Chin

Writing, review, and/or revision of the manuscript: K. Rai, K.C. Akdemir, L.N. Kwong, P. Fiziev, E.Z. Keung, E.A. Grimm, M.C. Barton, D.R. Milton, T.P. Heffernan, S. Ekmekcioglu, A.J. Lazar, J. Ernst, L. Chin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Sharma, N.S. Samant, M. Williams, J.B. Axelrad, D. Yang, E.A. Grimm, J.W. Horner, S. Ekmekcioglu

Study supervision: J.W. Horner

The work described in this article was supported by grants from the NIH (5U01 CA141508 and U01 CA168394, to L. Chin; R01ES024995 and U01 HG007912, to J. Ernst), NCI (1K99CA160578-01, to K. Rai), Cancer Prevention and Research Institute of Texas (R1204, to L. Chin), and National Science Foundation (1254200, to J. Ernst). The following authors were supported by fellowships: K. Rai (Charles A. King postdoctoral fellowship), J. Ernst (Alfred P. Sloan fellowship), and P. Fiziev (California Institute for Regenerative Medicine Training Grant TG2-01169 and UCLA Training Program).

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