Introduction

Uveal melanoma (UM) is the most common intraocular malignancy in adults. The incidence of UM ranges from 1-9 per million people annually worldwide, with approximately 6 per million people in Europe and the USA, and 0.5 per million people in Asia and Africa1. UM exhibits a considerable propensity for metastasis, and once metastasis spread is detected, the median survival time is less than one year2. Unlike cutaneous melanoma, which harbours mutations in proto-oncogenes such as BRAF, NRAS and KIT, UM harbours exclusive oncogenic mutations in G protein subunit alpha q (GNAQ), G protein subunit alpha 11 (GNA11)3, phospholipase C beta 4 (PLCB4) or cysteinyl leukotriene receptor 2 (CYSLTR2)4,5. Furthermore, genetic variants such as monosomy 3 and BRCA1-associated protein 1 (BAP1) mutations and gene expression profiles (GEPs) are well recognized for their role in UM progression6,7,8.

Recently, some progress has been made in the study of epigenetic mechanisms in UM9,10. A unique DNA methylation profile in UM has been demonstrated11. Furthermore, a comprehensive epigenetic study revealed that UM progression is largely driven by the loss of activity of Polycomb Repressive Complex 1 (PRC1), a key transcriptional regulator in development that modulates chromatin structure and catalyses histone H2A lysine 119 ubiquitination (H2AK119ub1)12. Preclinical experiments have revealed that small-molecule epidrugs targeting epigenetic factors involved in histone modification, such as histone deacetylases (HDACs) and the bromodomain and extraterminal domain (BET) family, display robust efficiency in the therapeutic intervention of UM13,14. Epigenetic dysregulation is involved in the development of UM, which affects chromatin structure and gene transcriptional activity, thereby activating oncogenes or repressing tumour suppressor genes, thus promoting UM cell proliferation, migration and invasion by activating downstream signalling pathways10. Despite considerable advances in elucidating the carcinogenesis of UM, the comprehensive epigenetic landscape has not yet been fully characterized. Emerging evidence suggests that epigenetic dysregulation not only contributes to tumorigenesis but may also be implicated in the molecular mechanisms underlying metastatic progression15,16.

Super-enhancers (SEs) are key epigenetic regulators that drive oncogene expression during the initiation and progression of malignancies16,17,18,19. Aberrant SEs are markedly enriched with histone modifications to drive the expression of oncogenes20. The transcription of SE-related oncogenes depends on the recruitment of master transcription factors (MTFs) and cofactors. Notably, MTFs regulate core oncogenes, for example, the MTF Kruppel-Like Factor 3 (KLF3) has been implicated in colorectal cancer21. In particular, oncogenes driven by SEs are vulnerable to MTF suppression19,22. Such epigenetic vulnerabilities have been demonstrated in several malignancies18. For instance, the fusion protein PAX3-FOXO1 establishes a myoblastic SE landscape and shows a profound dependence on BET family inhibitors in specific rhabdomyosarcoma patients23.

Here, we applied an integrative analytic pipeline of epigenetic and transcriptomic profiles to define UM-specific SEs and potential therapeutic targets. MTFs that are differentially expressed in UM were found to preferentially enrich oncogene-associated SEs. Subsequent bioinformatic analysis revealed the critical node-transcription factor AP-2 alpha (TFAP2A) as the main lineage-specific factor in UM, and its high-confidence targets were identified. TFAP2A is a key MTF that is highly expressed in UM and is closely associated with UM oncogenesis. Using CRISPR/Cas9 genomic editing, we eliminated TFAP2A expression and observed a significant reduction in tumour formation and cellular nutrient metabolism, suggesting its oncogenic properties. Pathway analysis and functional studies also revealed TFAP2A as the crucial node driving metabolic reprogramming. Moreover, we revealed Solute Carrier Family 7 member 8 (SLC7A8) as a key TFAP2A-occupied SE that contributes to UM tumorigenesis. Taken together, the results of our study revealed a UM-specific SE regulatory circuit that regulates UM oncogenesis and metabolic reprogramming.

Methods

Cell culture

The normal human melanocyte cell line PIG1 was provided by the Department of Ophthalmology, Peking University Third Hospital (Beijing, China). The UM cell lines 92.1 and MP46 and the human retinal pigment epithelial cell lines ARPE19 and HEK293T were purchased from American Type Culture Collection (Manassas, VA, USA). The human UM cell lines OMM2.3, OMM1 and Mel290 were provided by Professor Martine J Jager (Leiden University Medical Center, Leiden, The Netherlands).

PIG1 cells were cultured in DMEM/Ham’s F12K medium (GIBCO, Thermo Fisher Scientific, Waltham, MA, USA). HEK293T and Mel290 cells were cultured in DMEM (GIBCO), and the human UM cell lines 92.1, MP46, OMM1 and OMM2.3 were cultured in RPMI-1640 medium (GIBCO). All media were supplemented with 10% foetal bovine serum (GIBCO), penicillin (100 U/mL) and streptomycin (100 mg/mL). The cells were cultured at 37 °C in a humidified 5% CO2 atmosphere. All the in vitro cell lines were validated by short tandem repeat profiling and confirmed to be mycoplasma free.

Tissue specimens

Tumour samples were collected from surgically resected tumours by enucleation or endoresection at the Department of Ophthalmology, Ninth People’s Hospital, Shanghai Jiao Tong University School of Medicine (Shanghai, China). Fresh samples were snap-frozen and stored in liquid nitrogen. Written informed consent for this study was obtained from all patients in Shanghai Ninth People’s Hospital, Shanghai Jiao Tong University School of Medicine. All ethical regulations relevant to human research participants were followed. The study was in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Shanghai Ninth People’s Hospital, Shanghai Jiao Tong University School of Medicine (approval No. SH9H-2019-T185-2).

Lentivirus production and CRISPR-Cas9 transfection

HEK293T cells were cotransfected with the lentiviral envelope and packaging plasmid pCMV-VSVG (Addgene plasmid #8454), psPAX2 (Addgene plasmid #12260) or the lenti-CRISPR plasmid of sgRNA (Supplementary Table S1) and Lipofectamine 3000 (Invitrogen, Carlsbad, CA, USA). Twenty-four hours after transfection, the medium was replaced, and three days after transfection, the medium was centrifuged. The supernatant was collected, filtered and concentrated using Amicon Ultra tubes (100 kDa, Merck-Millipore, Darmstadt, Germany). The cells were seeded in 24-well plates and transduced with lentivirus containing 8 g/ml polybrene. Stable clones were selected using puromycin.

Western blotting

For western blotting, UM cells were harvested and washed twice with ice-cold PBS. The cells were lysed in RIPA lysis buffer supplemented with proteinase inhibitor cocktail and phosphatase inhibitor cocktail (Roche, Basel, Kanton Basel, Switzerland). Protein concentrations were determined with a Bio-Rad protein assay kit (Bio-Rad Laboratories, Hercules, CA) according to the following protocol. The samples were denatured at 95 °C for 10 min and separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis, followed by transfer to a nitrocellulose membrane (Bio-Rad). The membrane was incubated with primary antibody overnight at 4 °C. The membrane was then incubated with secondary antibody for 1 h at room temperature. Antibodies against the following antigens were used: TFAP2A (1:1000, 10319-3-ap, Proteintech), Actin (1:1000, ab8226, Abcam), Anti-rabbit IgG (H + L) (Alexa Fluor® 488 Conjugate) (1:5000, 4412S, Cell Signaling Technology).

RNA-seq

RNA was extracted using the RNeasy Mini Kit (Qiagen, Hilden, Germany), and 1 ng of RNA was used to construct the sequencing library using the KAPA Stranded RNA-seq Library Prep Kit (Illumina, San Diego, CA, USA). The generated library was evaluated using an Agilent 2100 Bioanalyzer and quantified via a qPCR technique. The barcoded libraries were combined, captured on an Illumina flow cell and then sequenced for 150 cycles from both ends on an Illumina Novaseq 6000 instrument. FastQC was used to perform quality control and cutadapt was used to remove the adaptors24. The trimmed paired-end reads were then aligned to the GRCh37 reference genome using HISAT2 v2.1.0. Each transcript abundance was assessed using StringTie, and the fragments per kilobase of exon model per million mapped fragments (FPKM) value for each gene and transcript was determined using the Ballgown package in R. Gene Ontology and pathway analyses were conducted, and volcano plots were generated using R. Differential gene expression analysis was performed using DESeq2 v3.7. Heatmaps were generated using the R package pheatmap. Curated gene sets (Molecular Signatures Database v6.2) were analysed using GSEA software version 3.0. The GSEA preranked instrument was applied with 1000 permutations, and only gene sets with a maximum list size of 500 and a minimum list size of 15 were taken into account. A pan-cancer analysis was conducted using the UCSCXenaShiny to explore gene expression patterns across multiple cancer types25. Additionally, TFAP2A genomic alterations, including copy number variations (CNVs), were analyzed using the cBioPortal platform25.

RNA extraction and real-time PCR(RT‒PCR)

RNA was extracted using TRIzol and measured using a NanoDrop ND-2000 (NanoDrop, Waltham, MA, USA). One microgram of RNA was treated with DNase (Invitrogen, Carlsbad, CA, USA). For reverse transcription, cDNA generation was performed using PrimeScript RT Master Mix (Takara, Kyoto, Japan). SYBR Green PCR Master Mix (Takara) was used for RT‒PCR analysis. The target genes were amplified on a 7500 Fast Real-Time PCR machine (Applied Biosystems, Foster City, CA, USA). The experiments were repeated in triplicate. The primers used are listed in Supplementary Table 3.

Chromatin immunoprecipitation sequencing (ChIP-seq)

ChIP-seq experiments were carried out with the ChIP-IT High Sensitivity Kit (Active Motif #53040) following the manufacturer’s protocol. Briefly, UM cells were cultured to 90% confluence, treated with 1% formaldehyde for 10 min and quenched with 125 mM glycine. Cells were washed twice with cold PBS, harvested, and lysed in Lysis Buffer supplemented with protease inhibitors. Chromatin was sheared using an EpiShear Probe Sonicator with 4 cycles of 30 s ON/30 s OFF pulses at 30% amplitude until DNA fragments reached 200–500 bp. After centrifugation, 50 μg of chromatin was immunoprecipitated overnight at 4 °C with 5 μg of target-specific antibody in ChIP Buffer. Immune complexes were captured with Protein G Magnetic Beads for 1 h at 4 °C, followed by sequential washes with Low Salt Wash Buffer and High Salt Wash Buffer. DNA was eluted in Elution Buffer and reverse crosslinked at 65 °C overnight. For library preparation, purified ChIP DNA fragments were size-selected (250–300 bp) using SPRIselect beads. Libraries were quantified by Qubit fluorometry (Thermo Fisher) and quality-checked by Agilent Bioanalyzer. Multiplexed libraries were sequenced on an Illumina NovaSeq 6000 system using the S4 flow cell, generating paired-end reads (2 × 150 bp) per sample. Antibodies against the following antigens were used: Histone H3 (tri methyl K9) (5 ug, ab176916, Abcam), Histone H3 (mono methyl K4) (5 ug, ab176877, Abcam), Histone H3 (tri methyl K27) (5 ug, ab192985, Abcam), Histone H3K27ac (5 ug, 39034, Active Motif), p300 (5 ug, 61401, Active Motif).

ChIP-seq analysis

The reads were mapped to the reference genome using BWA (v 0.7.17). MACS2 (https://github.com/taoliu/MACS/) was used to call peaks, with input used as a control. Gene ontology was analysed using the online tool GREAT, with the whole genome used as the background. Chromatin states were investigated using ChromHMM (http://compbio.mit.edu/ChromHMM/). Bedtools (http://bedtools.readthedocs.io/en/latest/) was used to find differential peaks and to determine normalized per million mapped reads (RPM), and enhancers were mapped to their nearest gene using the EDEN protocol according to previous methods23. Known and de novo motif enrichments were identified using HOMER and MEME. Enrichment peaks were visualized with IGV tools. For enhancer calling and enhancer‒gene associations, the Rank Ordering of Super-Enhancers (ROSE) algorithm was used to call enhancers and SEs according to H3K27ac enrichment in UM cells. ChIP-seq occupancy heatmaps and meta-profile plots were prepared using deepTools. Consensus TF DNA binding motifs (De Novo) were investigated using a 1 kb window centred on significant peaks using the HOMER suite. Gene ontology classification was performed using the Gene Ontology (GO) molecular function ontologies. P values were adjusted using the Benjamini–Hochberg multiple testing correction.

Cleavage under targets and tagmentation (CUT&Tag)

The CUT&Tag assay was completed as previously described26. Generally, 50 mg of tissue was collected, washed and mixed with activated concanavalin A-coated magnetic beads (Bangs Laboratories, Inc., Fishers, IN, USA.) at room temperature for 15 min. The mixture was resuspended in 100 μl DIG-wash buffer containing 2 mM EDTA and primary antibody (1:50 dilution) and incubated overnight at 4 °C. The secondary antibody was added at 1:50 dilution and incubated at room temperature for 1 h. After washing, 100 μl pA-Tn5 adapter complex (∼40 nM) was added and incubated at room temperature for about 1 h. After washing, the cells were resuspended in Tagmentation buffer (50 μl) and incubated at 37 °C for 1 h. Then, Proteinase K treatment was used at 55 °C for 30 min to terminate the tagmentation, followed by 70 °C incubation for 20 min to inactivate Proteinase K. DNA was then extracted with AMPure XP beads (Beckman Coulter) and eluted. Sequencing was conducted on the Illumina NovaSeq 6000 system with 150-bp paired-end reads. Differentially accessible peaks were identified using thresholds of fold change >2.0 (log2 ratio > 1) and p-value < 0.05. Antibodies against the following antigens were used: Histone H3K27ac (1:50, 39034, Active Motif), Anti-rabbit IgG (1:50, 4412S, Cell Signaling Technology).

Assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq)

ATAC-seq was conducted according to the manufacturer’s instructions. Briefly, UM cells were trypsinized and collected. The Tn5 transposition reaction was performed using the Nextera kit (Illumina). DNA was purified using the MinElute Kit (Qiagen). Library amplification was performed using NEBNext High-Fidelity 2X PCR Master Mix (NEB) with SYBR Green I dye (Invitrogen) and Nextera primers at 25 µM. Amplified libraries were purified with a PCR purification kit (Qiagen), and quality control was performed using the 2200 TapeStation (Agilent). Libraries were sequenced using Illumina NextSeq 500 paired-end 150-bp reads.

In vivo studies

Nude mouse models (6–8 weeks, female) were established by injecting 1 × 106 sgNC or sgRNA cell lines (OMM2.3-sgTFAP2A-1/2) into the tail vein. The fluorescence intensity of the tumours was measured every 7 days for a total of 30 days. At the end of the experiments, the mice were sacrificed, and the tumours were collected for further analysis. We have complied with all relevant ethical regulations for animal use. All experimental procedures were approved by the Institutional Animal Care and Use Committee of the Ninth People’s Hospital, Shanghai JiaoTong University School of Medicine. According to the institutional guidelines, the maximal permitted tumor diameter for subcutaneous xenografts was 20 mm (or approximately 4000 mg in mass). In all experiments, tumor growth remained within these approved limits.

Gas chromatography‒mass spectrometry (GC‒MS)

A total of 1 × 106 cells were harvested and washed three times with PBS, and metabolites were collected by adding 1 ml of ice-cold extraction buffer (50% methanol, 30% acetonitrile, and 20% water). The supernatant was obtained by centrifugation at 12,000 rpm for 10 min at 4 °C, and transferred to GC-compatible vials. The samples were stored at −80 °C. Prior to GC‒MS analysis, MS data were obtained. A Thermo Scientific Trace 1310 Gas Chromatograph coupled with a Thermo Scientific ISQ™ QT Mass Spectrometer was used for the analysis. The data were analysed using Thermo TraceFinder version 4.0 software.

Luciferase reporter assays

Luciferase assays were performed in UM cells with the reporter 8xGTIIC-Lux. The luciferase reporter was transfected together to normalize for transfection efficiency. Firefly luciferase activity was measured with an Infinite F200PRO plate reader using luciferin as a substrate. The data are presented as firefly/β-galactosidase activity. Each sample was transfected in triplicate, and this process was repeated three times.

Metabolite change measurement

The cells were seeded in 6-well plates and allowed to adhere overnight. The following day, the culture medium was replaced with fresh medium, and the cells were incubated for an additional 24 h. Glucose uptake was determined using the Glucose Uptake Colorimetric Assay Kit from BioVision (catalogue number K676-100). The absorbance was measured at 450 nm using a microplate reader. Lactate secretion was measured using the Lactate Colorimetric Assay Kit from BioVision (catalogue number K627-100). The absorbance was measured at 450 nm. Glutamine uptake was determined using the Glutamine/Glutamate Determination Kit from Sigma‒Aldrich (catalogue number GLN1). The absorbance was measured at 340 nm.

Colony formation

Cell colony formation was detected via crystal violet staining. A volume of 1 mL of complete medium containing 1 × 103 cells was added to each well of a 6-well plate. The plate was stained with 0.25% crystal violet after 1–2 weeks.

Flow cytometry assay

OMM1, OMM2.3 and PIG1 cells (3 × 104) were cultured in 24-well plates overnight. Following digestion with 0.25% trypsin (without EDTA) and washing with PBS, apoptosis was identified by double staining with Annexin V-FITC and propidium iodide (BD Biosciences, USA). Apoptosis and cell cycle were quantified using a flow cytometer (FACSCalibur, BD Biosciences, USA) and analyzed using the FlowJo software.

Statistics and reproducibility

GraphPad Prism 7 (GraphPad Software) was used for statistical analysis. All of the experiments were performed in triplicate, and the data are presented as the means ± standard deviations (SDs). The comparative threshold cycle method was used according to the ΔΔthreshold cycle method. The differences between two groups were analysed with the unpaired two-sided Student’s t-test or the Mann‒Whitney U test. A p-value < 0.05 was considered statistically significant and is indicated with asterisks. All experiments utilized 2–10 replicates and are representative of multiple independent experiments (typically 3).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Results

Active landscape of SE profiles in uveal melanoma

H3K27ac is a key characteristic of enhancers, particularly SEs, which are defined as having strong enrichment over extended regions27. To explore active enhancer profiles in UM, we designed an epigenomic pipeline to identify SE regulators in UM (Fig. 1A). First, using normal melanocytes as a comparator, we defined enhancers with recurrently gained activity across UM cell lines (Fig. 1B) by ChIP-seq. A total of SE-associated genes were found in OMM1, 92.1 and OMM2.3 cells, with 1,085 identified in OMM1, 1422 in 92.1, and 1303 in OMM2.3, respectively. Intriguingly, most SEs occupied some known lineage genes involved in melanocyte differentiation and melanin production, such as tyrosinase-related protein 1 (TYRP1), melanocyte-inducing transcription factor (MITF) and melan-A (MLANA) (Fig. 1B). We also revealed that oncogenes such as TFAP2A, RAS guanyl releasing protein 3 (RASGRP3), sex determining region Y-box transcription factor 10 (SOX10) and MITF (Fig. 1C) were significantly enriched in UM cells. In contrast, H3K27ac signals were low in the corresponding genes in normal melanocytes (Fig. 1B). TFAP2A is a transcription factor (TF) that binds to specific regions of DNA to control gene expression, notably to regulate melanocyte differentiation28. Our data suggest that SE regulatory elements represent active chromatin marks on lineage-specific genes and oncogenes.

Fig. 1: The epigenetic landscape of uveal melanoma.
figure 1

A Pipeline of the study. B H3K27ac binding at distal enhancers ranked by increasing signal intensity in cell lines. Super enhancers (SEs) were identified as those beyond the inflection point where a rapid increase in signal was observed (indicated by a dashed grey line). TF genes associated with SEs are indicated in blue. C ChIP-seq tracks for the TFAP2A, RASGRP3, MITF, and SOX10 SE loci. D Strategy for discovering master transcription factors (MTFs) from the RNA-seq and ChIP-seq datasets. E Motif analysis at the H3K27ac sites in the UM cell lines.

Full size image

TFs tend to bind to enhancers and SEs to define the cell fate, among which MTFs are produced at extremely high levels and can bind most of the enhancers in a stoichiometric manner. In accordance with the established MTF identification pipeline23 for tumours (Fig. 1D), we identified the MTFs regulated by SEs that presented considerable pattern enrichment in enhancers. These MTFs also presented relatively high levels of expression (average FPKM > 20) and were more highly expressed than in normal tissues. The results revealed a list of 26 TFs that were consistently overexpressed, suggesting a convergent underlying epigenetic state. Since MTFs are expressed at high concentrations to maintain their enrichment, we then filtered TFs with low expression levels as well as those expressed in normal cells (cutoff FKPM > 20). Therefore, we generated a series of potential TFs with significantly high levels of lineage-specific expression compared with those in normal tissues. Compared with normal choroidal tissue, MTFs presented significant motif enrichment in enhancers. In particular, motif analysis revealed significant enrichment of MITF and TFAP2A binding sequences in UM cell lines (medium-log p value: tumour 193.3; normal 56.49) (Fig. 1E), indicating that the recognition sequences of TFAP2A were highly significantly enriched at SE locations according to motif analysis. Taken together, these results revealed that TFAP2A, as a core MTF, was strongly enriched in SEs.

Biological features of SE-associated transcripts

Using an integrative analysis of epigenetic and transcriptional profiles, we identified an SE-driven MTF modulator in UM. We first searched for pathways activated by TFAP2A via GSEA (Fig. S1A) and predicted TFAP2A interactors using the STRING database (Fig. S1B) (https://string-db.org/network/), initially exploring the function of the transcription factor TFAP2A. Since the most uniquely and highly expressed MTFs were driven by SEs, to further characterize the effects of MTFs on chromatin, we next sought to investigate the histone modifiers H3K27ac and H3K4me1 and the binding of TFAP2A and EP300 binding to SEs. To further investigate whether TFAP2A expression was driven by an open chromatin structure, we also performed ATAC-seq to identify accessible chromatin regions in UM cells. We integrated these sequencing results in UM cell lines to identify a larger set of targets and enriched sites (Fig. 2A). Our results revealed that TFAP2A directly occupies an SE upstream of the TFAP2A locus to drive its expression, confirming that this MTF is self-promoted by an SE. The results revealed that accessible regions were cooccupied with H3K27ac and H3K4me1 signals. We next determined whether TFAP2A was enriched in promoter and intergenic regions (Fig. 2B). After peak calling, 19,887 (p < 0.000001) peaks were found in TFAP2A binding regions; these exhibited the highest statistical significance and the highest intensity. Genome-wide, the majority of the TFAP2A binding regions were located within introns and intergenic regions. Additionally, using ChromHMM analysis to define active chromatin regions, we found that TFAP2A predominantly resides in active transcription start sites and active enhancers, which harbour the active enhancer marks H3K27ac and H3K4me1 (Fig. 2C). Such interactions spanning broad one-dimensional sequence distances complicated the identification of target genes. Next, we determined the occupancy of TFAP2A in SEs over typical enhancers. Notably, TFAP2A bound a negligible proportion of typical enhancers (21.4%) but an abundance of SEs (61.55%) (Fig. S2A). Its predominance for SEs over typical enhancers was striking. We established the genomic loci within which TFAP2A enhancers may exist (Fig. S2B), and we investigated the topologically associated domain (TAD) structure around the TFAP2A locus on the basis of Hi-C data from Juicebox (Fig. S2C). We found that TFAP2A expression was modulated by a large ~2.4 Mb regulatory TAD region, where binding of the enhancer marker H3K27ac was correlated with the expression of TFAP2A.

Fig. 2: TFAP2A-dependent epigenetic transcriptional profiles in UM.
figure 2

A Normalized ChIP-seq tracks for H3K27ac, H3K4me1, TFAP2A and EP300, ATAC-seq and RNA-seq at the TFAP2A loci in OMM2.3 cells; ChIP-seq read signals (y axis) were normalized to the RPKM value in the corresponding sample. B TFAP2A peak distribution in the OMM2.3 cell line. The scatter plot is accompanied by a bar plot and upset plot showing the amounts of TFAP2A in intronic, exonic, intergenic or promoter-proximal sites. C Chromatin states in OMM2.3 cells (left) and relative TFAP2A density (right). States were defined using ChIP-seq data for all histone modifications using the hidden Markov modelling algorithm (ChromHMM). Frequency refers to the probability of each mark being present in a given state. D Heatmaps of TFAP2A density in OMM2.3 cells are shown accordingly. Normalized ChIP-seq tracks for H3K27ac, H3K4me1, and TFAP2A transcription factors at TF2P2A binding loci in OMM2.3 cells. ChIP-seq read signals (y-axis) were normalized to the RPKM value in the corresponding sample. E Representative heatmaps of H3K27ac density in the OMM2.3 cell line and UM tissue. Differences between cell lines and tumour tissue at common enhancer sites and significantly enriched enhancer sites are shown.

Full size image

Next, we created a map of enhancers in UM cell lines and tumours, which contained all the enhancer sites. We therefore discovered TFAP2A-SE loci that were highly enriched in UM cell lines and overlaid TFAP2A occupancy. We also compared the enhancer histone marker H3K27ac with that of TFAP2A sites and identified 12,932 TFAP2A-colocalized enhancers in the OMM2.3 cell line (Fig. 2D). To further determine whether these findings are applicable to human malignancies and highly relevant to UM oncogenesis, we next used CUT&Tag to examine SE occupancy in UM specimens. As expected, TFAP2A-associated SEs were highly enriched in tumours, suggesting their protumorigenic function, and H3K27ac enrichment in UM tumour tissues and cell lines was associated with a specific signal intensity (Fig. 2E).

The activation of TFAP2A is correlated with oncogenic activity

To elucidate the mechanism of epigenetic reprogramming mediated by TFAP2A, we first examined the expression of TFAP2A in UM tissues and adjacent normal choroidal tissues. We used immunofluorescence staining to examine TFAP2A expression in vivo. In accordance with the epigenetic results, TFAP2A expression was significantly upregulated in tumours and showed stronger nuclear expression in tumours than in normal adjacent control tissues (Fig. 3A). We then quantified the higher expression of TFAP2A in tumour tissues (Fig. 3B) and examined the expression of TFAP2A in UM cell lines via RT‒PCR analysis (Fig. 3C). The results revealed greater expression of TFAP2A in UM cell lines than in normal cell lines. As expected, TFAP2A was more highly expressed in most UM specimens, and this conclusion was also verified in the TCGA-UM cohort (Fig. 3D). Since copy number amplification is a common mode of oncogene activation in cancer, we next sought to identify the copy number of the TFAP2A gene. SCNA results using pancancer data revealed that the TFAP2A locus presented a remarkably amplified copy number (Fig. S3B), which was highest in the UM cohort (Fig. S3C). The correlation between TFAP2A and gene expression was highly significant (Spearman r = 0.63) (Fig. 3E). Multigene analysis also revealed a high frequency of TFAP2A amplification in the UM cohort (Fig. 3F). These showed that TFAP2A was a crucial master regulator of UM with aberrantly amplified copy numbers and pronounced gene expression.

Fig. 3: Focal amplification drives TFAP2A expression.
figure 3

A Representative images of TFAP2A immunofluorescence staining are shown at 20x magnification for UM and normal control samples. B TFAP2A immunofluorescence staining of normal and UM tissues. C TFAP2A RNA expression was measured in UM cell lines using real-time PCR. *P < 0.05, **P < 0.01, ***P < 0.001. D RNA expression analysis of TFAP2A expression in the TCGA-UM cohort. The samples were classified according to TFAP2A copy number. E Correlations between TFAP2A expression and copy number in the TCGA cohort of UM samples (Spearman correlation, p < 0.001). F Multigene analysis of UM-specific oncogenes as well as TFAP2A amplification was shown using cBioPortal.

Full size image

TFAP2A promotes UM formation in vitro and in vivo

The above findings prompted us to propose TFAP2A as a potent epigenetic stimulator that prevents the decommissioning of gene enhancers during UM oncogenesis. Small interfering RNA (siRNA) was performed, and we observed a significant reduction in cell proliferation in UM cell lines when TFAP2A was targeted by siRNAs (Fig. 4A), further suggesting that TFAP2A is an MTF in UM. To gain insight into the oncogenic properties of TFAP2A, we designed two sgRNAs in exon 1 of TFAP2A to silence its expression. sgRNAs targeting TFAP2A consistently reduced its expression in the OMM2.3 and 92.1 cell lines (Fig. 4B, S10). Similar results were obtained for the tumour-suppressive effect of TFAP2A knockdown (KD) using in vitro proliferation and apoptosis assays (Fig. 4C, Fig. S4A, B, Fig. S5). Compared with the control, stable TFAP2A-KD in UM cells using CRISPR-Cas9 dramatically compromised the proliferation and promoted the apoptosis of UM cells (Figs. 4C, S4A, B), suggesting the reduced proliferative potential of UM, and we found that it regulated UM cell cycle distribution in vitro (Fig. S4C). We also knocked down TFAP2A in 786-O renal cell carcinoma cells using the same sgRNAs and did not observe its ability to inhibit cell proliferation (Fig. S5). Furthermore, we interfered with SOX10 expression with siRNA and found no significant change in TFAP2A expression (Fig. S6), confirming that TFAP2A directly promotes UM formation in vitro. We next evaluated the effect of TFAP2A knockdown in vivo in UM cells stably expressing scramble or TFAP2A-sgRNAs. TFAP2A-KD resulted in a significant decrease in UM tumour growth in vivo (Fig. 4D, E). Taken together, our data strongly suggest that TFAP2A is a UM-specific MTF for UM formation and a potential driver of UM progression.

Fig. 4: TFAP2A knockdown suppresses UM tumour growth.
figure 4

A CCK8 analysis of OMM2.3 and 92.1 cells transfected with the indicated siRNA against TFAP2A. B Western blot analysis of TFAP2A expression in cells stably transfected with the indicated sgRNAs. GAPDH was used as a loading control. C CCK8 analysis of OMM2.3 and 92.1 cells stably transfected with the indicated sgRNAs against TFAP2A. D In vivo tumour formation by OMM2.3 and TFAP2A-KD cells. K‒M curves for the sgNC and sgTFAP2A groups. E IVIS signals in the control and TFAP2A depletion groups. *P < 0.05, **P < 0.01, **P < 0.001.

Full size image

TFAP2A-biologically associated SEs recapitulate the transcriptional machinery of primary tumours

To examine pathways downstream of TFAP2A, we used CRISPR-Cas9 to knock out TFAP2A expression and conducted RNA-seq experiments. After TFAP2A silencing, the expression of 222 genes was significantly elevated, and 212 genes were suppressed (Fig. 5A, B). Notably, SE-mediated transcription was preferentially impacted by TFAP2A depletion. TFAP2A silencing affected the majority of SE gene transcription (Fig. S2D). We next examined the gene expression datasets for verification. The increased expression of these SE target genes in UM was validated via pathway analysis. The functional annotation of TFAP2A targets revealed few TFAP2A-specific processes, while most were common and strongly associated with translation and metabolism (Fig. 5C). Notably, metabolic processes were highly involved, in which a large category of genes were involved in KEGG HSA00270 CYSTEINE AND METHIONINE METABOLISM and KEGG HSA00250 ALANINE ASPARTATE AND GLUTAMATE METABOLISM (Fig. 5D). The TFAP2A-regulated genes were subjected to pathway analysis, which allowed the identification of TFAP2A signatures. The TFAP2A-associated genes were enriched mostly in metabolic processes such as “biosynthesis of amino acids”, “cysteine and methionine metabolism”, and “metabolic pathways” (Fig. 5E, F). Therefore, we speculated that TFAP2A may be involved in the metabolic processes of UM oncogenesis.

Fig. 5: Transcriptional analysis revealed key metabolic changes after TFAP2A depletion.
figure 5

A Volcano plot of gene expression in OMM2.3 sgNC and TFAP2A-KD cells. B Venn diagram of DEGs (downregulated and upregulated genes) and ChIP-seq TFAP2A targets. C Pathway analysis of cells stably transfected with the indicated sgRNA against TFAP2A. D KEGG analysis of TFAP2A expression in cells stably transfected with sgNC and TFAP2A sgRNAs. E Network combining the GSEA results in the sgNC and sgTFAP2A cells. F Representative ridge plot of significant pathways in the sgNC and sgTFAP2A cells.

Full size image

Direct activation of SE-associated gene transcription by TFAP2A

To gain insight into the relationship between TFAP2A enhancer activity and target gene expression, ChIP-seq data were integrated with RNA-seq data to compare SE-associated expression under various conditions. We first compared the enrichment of TE and SE H3K27ac signals in UM cell lines. As defined by ROSE algorithms, SEs are associated with significantly high H3K27ac signals compared with TE regions (Fig. 6A). Notably, the expression of SE-associated genes was significantly greater than that of TE-associated genes (Fig. 6B). In addition, we separated the SE and TE genes on the basis of TFAP2A occupancy. Interestingly, we found that TFAP2A-associated genes presented much greater gene expression than did genes without TFAP2A occupancy, especially SE genes (Fig. 6C), suggesting that TFAP2A may promote lineage-specific gene expression through SEs. To explore high-confidence targets of TFAP2A, we integrated ChIP-seq and RNA-seq data to identify target genes. We first analysed TFAP2A ChIP-seq data and overlapped with SE regions to identify TFAP2A-bound SEs in UM cell lines. RNA-seq of genes whose expression was downregulated by TFAP2A-KD was subsequently used to identify direct targets of TFAP2A. Finally, after removing genes with low expression in UM, we identified SLC7A8 as the prominent node of TFAP2A targets (Fig. S1C). To gain insight into the epigenetic role of SE‒MTF interactions, we performed ChIP‒seq assays to examine the epigenetic landscape of the SLC7A8 locus. The data revealed a 50 kb SE encompassing the SLC7A8 promoter with prominent H3K27ac and H3K4me1 levels, whereas these enhancer markers were extremely low in normal cells (Fig. 6D). In addition, TFAP2A was found to be involved in the SE region (Fig. S7). To investigate the role of TFAP2A in regulating target gene expression, we depleted TFAP2A using two independent sgRNAs. To further determine whether these in vitro findings are applicable to human malignancies, pancancer analysis revealed significant high expression of the SLC7A8 gene across the cohort (Fig. 6E). Next, we compared the TCGA-UM cohort of two distinct groups, those with low and high SLC7A8 expression levels, with respect to patient prognosis. We distinguished the high and low groups by the median value of the SLC7A8 expression level. Notably, high SLC7A8 expression was associated with poor prognosis in the TCGA cohort for overall survival and in the Laurent cohort (GSE22138)29 for metastasis-free survival (Fig. S8), suggesting a potential role for SLC7A8 in patient survival.

Fig. 6: TFAP2A orchestrates super enhancer activity.
figure 6

A H3K27ac signal in the super enhancer (SE) and typical enhancer (TE) regions in OMM2.3 cells, as revealed by ROSE analysis. B Boxplots of SE- and TE-associated gene expression in OMM2.3 cells. C TFAP2A-associated gene expression in the TE and SE regions in OMM2.3 cells. RNA expression was normalized using FPKM. D IGV plot showing that SLC7A8-SE regions were enriched in the H3K27ac/H3K4me1 signal and in TFAP2A and EP300 binding sites. E Pancancer analysis showing that the expression of SLC7A8 was the highest in the UM dataset. *P < 0.05, **P < 0.01, ***P < 0.001. ****P < 0.0001.

Full size image

TFAP2A silencing inhibits UM metabolism

We next determined whether TFAP2A mediated glucose metabolism under various conditions. As expected, TFAP2A silencing dramatically inhibited glycolysis, as determined by lactate secretion analysis. Moreover, TFAP2A depletion significantly reduced glutamine and glucose uptake (Fig. 7A, B), indicating that TFAP2A silencing inhibited basal energy substance uptake by tumour cells. In addition, GC‒MS was used to collect intracellular metabolites from TFAP2A wild-type and KD cells (n = 4 biological replicates) (Fig. 7C, D). The heatmap showed that TFAP2A induced metabolic reprogramming in UM. SLC7A8 is the prominent downstream node of TFAP2A and SLC7A8 serves as an amino acid transporter in metabolic processes. We determined that the knockdown of SLC7A8 could impair colony formation and basic metabolite uptake in UM cells, similar to the effects of TFAP2A-KD induction (Fig. S9). Our observations support the mechanism that elevated TFAP2A expression drives amino acid uptake through SLC7A8-mediated pathways (Fig. 7E).

Fig. 7: Inhibition of TFAP2A compromises metabolic reprogramming in UM cells.
figure 7

A, B OMM2.3 and 92.1 cells were selected and cultured in fresh medium for 24 h. Subsequently, glucose uptake, lactate secretion, and glutamine consumption were analysed and normalized. C Heatmap showing the amounts of lipid metabolites in the control and TFAP2A-depleted OMM2.3 samples. Gas chromatography‒mass spectrometry (GC‒MS) was used to collect intracellular metabolites from TFAP2A wild-type and KD cells (n = 4 biological replicates). D Heatmap showing the amounts of amino acid metabolites in the control and TFAP2A-depleted OMM2.3 samples (n = 4). E The overall schematic diagram of the study. *P < 0.05, **P < 0.01, ***P < 0.001.

Full size image

Discussion

Oncogene mutations and chromosomal aberrations are the most recurrent genetic aberrations in primary uveal melanomas30. However, epigenetic aberrations associated with distinct oncogenes and tumour suppressor genes have yet to be elucidated. We conducted an integrative analysis of UM-specific SEs and gene activation patterns to clarify the significance of the regulation of SEs in the characteristics of malignancy. We sought UM-specific SEs to determine whether SEs are responsible for gene regulation in dysfunctional signalling pathways in UM. In particular, we were able to identify UM-specific SE target gene pairings, which are strongly linked to characteristics of malignancy, such as RASGRP3 and MITF31,32,33. Through a comprehensive analysis of multiomics data, our epigenomic results revealed a highly overexpressed MTF, TFAP2A, which has biological significance in accelerating UM tumorigenesis. RNA-sequencing data comparing primary UMs with adjacent normal tissues revealed candidate TFs with robust expression. Our findings suggest that the UM epigenome is characterized by epigenetic aberrations at enhancers that activate a common, aberrant transcriptional program critical for UM growth and survival. Notably, UMs maintain pluripotency while having exceptionally high levels of TFs.

SEs are a specialized clusters of active enhancers densely enriched with regulatory circuitry and active histone marks34. And SEs encompass the strongest potent enhancers connected to the central regulating MTF machinery35. Intriguingly, we found that the core MTF “TFAP2A” most frequently occupied the strong enhancer chromatin state. Our findings highlight a melanotic transcriptional program mediated by TFAP2A in UM oncogenesis. SE-associated genes were much more highly expressed than typical and normal genes were, and the enhancers hijacked by MTFs were the most highly expressed.

Small-molecule inhibitors or dietary management for targeting cancer metabolism are attractive strategies for anticancer treatment36. In addition to growth hormones, the homeostasis of amino acids (AAs) is critical for the regulation of mTORC1 activity37. In particular, leucine18 is essential for mTORC1 activation. In a sodium-independent manner, LAT2 is responsible for the intracellular absorption of LNEAA in exchange for glutamine. In addition to identifying the molecular pathways related to UM, our research enabled us to identify a pool of candidates whose expression was strongly associated with UM. Among these proteins, we focused on SLC7A8, a putative transcription coregulator whose overexpression in UM has not been previously reported. SLC7A8 is a key amino acid supplier to the metabolic programs that sustain tumour amino acid metabolism38. In particular, SLC7A8 is controlled by the master regulator TFAP2A and promotes UM oncogenesis by activating metabolic pathways. For the treatment of advanced UM, overcoming mTORC1 pathway activation might be a potential therapy. Furthermore, the branched-chain amino acid (BCAA) catabolism replenishes the tricarboxylic acid cycle and is dysregulated in UM, whereas targeting essential amino acid (EAA) metabolism might be a novel treatment approach for UM39. In this study, we characterized the epigenomes and metabolome specifically to investigate UM oncogenesis. Detailed characterization of these models might reveal pathways and molecular markers shared by UM.

Moreover, in vitro TFAP2A expression in several UM cell lines was reciprocally altered following the modulation of TF function. Consistent with the protumorigenic activity of TFAP2A in UM, CRISPR-mediated KD of TFAP2A in UM cells decreased cell migration and further inhibited UM tumour growth. The function of TFAP2A in cancer remains contentious. In contrast to our findings, early investigations of several malignancies revealed a correlation between high TFAP2A expression and melanocyte differentiation28. Besides, TFAP2A inhibition led to decreased cell motility and invasion, confirming the involvement of TFAP2A as an oncogene. Conversely, the protumorigenic features of TFAP2A have been identified in melanomas. We identified SLC7A8 as a target of TFAP2A in UM by integrating transcriptomics and proteomics approaches. SLC7A8 regulates the cellular transport of BCAAs in exchange for glutamine. High SLC7A8 expression correlated with UM progression. In addition, the inhibition of LAT-dependent AA transport has yielded encouraging outcomes in preclinical models. CRISPR-mediated TFAP2A KD reduced SLC7A8 levels in UM cells, mechanistically linking TFAP2A to SLC7A8 expression through transcriptional regulation. These findings are important for the exploration of cancer therapeutic targets at SEs in UM. By mapping the enhancer landscape and the target in UM, we have uncovered the therapeutic potential of drugs targeting SE-driven transcription and amino acid metabolism in UM.