Neddylation Blockade Diminishes Hepatic Metastasis by Dampening Cancer Stem-Like Cells and Angiogenesis in Uveal Melanoma

Metastasis is the major and direct cause of death in patients with cancer at the terminal stage. Unfortunately, there are no effective therapies for metastatic patients. The complicated metastasis process is principally composed of intravasation, circulation, extravasation, and colonization in the target organ (1). Intravasation is referred to as the process that the cancer cells in the primary tumor sites break out of physical barriers in the surrounding tissues, and enter the blood or lymphatic stream to become circulating tumor cells (CTCs; ref. 2). Only a small portion of CTCs that acquire stemness features can successfully escape anoikis, immune response, and shear stress (2). The CTCs undergo extravasation in which they leave from the bloodstream for distant organs and tissues, where they are called metastasis initiating cells (MICs). MICs interact with the microenvironment (niche) to establish their own niches for colonization and eventually form clinically overt metastatic foci. CTCs and MICs share many features (e.g., self-renewal, quiescence, and asymmetric division) with cancer stem-like cells (CSC; ref. 3).

In the clinic, patients with some types of cancer (e.g., breast cancer, lung cancer, and cutaneous melanoma) manifest metastasis at multiple different organ sites (4). Patients with certain other types of cancer [e.g., prostate cancer, pancreatic cancer, and uveal melanoma (UM)] manifest metastasis at a single organ site (e.g., prostate cancer to bone, pancreatic cancer, and UM to liver; ref. 4). The clinical single site-specific organ metastasis pattern in these types of cancer may provide a simplified realization window and study model to circumvent metastasis (4).

UM biologically distinct from cutaneous melanoma may provide a naturally clinical phenotype and typical example of organ-specifically liver metastasis (5). UM, the most common ocular malignancy in adults, usually originates from melanocytes of the choroid, ciliary body, and iris (6). Although successful treatment of the primary tumor with enucleation or radiotherapy can be achieved, half of the UM patients develop metastasis. Eighty-five percent of the patients with metastatic UM exhibit single liver metastasis (7). Little is known about the underlying mechanism of such liver-specific metastasis; there is no effective therapy for patients with metastatic UM, with a median survival of less than 12 months (8). Whole-genome sequencing has demonstrated that mutually exclusive gain-of-function mutations in GNAQ or GNA11 are found in 80% of UM, which may lead to activation of the mitogen-activated protein kinase (MAPK) pathway and transcriptional factor YAP pathway to promote growth and migration of UM cells (9, 10). However, no direct in vivo evidence supports that GNAQ/GNA11-MAPK and GNAQ/GNA11-YAP act as drivers in metastasis of UM.

Most of stemness-related proteins are of short-life with stability sensitive to posttranslational modifications (11). In addition, exocytosis, intracellular trafficking, and secretion of inflammatory cytokines are regulated by protein maturation and posttranslational modifications (12). Neural precursor cell expressed, developmentally downregulated 8 (NEDD8)-conjugation, also called neddylation, is first activated by an E1 enzyme [NEDD8-activating enzyme (NAE); a heterodimer consisting of NAE E1 subunit 1 (NAE1) and ubiquitin-like modifier activating enzyme 3 (UBA3)], transferred to an E2 enzyme (UBC12), and then conjugated to target substrates through cullin-RING ligases (CRL; ref. 13). Neddylation controls protein turnover of a number of CRL targets with essential roles in regulation of oncogenic transformation and pathogenesis. MLN4924 (pevonedistat), a potent and selective first-in-class NAE1 inhibitor (14), is currently in phase I clinical trials in some solid tumors and hematologic malignancies (15, 16).

We hypothesized that blocking the neddylation pathway might disturb the protein homeostasis between synthesis and degradation, which might in turn impact the stemness-related proteins to diminish CSCs, and decrease secretion of the microenvironment cytokine(s) in UM, both of which are fundamental for metastasis (12). The hypothesis was supported by the recent observation in a zebrafish xenograft model that MLN4924 inhibits migration and proliferation of UM cells (17) with the mechanism yet unknown. We tested the hypothesis to find that the neddylation blockade diminished organ-specifically hepatic metastasis by disrupting features of CSCs, niche secreting VEGF-C and its dependent angiogenesis in UM. The findings revealed that MLN4924 may be a promising agent for the treatment of UM patients with hepatic metastasis.

Genetic status of the known altered genes in human primary (92.1, Mel270) and metastatic (Omm1, Omm2.3) UM cells used in this study is summarized in Supplementary Table S1. 92.1 cells were established in Leiden University Medical Center, Leiden, the Netherlands (18). Mel270 and Omm2.3 cells originally derived from patients at the Bascom Palmer Eye Institute, University of Miami School of Medicine, Florida (19), and Omm1 cells were established by Luyten G at Rotterdam University Hospital, the Netherlands (20). These UM cell lines were cultured in RPMI 1640 medium (Invitrogen) supplemented with 10% fetal bovine serum (Hyclone) and 2 mmol/L L-glutamine (7, 21) and authenticated in HAKE Genetics Co., Ltd. by using short tandem repeat matching analysis at August 8, 2017. The cells were kept at 37°C in a humidified incubator with 5% CO₂. No mycoplasma (ThermoFisher Scientific) contamination was detected.

Human tissues from primary tumors were collected from enrolled patients (n = 45) with UM in The Sun Yat-sen University Zhongshan Ophthalmic Center during 2012–2015, after informed consent according to the institutional guidelines and the Declaration of Helsinki principles. The seventh edition of the American Joint Committee on Cancer tumor, node, metastasis (TNM7) classification for eye cancer was used (22, 23). Detailed information of UM patients is shown in Supplementary Table S2. The studies were approved by Institutional Review Board, The Sun Yat-sen University Zhongshan Ophthalmic Center.

IHC staining with anti-NAE1 antibody and quantification of NAE1 expression were performed as described (17, 21, 24), with details provided in the Supplementary information. The association between NAE1 expression and clinicopathologic features in UM patients is summarized in Supplementary Table S3.

All above methods were performed as described previously reported (21, 25), with details provided in the Supplementary information.

Male NOD-SCID mice (4–6-week-old) purchased from Vital River Laboratory Animal Technology Co. and bred at the animal facility of Sun Yat-sen University were used for evaluation the in vivo antineoplastic efficacy of MLN4924 against UM as previously reported (21). After 4 weeks, the mice were randomly separated into two groups (n = 8 per group) and received treatment with vehicle (10% 2-hydroxypropyl-β-cyclodextrin) or MLN4924 (60 mg/kg/day, i.p.) for 14 days. The mice were euthanized, and tumor xenografts were immediately removed, recorded, fixed, and stored at −80°C.

The above methods were examined as previously reported (21, 24, 26), with details provided in the Supplementary information.

Twenty-four hours after Omm2.3-luciferase cells (1 × 10⁶ in 50 μL PBS per mouse) were intrasplenically injected into the NOD-scid-IL2Rg^−/− (NSI) mice (25, 27), the mice were randomly separated into two groups (n = 6 per group) and received treatment with vehicle or MLN4924 (90 mg/kg, bid, i.p.) 5 days on/2 days off for four cycles. Liver metastasis was detected by ex vivo bioluminescence imaging using the IVIS Lumina II (PerkinElmer) or counted the nodules on liver paraffin sections after Hematoxylin and eosin (H&E) staining. All animal studies were conducted with the approval of the Sun Yat-sen University Institutional Animal Care and Use Committee.

All experiments were performed three times, and results are reported as mean ± standard deviation (SD), unless otherwise stated. GraphPad Prism 5.0 was used for statistical analysis. Comparisons between two groups were analyzed by two-tailed Student t test and comparisons of multiple groups by one-way ANOVA with post hoc intergroup comparison with Tukey test. A P < 0.05 was considered statistically significant.

We first examined certain key proteins involved in global NEDD8 conjunction in human UM cells. Western blotting results indicated that the levels of NAE1, UBA3, and UBC12 were higher in the human UM cells than those in ARPE-19 cells (Fig. 1A). Correspondingly, cullin1 was neddylated in human UM cells (Fig. 1A). Further, qRT-PCR analysis revealed that the mRNA levels of NAE1, UBA3, UBC12, and NEDD8 displayed 3.9- to 6.0-, 2.9- to 4.1-, 5.0- to 7.9-, and 4.7- to 7.0-fold increase in the UM cells compared with ARPE-19 cells, respectively (P < 0.01, Fig. 1B–E). These results suggest that the readily detectable status of these key proteins may occur at the transcriptional level.

After verifying the specificity of antibody against NAE1 (Supplementary Fig. S1), we conducted IHC analysis of UM specimens. The results showed that no staining of NAE1 in the adjacent normal tissues was detected. In contrast, readily detectable expression of NAE1 scored from low to high was observed in 77.8% (35/45) patients with UM (Fig. 1F–H). Of importance, NAE1 expression was positively correlated to the largest basal diameter (P = 0.0373) and thickness (P = 0.0461) of the primary tumors we analyzed, two important independent predictors of UM metastatic death (28, 29) and the TNM stage (P = 0.0374; Supplementary Table S3).

To explore whether the neddylation pathway was a therapeutic target for UM, we first examined the specificity of MLN4924 against the neddylation pathway in UM cells. The results indicated that MLN4924, but not proteasome inhibitors MG132 and bortezomib, suppressed the global NEDD8 conjunction or cullin1 neddylation (Fig. 2A). Consequently, the substrates of CRLs (e.g., p21 and p27) were accumulated over all treatments; however, cyclin B1, a non-CRL substrate, was accumulated upon MG132 and bortezomib treatment but not upon MLN4924 treatment (Fig. 2A). Similarly, MLN4924 treatment increased the protein levels of the CRL substrates (e.g., NRF2, phospho-IκBα) in a concentration-dependent manner (Fig. 2B). MLN4924 treatment also decreased cullin1 neddylation in a time-dependent manner (Fig. 2C). Consistent with previous report (30), MLN4924 treatment inhibited 70% to 97% and 90% to 95% (P < 0.001) of the NF-κB-dependent reporter gene activity in Mel270 and Omm2.3 cells, respectively (Fig. 2D). These data suggest that MLN4924 specifically inhibits the neddylation pathway, leading to accumulation of CRL substrates.

We next evaluated the effect of MLN4924 on cellular growth. MTS results showed that MLN4924 concentration-dependently inhibited the growth of UM cells (IC₅₀ values: 433–890 nmol/L) but not of ARPE-19 cells (IC₅₀ value: >10 μmol/L), hinting a therapeutic window in UM cells (Fig. 2E). MLN4924 concentration-dependently inhibited the clonogenicity of UM cells in soft agar, which can better reflect malignant behaviors of tumor cells (IC₅₀ values: 93.8–134.4 nmol/L; Fig. 2F).

We next ascertained the ability of MLN4924 to induce apoptosis in UM cells. MLN4924 increased the dead cell population in a concentration- (Fig. 3A) and time-dependent manner (Supplementary Fig. S2A) in UM cells. Western blotting analysis showed a concentration- (Fig. 3B) and time-dependent (Supplementary Fig. S2B) PARP cleavage and caspase-3 activation. Additionally, treatment with MLN4924 increased cytochrome c release from the mitochondria into the cytosol detected by immunoblotting (Fig. 3C) as well as the cell population with loss of mitochondrial potential (ΔΨm) detected by flow cytometry (Supplementary Fig. S2C). These data together suggest that MLN4924 induces mitochondrial damage and triggers intrinsic apoptosis pathway.

To elucidate the mechanism of MLN4924-induced apoptosis, we evaluated the expression of apoptosis-related proteins. The results showed that the levels of pro-survival proteins survivin and Bcl-X_L were decreased, whereas those of pro-apoptotic protein Bim were slightly increased (Fig. 3D). No change in Bcl-2, Bid, Bax, and Noxa was observed (Fig. 3D).

Because Bcl-X_L (a known NF-κB target gene) was overexpressed in 10% of patients with UM extrapolated from The Cancer Genome Atlas (TCGA) database (http://www.cbioportal.org), which apparently correlated with poor overall survival (P = 0.0174; Fig. 3E). The mRNA levels of BCL2L1 gene were lowered by up to 63.9% and 56.9% in the MLN4924-treated Mel270 and Omm2.3 cells, respectively (Supplementary Fig. S2D). We next examined the role of Bcl-X_L in MLN4924-induced apoptotic cell death in UM cells. Forced overexpression of Bcl-X_L attenuated MLN4924-induced apoptosis in UM cells as reflected by caspase-3 activation (Fig. 3F) and trypan blue staining cells (19.3% decrease in Mel270, P < 0.01; 10.2% decrease in Omm2.3, P < 0.05; at 1.0 μmol/L treatment; Fig. 3G). Conversely, silencing Bcl-X_L by siRNA duplexes potentiated the lethal effect of MLN4924 in UM cells (Supplementary Fig. S2E and F; increased dead cells by 25.6% and 27.9% in Mel270; 23.8% and 24.2% in Omm2.3, P < 0.01; at 1.0 μmol/L treatment).

Next, we evaluated the synergistic effect between MLN4924 and the conventional chemotherapeutic agent vinblastine in UM cells (17). MTS results showed that combinational treatment with MLN4924 and vinblastine synergistically (CI values <1) inhibited cell growth in Mel270 and Omm1 cells (Supplementary Fig. S3A), as evaluated by using the median-effect method of Chou and Talalay (31). Flow cytometer and Western blotting results suggest that combinational treatment with MLN4924 and vinblastine induced enhanced cell apoptosis as reflected by the increase of Annexin V⁺ cells, PARP cleavage, and caspase-3 activation, respectively (Supplementary Fig. S3B and C).

Previous studies have shown that MLN4924 can induce DDR and cell cycle perturbation to inhibit cell growth in several types of cancer (32, 33). We first assessed the key components involved in DDR. Western blotting analysis showed that MLN4924-treated UM cells displayed an appreciably increased γ-H2AX, an indicative hallmark of DNA double strand breaks (DSB; Supplementary Fig. S4A). γ-H2AX not only indicated the existence of MLN4924-mediated DSBs but also reflected the activation of DDR. Indeed, the levels of p53 and phospho-p53 (S15) were increased by MLN4924 (Supplementary Fig. S4A).

Ataxia telangiectasia-mutated (ATM) kinase and ATM-Rad3-related (ATR) kinase function at the level of sensors and transducers of DNA damage signaling pathway in response to DSBs and single strand breaks (SSB), respectively (34). Our results showed that the phosphorylation of ATM was concentration-dependently increased in the MLN4924-treated UM cells (Supplementary Fig. S4A), suggesting a steady activation of ATM after MLN4924 treatment. A slight and nonconcentration-dependent increase in ATR phosphorylation was observed in the cells treated with certain lower concentrations of MLN4924 (Supplementary Fig. S4A). These findings together with the γ-H2AX change further hint that DSBs may be the major DNA damage form in response to neddylation inhibition.

We also detected DDR checkpoint kinase 1 (Chk1) and Chk2. The results revealed that the phosphorylation of both Chk1 and Chk2 was steadily increased by MLN4924 (Supplementary Fig. S4A). These data collectively indicate that neddylation inhibition induces DSBs and elicits DDR predominantly by ATM, Chk1, and Chk2.

Increased reactive oxygen species (ROS) generation is involved in MLN4924 action mechanism and may lead to DNA damage (35). To explore the reason that MLN4924 induced DDR in UM, we detected the intracellular ROS levels. ROS generation was found to be increased by 1.4- to 2.7-fold in MLN4924-treated Mel270 cells (Supplementary Fig. S4B). Pretreated with N-acetylcysteine, the ROS scavenger, significantly prevented MLN4924-mediated DNA damage, as reflected by γ-H2AX in UM cells (Supplementary Fig. S4C). These data suggest that ROS generation may mediate DNA damage in UM cells upon MLN4924 exposure.

Flow cytometry analysis showed that MLN4924 treatment led to increase in G₂–M phase population in a concentration- and time-dependent manner in UM cells, respectively (Supplementary Fig. S4D and E). Correspondingly, overt decrease of phospho-histone H3 (S10), a hallmark of M phase, was noted in the MLN4924-treated UM cells, suggesting a stalled G₂ phase (Supplementary Fig. S4F). Further analysis of G₂–M checkpoint mechanism revealed that a remarkable increase of Wee1 (a well-defined CRL substrate and an inhibitor of G₂–M phase transition), phospho-Cdc25C (S216), and phospho-Cdc2 (Y15) was observed in the MLN4924-treated UM cells (Supplementary Fig. S4F).

We next examined whether MLN4924 was synergistic with MK-8776 (Chk1 inhibitor) and MK-1775 (Wee1 inhibitor). The results indicated that the combination between MLN4924 and MK-8776 or MLN4924 and MK-1775 synergistically retarded the cell growth as measured by MTS assay (Supplementary Fig. S5A) and induced enhanced apoptosis, reflected by enhanced specific cleavage of PARP as detected by Western blotting analysis (Supplementary Fig. S5B) in UM cells.

We evaluated the in vivo antineoplastic efficacy of MLN4924 against UM using NOD-SCID mouse xenograft model. The mice (4–6-week-old) were subcutaneously injected with Omm1 cells. When the tumor reached ∼50 mm³, the mice were randomly divided into two groups (n = 8) and received treatment with vehicle or MLN4924 for 14 days. The tumor size was much smaller in MLN4924-treated group than vehicle-treated mice (Fig. 4A). MLN4924 administration abrogated the tumor outgrowth (Fig. 4B). Consistently, tumor weight was decreased by 58% (P < 0.01) in the MLN4924-treated mice (Fig. 4C). MLN4924 administration inhibited tumor proliferation as reflected by the expression of Ki67. Moreover, the expression of p27, a well-known CRL substrate, was increased in tumor tissues from the MLN4924-treated mice (Fig. 4D). Western blotting results of cell lysates showed inhibited neddylation pathway with the protein levels of CRL substrates p21 and p27 notably increased. The expression of Slug, Bcl-X_L, and survivin was, however, decreased in MLN4924-treated group (Fig. 4E), which were consistent with the in vitro findings. These data demonstrate the in vivo antineoplastic efficacy of MLN4924.

Because proteasome-ubiquitin cascade is involved in self-renewal of CSCs, and CSCs were reported to exist in UM cells (36), we determined whether MLN4924 treatment conferred elimination of CSCs in UM. We first employed melanosphere formation and serially replating assay. As expected, MLN4924 treatment inhibited the self-renewal of spherogenic UM cells (P < 0.05 for all comparisons) in UM cells (Fig. 5A). Aldehyde dehydrogenase activity (ALDH) was a widely accepted biomarker for CSCs in solid tumors (37). We observed that MLN4924 treatment reduced ALDH⁺ cells by 28.6%, 43%, 2.1%, and 8% in Mel270, 92.1, Omm1, and Omm2.3 cells, respectively (Fig. 5B; Supplementary Fig. S6). Of importance, in vivo limiting dilution assay showed that neddylation inhibition reduced UM CSCs frequency by 98% (Control: 2.31 × 10⁻⁶; MLN4924: 3.84 × 10⁻⁸; Fig. 5C; Supplementary Table S4). To explore the reason that MLN4924 inhibited CSC properties in UM cells, we determined the levels of epithelial-to-mesenchymal transition (EMT)- and stemness-related transcriptional factors (38). The results demonstrated that the levels of Slug were decreased in the MLN4924-treated UM cells (Fig. 5D).

In order to elucidate the mechanism by which MLN4924 regulates Slug, we examined whether neddylation inhibition affected protein stability of Slug. Time-chase experimental results showed that MLN4924 treatment led to 1.9- and 1.6-fold increase in turnover rates of Slug protein in 92.1 and Omm2.3 cells, respectively (Fig. 5E). The MLN4924-mediated decrease in Slug protein was effectively rescued by the proteasome inhibitor MG132 treatment (Fig. 5F). Because FBXO11 was the bona fide E3 ligase of Slug (11), we examined FBXO11 expression. Consistent with the previous study (35), MLN4924 treatment increased the levels of FBXO11 (Fig. 5G). Overexpression of FBXO11 but not the F-box deletion mutant of FBXO11 (FBXO11-ΔF) decreased Slug protein level (Fig. 5H), suggesting that the amino-terminal F-box domain of FBXO11 may be essential for Slug protein degradation. Taken together, the data suggest that MLN4924 decreases Slug through protein degradation in UM cells.

To functionally characterize the role of Slug in MLN4924-mediated elimination of CSCs in UM cells, we first analyzed the impact of ectopic expression of Slug on stemness traits. The results showed that Slug overexpression increased the ability of melanosphere formation and serially replating in CSC culture medium compared with the cells transfected with empty vector (P < 0.001; Supplementary Fig. S7A-C). Similarly, ectopic expression of Slug in UM cells increased 1.6- and 4.5-fold the subpopulation of ALDH⁺ cells (Supplementary Fig. S7D and E). Conversely, Slug knockdown decreased the ability of melanosphere formation and serially replating (P < 0.001) as well as percentage of ALDH⁺ cells by 87% and 70% in 92.1 and Omm2.3 cells, respectively (Supplementary Fig. S7F-J). The MLN4924-mediated decrease in melanosphere formation was rescued by forced expression of Slug in UM cells (Supplementary Fig. S7K). Collectively, these results indicate that Slug positively regulates CSC properties, and that decreasing Slug by MLN4924 confers eradication of CSCs in UM cells.

It was previously reported that MLN4924 blocked proliferation and migration of UM cells in a zebrafish xenograft model (17). By using wound-healing scratch assay, we similarly found the inhibitory effect of MLN4924 on the scratch healing ability of 92.1 and Omm2.3 cells (Supplementary Fig. S8A). Transwell Boyden chamber assay further revealed that the migration and invasion capacities were suppressed in the MLN4924-treated UM cells, respectively (P < 0.001; Supplementary Fig. S8B and C). Western blotting analysis showed that the levels of metastasis-associated MMP-9 and MMP-2 were decreased after MLN4924 treatment (Supplementary Fig. S8D). Taken together, our data suggest that MLN4924 diminishes migration and invasion of human UM cells.

Using a liver metastasis mouse model by intrasplenic injection of Omm2.3-Luc cells in NSI mice, we found that MLN4924 treatment reduced bioluminescence signal by 76% in liver (P < 0.01; Fig. 6A). H&E staining of liver sections indicated 55% decrease in the number of metastatic nodules in livers of the MLN4924-treated mice (P < 0.01; Fig. 6B). These results suggest that MLN4924 treatment reduces hepatic metastasis in UM. Of note, we observed that the microvascular density was decreased by 74% (P < 0.05) upon MLN4924 treatment (Fig. 6C), indicating impaired angiogenesis. Because the conditioned medium (CM) derived from cancer cells can promote the migration and tube formation of HUVECs (24), we examined the effect of the CM derived from MLN4924-treated UM cells. The results showed that the ability of migration (decreased by 44%–74% and 51%–69%) and tube formation (decreased by 73%–86% and 74%–88%) of HUVECs when exposed to the CM derived from MLN4924-treated Omm1 and Omm2.3 cells, respectively (Fig. 6D and E). With an independent approach of CAM assay, we noted that the CM derived from MLN4924-treated UM cells caused 55% to 70% reduction of angiogenesis when compared with the CM derived from DMSO-treated UM cells (Fig. 6F). Taken together, these results suggest that the neddylation pathway blockade impairs angiogenesis.

Given the critical role of VEGF in both angiogenesis and lymphangiogenesis (39) and the existence of VEGF in the CM of UM cells (40), we examined the impact of MLN4924 on the secreting ability of VEGF in UM cells. Among the all tested five angiogenic factors (VEGF-A, VEGF-B, VEGF-C, VEGF-D, and b-FGF) by qRT-PCR analysis, only the mRNA level of VEGF-C was inhibited in MLN4924-treated Omm1 and Omm2.3 cells (Fig. 6G; Supplementary Fig. S9A). The mRNA level of VEGF-A was undetectable (data not shown). We therefore chose VEGF-C in the subsequent experiments. ELISA assay showed that the protein levels of VEGF-C in the CM derived from MLN4924-treated UM cells decreased by up to 70% (Fig. 6H). Ectopic overexpression of VEGF-C cDNA in UM cells rescued, while treatment with neutralizing anti-VEGF-C antibody or silencing VEGF-C augmented, the inhibitory effect of CM derived from the MLN4924-treated UM cells on migration and matrigel tube formation of HUVECs (Fig. 6I–L; Supplementary Fig. S9B–G). These results indicated that MLN4924 suppresses angiogenesis by impairing the VEGF-C secretion in UM.

Given that NF-κB can bind the promoter of VEGF-C gene to initiate its transcription (39), and that treatment with MLN4924 decreases NF-κB transcriptional activity (Fig. 2E; ref. 30), we examined whether NF-κB was a mediator of MLN4924-mediated suppression in VEGF-C. Luciferase reporter assay showed that MLN4924 reduced VEGF-C gene promoter activity by 49% to 72% and 39% to 52% in Omm1 and Omm2.3 cells, respectively (Supplementary Fig. S9H). Overexpression of p65 increased 2.8- and 3.3-fold VEGF-C mRNA levels, and rescued VEGF-C decrease in MLN4924-treated UM cells (Supplementary Fig. S9I). The results suggest that NF-κB is responsible for MLN4924-mediated suppression in VEGF-C.

In the present study, we found that NAE1 expression was readily detectable in UM cells, and that inhibition of the neddylation pathway by MLN4924 suppressed the proliferation, survival, migration and invasion, and hepatic metastasis. The inhibitory effect of neddylation blockade on proliferation, which was confirmed by xenografted UM tumor in NOD-SCID mice, was involved in activation of ATM-Chk1-Cdc25C DDR, and G₂–M phase arrest. Moreover, MLN4924 treatment eliminated CSCs with stemness-related transcriptional factor Slug suppressed, and disturbed the paracrine secretion of VEGF-C and its dependent angiogenesis. With an NSI mouse model, we identified that MLN4924 inhibited hepatic metastasis in UM.

We demonstrated that MLN4924 suppressed CSC properties in UM. To our knowledge, this is the first study to provide in vitro and in vivo evidence to demonstrate that inhibition of the neddylation pathway eliminates CSCs in solid tumors. Consistent with our findings, Knorr and colleagues showed that in vitro treatment of MLN4924 exerts a cytotoxic effect on myelodysplastic syndrome and acute myelogenous leukemia stem and progenitor cell populations (41). In contrast, Zhou and colleagues reported that MLN4924 at low nanomolar concentrations (0–100 nmol/L) exerted a stimulating effect on tumorsphere formation of non-small cell lung cancer cells when cultured in serum-free medium (42). The discrepancy may be attributed to their serum-starved culture condition, suggesting that blockage of neddylation modification to regulate CSC self-renewal may be serum-, drug concentration-, and cancer type-dependent. At this view point, we carefully and systemically evaluated the functional parameters of CSCs (e.g., ALDH⁺ cells, melanosphere and serially replating capacity, and the frequency of CSCs) in UM. The results unequivocally revealed that the CSC features were diminished after MLN4924 treatment.

Previous studies have demonstrated that Slug alone or in collaboration with transcriptional factor SOX9 is required to determine the characteristics of CSCs in breast and ovarian cancer (43). However, the role of Slug in maintaining the self-renewal of UM CSCs has not been reported. By screening the EMT- and stemness-related proteins in UM cells, we found that the change of Slug is most impressive in the MLN4924-treated UM cells. We therefore determined the role of Slug in the MLN4924-mediated elimination of CSCs in UM. Our results including Slug cDNA rescue experiments indicated that Slug is essential for sustaining the CSC properties in UM. Mechanistically, MLN4924 treatment leads to the degradation of Slug through the E3 ubiquitin ligase FBXO11, which was consistent with the previous report (35).

In the complex colonization process, establishment of tumors' own angiogenesis may be a turning point. Regarding UM, we found that VEGF-C is a mediator of the angiogenesis in liver. This is supported by several lines of evidence. First, the CM derived from UM cells can promote angiogenesis, indicating that UM cells may secrete certain cytokine(s). Second, MLN4924 treatment significantly attenuated the UM cells secreting such cytokine(s). Third, ectopic overexpression of VEGF-C decreased the suppression effect of MLN4924. Fourth, treatment with neutralizing antibody against VEGF-C suppressed the angiogenesis-promoting effect of UM CM and potentiated the effect MLN4924. Finally, MLN4924 significantly decreased the hepatic metastasis and angiogenesis in mice.

Mechanistically, we found that NF-κB p65 was critical in the MLN4924-mediated inhibition of VEGF-C gene transcription in UM cells. MLN4924 treatment led to accumulation of IκBα, which elicited p65 degradation in UM cells, which was consistent with the precious report in lymphoma cells (39). In the canonical NF-κB pathway, IκBα phosphorylation at S32 and S36 and the resultant polyubiquitination and degradation of IκBα protein are critical steps for nuclear localization of p65. Polyubiquitination of IκBα is regulated by SCF^βTRCP, which is controlled by NAE (44). Consistent with this notion, a dose-dependent IκBα phosphorylation at S32 and S36 was noted in the MLN4924-treated UM cells.

Proliferation and evading apoptosis are also critical for forming overt metastases as well as growth of primary tumor. We found that MLN4924 at nanomolar concentrations exert broad tumoricidal activity across a panel of UM cells. Phase I studies of pevonedistat (MLN4924) showed that maximal plasma concentration (C_max) reached to 2.7 to 8.1 μmol/L at maximum tolerated dose (45), which is sufficient to kill UM cells, as extrapolated from our data.

MLN4924 treatment induces apoptosis with mitochondrial damage in UM cells. Bcl-X_L exhibited a marked decrease in the MLN4924-treated UM cells. In line with our findings, Nimati and colleagues reported that Bcl-X_L is an attractive therapeutic target in UM patient-derived xenografts (46). Targeting Bcl-X_L may have a particular implication because BCL2L1 overexpression is a poor prognostic factor based on the analysis of UM patients obtained from the TCGA database.

Our results highlighted that prolonged (up to 48 hours) exposure of UM cells to MLN4924 changed cell cycle distribution featured G₂–M phase also observed in pancreatic cancer (47), and gastric cancer cells (48), but not S phase accumulation. The increased γ-H2AX together with increased phospho-ATM and phospho-Chk1 unequivocally indicated that MLN4924 induced DSBs, because ATM-Chk1-Cdc25C axis is predominantly responsible for DSBs. However, MLN4924 treatment also caused phosphorylation in ATR and Chk2, suggesting occurrence of SSBs. Consistent with previous report (35), MLN4924-induced DNA damage due to ROS generation in UM cells. Taken together, MLN4924 causes accumulation of CRL substrates, which are responsible for G₂–M phase arrest and DDR.

BRCA1 associateprotein-1 (BAP1) mutations are reportedly present in the majority of metastatic UM (49). BAP1, a de-ubiquitinating enzyme, is a member of the polycomb-group proteins (PcG) of highly conserved transcriptional repressors required for long-term silencing of genes that regulate cell fate determination and stemness features. It is possible that mutation of BAP1 is involved in the efficacy of neddylation inhibition in metastatic UM. However, silencing BAP1 or forced overexpressing BAP1 in Mel270 and Omm2.3 cells did not alter the sensitivity to MLN4924 (data not shown).

In conclusion, neddylation inhibition dampens Slug to eliminate CSCs, disturbs paracrine secretion of VEGF-C and its dependent angiogenesis. Blockade of neddylation lowers Bcl-X_L to induce intrinsic apoptosis, and elicits DNA damage to induce cell cycle arrest in UM cells (proposed model, Fig. 6M). From a therapeutic standpoint, our work presents novel insights into molecular strategies to effectively treat organ-specifically hepatic metastasis in UM. These findings warrant a clinical trial of pevonedistat in hepatic metastatic patients with UM.

No potential conflicts of interest were disclosed.

Conception and design: Y. Jin, J. Pan

Development of methodology: Y. Jin, Y. Wang, B. Jin, J. Zhou, J. Zhang, J. Pan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Jin, P. Zhang, Y. Wang, B. Jin, J. Zhou, J. Zhang, J. Pan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Jin, Y. Wang, B. Jin, J. Pan

Writing, review, and/or revision of the manuscript: Y. Jin, J. Pan

Study supervision: J. Pan

The authors thank Jager MJ (Leiden University Medical Center, Leiden, the Netherlands) for generously providing 92.1, Mel270, Omm1, and Omm2.3 cells. This study was supported by grants from National Natural Science Funds (No. U1301226 and No. 81025021 to J. Pan); the Research Foundation of Education Bureau of Guangdong Province, China (Grant cxzd1103 to J. Pan). Present address of Yanli Jin: Institute of Tumor Pharmacology, College of Pharmacy, Jinan University, Guangzhou, China.

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.