AP-2α–Mediated Activation of E2F and EZH2 Drives Melanoma Metastasis

Localized melanoma is readily cured by surgical excision and stage I disease carries a 5-year melanoma-specific survival rate of 98% (1). Melanoma prognosis dramatically worsens with metastasis, causing the 2-year survival rate of patients with distant metastases to be only 10% (2). Currently, adjuvant therapies for advanced stage melanoma predominantly involve combination MAPK inhibitors (such as dabrafenib and trametinib therapy) or immunotherapy (such as anti-PD-1 therapy; ref. 3). However, relapse and severe adverse responses to therapy continue to occur in the majority of patients receiving these treatments. Furthermore, there are no targeted therapies approved for tumors lacking BRAF mutations, such as NRAS-activated melanomas. Thus, our current repertoire of adjuvant therapeutic options is insufficient to elicit broadly effective antimetastatic responses and there is a dire need for additional targeted therapies.

Transcription factor AP-2α, encoded by the TFAP2A gene, is a retinoic acid inducible protein critical to maintain the viability of migratory neural crest cells during vertebrate development. Studies in AP-2α–deficient zebrafish embryos have shown neural crest cells, including pigment-producing melanophores, die by apoptosis soon after initiating migration (4). In healthy vertebrate embryos, AP-2α maintains cellular viability and drives differentiation of melanocytes in cooperation with MITF (5), a master regulator of the melanocytic lineage.

Owing to its developmental ancestry, metastatic melanoma retains a neural crest-like resemblance. Using a chick embryo model, Kulesa and colleagues transplanted human melanoma cells in vivo adjacent to premigratory neural crest cells (6). Melanoma cells traveled along stereotypical neural crest migratory pathways to populate distal locations while avoiding canonical neural crest–free zones, implying a mechanistic parallel between melanoma migration and its progenitor neural crest and suggesting the involvement of AP-2α in melanoma metastasis. The frequently cited phenotype switching model of melanoma, which attributes melanoma cells a plasticity involving a reversible switch between proliferative and invasive phenotypes, recognizes a dedifferentiated neural crest stem cell (NCSC)-like subset of invasive cells. The NCSC subpopulation is characterized by high expression of the AP-2 paralog AP-2β and the outgrowth of NCSCs within a melanoma can be prevented by retinoic acid receptor antagonists, implying a necessity for the AP-2 transcription factors for maintenance of this invasive phenotype (7).

A consistent influence of AP-2α in cancer remains unclear. TFAP2A is overexpressed in basal-squamous bladder cancer and is associated with lymph node metastasis (8). Likewise, TFAP2A overexpression in nasopharyngeal carcinoma positively correlates with tumor stage, local invasion, and decreased overall survival (9). Contradicting studies in breast and colon cancer have reported high expression of AP-2α drives proliferation and invasiveness (10) or can act as a tumor suppressor (11). In melanoma, early publications showed that loss of nuclear AP-2α was associated with disease progression (12) and exogenous overexpression of TFAP2A reduced primary tumor growth and metastasis (13, 14). However, a recent clinical study challenged this paradigm by correlating high TFAP2A expression in melanoma lymph node metastases with poor prognoses and an increased risk of recurrence (15).

The influence of the developmentally cooperative MITF in melanoma is heavily dependent on its level of expression (reviewed in ref. 16). While high expression of MITF drives melanomas into a well-differentiated and lowly invasive phenotype, depletion paradoxically elicits senescence. This led to the hypothesis that melanomas show lineage addiction, a cell-type specific reliance on a developmentally critical factor, to the survival cues provided by basal MITF expression. This observation was furthered by the MITF rheostat model explaining that a spectrum of MITF activity differentially alters phenotypes. MITF^Low melanomas are lowly proliferative but highly invasive, and MITF^High melanomas are lowly invasive but highly proliferative. However, either prolonged depletion or overexpression of MITF can elicit growth arrest.

The disparate articles surrounding the role of AP-2α in melanoma led us to consider whether the level of TFAP2A expression differentially influences melanoma progression. We hypothesize that the basal expression of TFAP2A observed in advanced melanomas drives a cytoprotective transcriptome that parallels neural crest migration and is necessary to maintain viability during metastasis. The early studies examining phenotypes driven by overexpression of TFAP2A would have overlooked the cytoprotective effects afforded from basal expression of TFAP2A during metastasis. Experimentally testing this hypothesis and determining prosurvival targets of AP-2α can therefore uncover novel antimetastatic therapies specific to advanced melanoma. Herein, we show AP-2α is necessary for melanoma metastasis via transcriptional activation of E2F pathway targets including EZH2, a druggable driver of metastasis.

All cells were propagated and cultured in accordance with instructions from the providing institution and utilized with 15 or fewer passages. The A375 and SKMEL28 cell lines were obtained from ATCC. The M21 and TKLP cell lines were obtained as described in the Acknowledgements section. The A375, SKMEL28, M21, and TKLP cell lines were cultured in DMEM medium supplemented with 10% FBS, 1% pen-strep, and 0.2% Plasmocin and split 1:7 upon reaching confluency. Cell lines were not further authenticated. The patient-derived cell lines MB4667, MB4562, and MB4479 were obtained from the Center for Rare Melanomas Biorepository at the University of Colorado (Boulder, CO). Patient-derived cell lines were cultured in RPMI1640 medium supplemented with 10% FBS, 1% Pen-Strep, and 0.2% Plasmocin and split 1:7 upon reaching confluency. Cells were tested for Mycoplasma contamination using LONZA MycoAlert Mycoplasma Detection Kit. All cells were cultured at 37°C with 5% CO₂.

All mouse experiments were approved by and performed in accordance with the University of Iowa Institutional Animal Care and Use Committee (IACUC).

A375 was obtained from ATCC. To generate complete genetic knockout (KO) of TFAP2A, we used single-guide RNAs (sgRNA) independently targeting the + and − DNA strands of exon 3 designed as gBlocks Gene Fragments from Integrated DNA technologies as described previously (17) with the following sequences: Guide A: ATCAAACTGTAATTAAGAA and Guide B: TTCTACATGCTGCAACAAA. gBlocks were cloned into the Zero Blunt Topo vector (Invitrogen). sgRNAs were cotransfected using Lipofectamine 2000 (Invitrogen) alongside a bicistronic GFP-Cas9^D10A “Nickase” expression vector (Addgene #44720; ref. 18) to reduce the likelihood of off-target mutations. GFP-positive cells were sorted into 96-well plates after a 24-hour incubation and screened by PCR using Platinum Blue PCR SuperMix (Invitrogen) using the following primer sequences: Forward: AGCTAGCCTGTTGGCATTACC, Reverse: CCTCTTGAGTTGCAAAGCCC. Mutations were confirmed by Sanger sequencing and knockout confirmed by Western blot analysis.

Western blots were performed with anti-AP-2α (Abcam, ab108311), anti-GAPDH (Santa Cruz Biotechnology, sc-32233), anti-EZH2 (Novus, NBP2-17087), anti-RRM2 (Novus, NBP1-31661), anti-RAD51 (Santa Cruz Biotechnology, sc-8349), anti-TYMS (Invitrogen, PA5-27867), anti-CDC6 (Novus, NBP2-15837), anti-E2F1 (Santa Cruz Biotechnology, sc-193), anti-E2F2 (Invitrogen PA5-79179), and anti-E2F8 (Bethyl, A303-039A-M).

A375 (passage 1) or TFAP2A^KO2 were luciferized via transduction with bicistronic EF1a-Luciferase-P2A-GFP lentiviruses (GenTarget, LVP438-PBS). A total of 10⁶ cells were mixed with Matrigel (Corning) and subcutaneously xenografted into the flanks of NOD/SCID mice. Primary tumors were excised by surgical physicians blinded to the xenograft type in survival surgeries 28 days after xenograft. Mice were bioluminescently imaged 28 days after surgery, euthanized, and lungs and livers were bioluminescently imaged ex vivo using an IVIS Illumina after allowing 10 minutes for perfusion. Images were processed and quantified using the Living Image software. Quantifications were compared using two-tailed Student t tests.

Subconfluent cells were plated and transduced for 1 hour in a minimal amount of serum-free (Opti-MEM, Gibco) medium. Cells were utilized for soft agar assays or RNA/protein extraction 96 hours after transduction. Short hairpin RNAs (shRNA) were obtained from Sigma-Aldrich [shNT: SHC002V, shTFAP2A (Human): TRCN0000004926, shTfap2a (Mouse): TRCN0000012049].

Soft agar assays were performed as described previously (19). Unless otherwise noted, 5,000 cells per replicate were plated for established cell lines or 7,500 cells per replicate for patient-derived xenografts (PDX) and incubated between 2 and 4 weeks, depending on proliferation rate. The parental A375 cell line was used for these assays with minimal passaging (≤3 passages). Assays were performed in triplicate, with one representative picture per replicate quantified and compared using Student t tests.

For the parental A375 versus TFAP2A^KO single-cell RNA sequencing (scRNA-seq), and for shNT versus shTFAP2A scRNA-seq, subconfluent (≈ 40% confluent) cells were trypsinized, prepared according to 10X Genomics sample preparation guidelines, and counted by hemocytometer. Cells were suspended in 0.04% BSA/PBS and libraries were prepared using 10X Genomics v3 Chemistry targeting 5,000 cells per condition and sequenced on an Illumina HiSeq 4000 with 150 bp, paired end reads. All libraries were above the 10X Genomics recommended minimum of 20,000 average reads per cell.

A custom reference genome accounting for polymorphisms within the A375 genome was generated by aligning publicly available A375 whole-genome sequencing reads (NCBI SRA: SRR8639175) to the GRCh38 reference genome using BWA-MEM, converted to a BAM file (Samtools), duplicates removed (GATK), and sorted (GATK). This alignment was then used to create an A375-specific genome assembly with RGAAT v2 (20). 10X Genomics scRNA-seq reads were then processed and aligned to this assembly with Cell Ranger (Cell Ranger Count function). All scRNA-seq figures in this article were generated using R.

For Seurat analysis of A375 and its derivatives, cells with fewer than 200 RNA feature counts and greater than 10% mitochondrial contamination were removed via filtering. Counts were then normalized (NormalizeData function with default parameters), followed by the FindVariableFeatures function with the following parameters: selection.method = “mean.var.plot,” nfeatures = 7,000. To generate the t-SNE plot shown in Fig. 2D, the RunPCA function was performed with default parameters followed by the RunTSNE function with following parameters: theta = 0.0, perplexity = 100. Clusters were then determined by the FindNeighbors function with dims 1:30 followed by the FindClusters function with a resolution of 0.32, which identified 6-cell clusters. Seurat analysis of the mouse datasets was performed using standard parameters following alignment to mm10 (Cell Ranger Count). Seurat differential expression analysis was performed using MAST. For the linked interactive datasets, libraries were aggregated (Cell Ranger Agg function) and reprocessed to calculate a three-dimensional t-SNE (Cell Ranger Reanalyze function with the following parameters: tsne_max_iter = 10,000, tsne_max_dims = 3, tsne_theta = 0), then uploaded to the Broad Institute Single Cell Portal. The projection and cluster numbering of these datasets will differ from the Seurat and Monocle3 generated plots within article figures above. The shNT versus shTFAP2A dataset is hosted at: https://singlecell.broadinstitute.org/single_cell/study/SCP1161. The parental A375 versus TFAP2A^KO4 dataset is hosted at: https://singlecell.broadinstitute.org/single_cell/study/SCP1160. scRNA-seq data are also hosted by NCBI Gene Expression Omnibus (GEO) under accession GSE162362.

RNA was isolated from the four TFAP2A knockout clones using the RNeasy Plus Mini Kit (Qiagen) and sequenced with 75 bp, single end reads. Reads from parental A375, A375 shNT, and A375 shTFAP2A scRNA-seq had the 28 base 10X Genomics UMIs/barcodes removed from the start of R1 and the first low-quality base removed from the start of R2 using the HEADCROP operation of Trimmomatic. All reads were aligned using RNA STAR with parameters for Cufflinks compatibility. Differential expression was then calculated using Cufflinks/Cuffmerge/Cuffdiff. The RNA-seq heatmap in Fig. 2 was generated using Morpheus from the Broad Institute. Pathway analysis was performed with the Ingenuity Pathway Analysis software (Qiagen). RNA-seq data are hosted by NCBI GEO under accession GSE162362.

RNA was isolated using the RNeasy Plus Mini Kit (Qiagen) and quantified using a Nanodrop (Thermo Fisher Scientific). A total of 1 μg of RNA was used for reverse transcription using the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems) and qRT-PCRs were performed on a QuantStuido 6 Flex instrument (Applied Biosystems) using TaqMan Universal PCR Master Mix (Applied Biosystems) and predesigned primer/probe sets purchased from Thermo Fisher Scientific. qRT-PCRs were performed in technical triplicate.

Unsynchronized cells were trypsinized and fixed for 1 hour with cold 70% ethanol, washed with PBS, and suspended with cold 10% FBS/PBS. RNA was eliminated with RNase A (Qiagen) and cells were stained with propidium iodide (G-Biosciences) for 30 minutes on ice in the dark. DNA quantification was performed on a FACS Becton Dickinson LSR II flow cytometer as described in ref. 21.

AP-2α chromatin immunoprecipitation sequencing (ChIP-seq) was performed using anti-AP-2α 3B5 from Santa Cruz Biotechnology (SC-12726 X) as described previously (22). ChIP-seq data are hosted by NCBI GEO under accession GSE162363.

ChIP was performed as described above using Mouse IgG (Sigma, 15381), anti-H3K9/K14Ac (Invitrogen, 49-1010), anti-H3K27Ac (Millipore, 07-360), anti-GATAD2B (Abcam, ab76924), anti-HDAC2 (Invitrogen, PA1-861), anti-CHD4 (Abcam, ab70469), anti-RBBP4 (Invitrogen, PA5-95325), and anti-AP-2α 3B5 (Santa Cruz Biotechnology, SC-12726 X). ChIP-qPCR relative enrichment was calculated with ΔΔC_T relative to nonspecific immunoprecipitation at a negative control amplicon. Custom TaqMan probes and primers were designed underlying AP-2α ChIP-seq peaks and purchased from Integrated DNA Technologies with the sequences (5′-3′) listed in Table 1.

BioID was performed as described previously (23). Briefly, TFAP2A was ligated into the NheI and EcoRI sites within the MCS-BioID2-HA plasmid (Addgene, #74224). The TFAP2A stop codon was deleted and the biotin ligase fused to the carboxy-terminus of AP-2α with the Q5 Site Directed Mutagenesis Kit (New England Biolabs). Biotin-depleted medium was created by overnight incubation of culture medium with Dynabeads MyOne Streptavidin C1 magnetic beads (Invitrogen) with gentle rotation at +4°C and passed through a 0.2 μm vacuum filter. A375 was biotin starved by incubation for 72 hours with biotin-depleted medium, after which, it was transiently transfected with the AP-2α-BioID2 fusion expression plasmid with Lipofectamine 2000 and pulsed with 50 μmol/L biotin overnight. Cells were trypsinized, snap frozen, and sent to the Sanford Burnham Proteomics Facility for lysis, purification, and tandem mass spectrometry. Analysis was restricted to nuclear proteins (per Ingenuity Pathway Analysis) with a MS-MS count ≥ 5.

Coimmunoprecipitations (co-IP) were performed by isolating nuclear extracts from the A375 cell line using NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Fisher Scientific). Mouse IgG (Sigma, 15381), or anti-AP-2α 3B5 (Santa Cruz Biotechnology, SC-12726 X) were coupled to Dynabeads Protein G magnetic beads (Invitrogen) blocked in 5% BSA/PBS. The Western blot protocol listed above was used with anti-HDAC2 (Invitrogen, PA1-861), or anti-MTA2 (Santa Cruz Biotechnology, sc-55566) primary antibodies. The HDAC2 blot was performed with anti-rabbit light chain secondary antibody (Abcam, ab99697) and the MTA2 blot was performed with anti-mouse IgG secondary antibody (Cell Signaling Technology, 7076S).

Anti-AP-2α ChIP was first performed as described above. Beads were washed five times with RIPA buffer. Precipitated DNA was eluted by incubating beads with 10 mmol/L DTT for 45 minutes at 37°C, vortexing every 5 minutes. DTT was diluted 1:20 in Farnham Lysis Buffer and a second immunoprecipitation was performed with anti-HDAC2, anti-GATAD2B, or anti-CHD4 antibodies. qPCR was performed as described above with the EZH2 Promoter and Negative Control primer/probe sets.

The parental A375 cell line or TFAP2A^KO2 were cotransfected with either DsiRNA targeting both HDAC1 and HDAC2 (Integrated DNA Technologies, hs.Ri.HDAC1.13.2, hs.Ri.HDAC2.13.2) or a negative control DsiRNA (Integrated DNA Technologies, NC-1) using Lipofectamine RNAiMAX transfection reagent. RNA was harvested 96 hours after transfection and qRT-PCR was performed as described above.

For the cell line soft agar experiments, tazemetostat (MedChemExpress) was dissolved in DMSO for a stock concentration of 2 mmol/L and placed into the cell containing top layer of agarose and overlaying medium at the noted concentrations. For PDX soft agars, PDXs were pretreated with the noted concentrations of tazemetostat for 48 hours prior to seeding in addition to tazemetostat being placed into the cell containing top layer of agarose and overlaying medium. For cell-cycle analysis, cells were treated with 2 μmol/L tazemetostat for 96 hours prior to sample preparation.

A total of 10⁶ cells per mouse luciferized A375 cells were mixed with Matrigel (Corning) and subcutaneously injected into the flanks of NOD/SCID mice to create xenografts. Mice were randomly divided and began oral treatment with tazemetostat (Glpbio) or a DMSO (vehicle) control placed in their water starting 96 hours after xenograft. Tazemetostat was suspended (80 mg/mL) in DMSO for cryopreservation at −80°C, thawed prior to use, diluted in water to a final concentration of 4 mg/mL. Drug containing water was replaced daily for the 24-day treatment regimen. Consumption of drug-containing water was monitored so mice received approximately 800 mg/kg of tazemetostat per day in a nonrestrictive volume (≥ 10 mL available per mouse per day). Primary tumors were excised by surgical physicians blinded to the treatment regimen in survival surgeries 4 weeks after implantation. For the survival analysis, animals were either found deceased, or were euthanized according to IACUC protocol. Mice were euthanized for body condition score approaching 2 or less or flank tumor recurrences exceeding 2 cm in diameter.

IHC was performed on formalin-fixed, paraffin-embedded tissue sections by the University of Iowa Comparative Pathology Laboratory. IHC for H3K27Me3 was performed using anti-Histone H3K27Me3 (Millipore, 07-449).

Following surgical resection, tumors were disaggregated into single-cell suspensions. A total of 10⁵ cells per mouse were mixed with Matrigel (Corning) and subcutaneously injected into the flanks of naïve NOD/SCID mice to create xenografts. Mice were bioluminescently imaged 4 weeks after injection and subsequently euthanized. Livers from two vehicle and two tazemetostat secondary xenograft mice were disaggregated into single-cell suspensions for flow cytometry.

Following euthanasia, livers from 2 mice receiving vehicle-treated secondary xenografts or tazemetostat-treated secondary xenografts were disaggregated into single-cell suspensions. The suspension was analyzed on a Becton Dickinson Aria Fusion Cell Sorter examining GFP-positive metastatic melanoma cells residing in the livers. The percentages of metastatic cells were calculated using equal gating for both samples.

Statistics were calculated assuming equal variance. Bioluminescence was quantified using the Living Image software and experimental conditions were compared with two-tailed Student t tests. Soft agar assays and qPCRs were performed in triplicate and compared with two-tailed Student t tests. For RNA-seq, statistically significant differential expression was calculated by the Cuffdiff software. scRNA-seq differential expression was calculated by MAST in Seurat. Survival durations of tazemetostat- versus vehicle-treated mice were compared using Kaplan–Meier analysis. Primary tumor caliper measurements were compared with two-tailed Student t tests. All t test t values and degrees of freedom were calculated by GraphPad Prism. In all figures, *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001. P values less than 0.05 were considered significant. Error bars are shown as mean ± one SEM. Specific tests used, representations of P values, and sample sizes are also listed in figure legends.

Datasets generated in this article are available from NCBI GEO (accession numbers GSE162362 and GSE162363).

To determine whether AP-2α maintains the viability of metastasizing melanoma, we used CRISPR/Cas9 to disrupt TFAP2A in the A375 human metastatic melanoma cell line and generated multiple knockout clones devoid of AP-2α protein by Western blot analysis (Fig. 1A). We luciferized and orthotopically xenografted both the parental, AP-2α–positive cell line and a TFAP2A KO clone into the flanks of NOD/SCID mice. Four weeks after implantation, investigators blinded to the xenograft type excised flank tumors in survival surgeries to prevent primary tumor overgrowth and allow outgrowth of metastases. Pre-surgery caliper measurements of flank tumors confirmed the TFAP2A knockout clone retains tumorigenic and proliferative capability evidenced by no differences in primary tumor size comparing the parental cell line and a TFAP2A knockout clone (Fig. 1B). Bioluminescence imaging showed AP-2α–positive melanoma rapidly metastasized to the thoracic and abdominal cavities, but knockout of TFAP2A virtually eliminated metastases (Fig. 1C and D). Histologic staining (Fig. 1E) and ex vivo bioluminescence imaging (Fig. 1F) of isolated lungs and livers of xenografted mice confirmed in vivo bioluminescence corresponds to distant metastases.

Anchorage-independent colony formation in soft agar is a hallmark of metastatic cells (24). Colony formation was drastically reduced by loss of AP-2α in TFAP2A knockout clones and confirmed with knockdown of TFAP2A in a panel of melanoma cell lines and supports the findings of metastasis in xenografts (Fig. 1G; Supplementary Fig. S1A). Control clones, AP-2α–positive clones derived from the A375 cell line, retained anchorage-independent colony formation capability (Supplementary Fig. S1B) and confirmed the reduction of colony formation in soft agar results from the loss of AP-2α rather than clonal selection. The dependency on AP-2α for avoidance of anchorage-independent cell death and metastasis parallels the role of AP-2α in maintaining viability of migratory neural crest cells (4).

To elucidate the genes regulated by AP-2α that contribute to this prometastatic phenotype, we used RNA-seq to determine differential expression consistent between four CRISPR knockout clones of TFAP2A relative to the parental A375 cell line and A375 with shRNA-mediated knockdown of TFAP2A relative to a non-targeting control. RNA-seq revealed consistent and profound alterations of the transcriptome with loss of AP-2α (Fig. 2A). Pathway analysis indicated that the loss of AP-2α included inhibition of the E2F family of transcription factors, with four E2F pathways significantly inhibited with an activation z-score ≤ −2. The E2F transcription factors are cell-cycle effectors of c-MYC that frequently become hyperactivated during tumorigenesis and drive cell survival and proliferation. E2F1 is an extensively studied oncoprotein that is overexpressed in melanoma and drives colony formation, invasion, and metastasis (25, 26), making E2F1- and E2F-regulated targets likely candidates for AP-2α–driven metastasis. Furthermore, in apparent parallel to high AP-2α driving poor prognoses in node-positive melanoma (15), melanoma lymph node metastases frequently show E2F1 gene amplification (26).

Per pathway analysis, the AP-2α–regulated subset of E2F signaling comprises 129 genes exemplified by activation of the E2F targets EZH2 (27), RAD51 (28), and RRM2 (29), and repression of GM2A (30), LACTB (31), and EYA4 (Fig. 2A; ref. 31). The repression of the three E2F paralogs (E2F1, E2F2, and E2F8) after loss of AP-2α was observed in TFAP2A knockout clones by qRT-PCR and a panel of melanoma cell lines with knockdown of TFAP2A further confirmed consistent activation of these E2F paralogs by AP-2α (Fig. 2B). Consistent with the cell-cycle influence of this signature, propidium iodide staining revealed a defect in cell-cycle progression of TFAP2A knockout clones as demonstrated by a reduced proportion of cells present in G₂–M phases and an increased proportion of cells in S-phase (Fig. 2C; Supplementary Fig. S2).

Melanoma tumors and cell lines are known to be composed of a heterogeneous population, and it is unclear whether all cells within a melanoma are metastatic or if metastasis is restricted to a more aggressive subpopulation within tumors. If AP-2α–activated E2F signaling is restricted to a highly metastatic subset of cells, screening tumors for analogous subpopulations would be a powerful biomarker predicting metastatic potential. To determine the prevalence of E2F-activated cells, we performed scRNA-seq on the A375 cell line and identified four clusters representing a majority of cells that show a high degree of E2F activation as illustrated by the coexpression of E2F1 and EZH2 shown in Fig. 2D.

To directly test the influence of AP-2α on E2F activation, we performed scRNA-seq on A375 transduced with shNT versus shTFAP2A lentiviruses (https://singlecell.broadinstitute.org/single_cell/study/SCP1161/shrna#study-visualize) and a TFAP2A CRISPR knockout (https://singlecell.broadinstitute.org/single_cell/study/SCP1160/crispr). As shown in Fig. 3A, scRNA-seq revealed a high degree of heterogeneity within this cell line with the loss of AP-2α separating subpopulations as seen with UMAP cluster visualization. scRNA-seq revealed subpopulations of cells with a high degree of AP-2α–dependent E2F pathway activation typified by repression of EZH2^High and E2F1^High subpopulations with loss of TFAP2A (Fig. 3B, additional E2F pathway targets are shown in Supplementary Fig. S3A and differential expression with TFAP2A knockout is shown in Supplementary Fig. S3B and S3C). qRT-PCR confirmed that loss of TFAP2A reproducibly repressed pathway targets in multiple melanoma cell lines (Fig. 3C), and E2F pathway repression was consistently observed in TFAP2A knockout clones, but not AP-2α–positive control clones by Western blot analysis (Fig. 3D).

To determine whether AP-2α is regulating E2F pathway genes individually or through a single upstream mediator, we examined AP-2α genome occupancy by ChIP-seq in the M21 and A375 melanoma cell lines and a TFAP2A knockout negative control. This approach revealed consistent promoter occupancy by AP-2α near the transcription start sites of E2F pathway genes (Fig. 3E; Supplementary Fig. S4), strongly suggesting direct and individual gene regulation by AP-2α. Analysis of an A375 ATAC-seq dataset (32) revealed that E2F pathway promoter occupancy by AP-2α showed clear overlap with locations of accessible chromatin (Fig. 3E; Supplementary Fig. S4), raising the potential that AP-2α activates the E2F pathway genes by maintaining promoter accessibility.

During development AP-2α acts as a pioneer factor, a transcription factor whose DNA occupancy holds chromatin in a euchromatic, transcriptionally active state (33) with sites of occupancy coinciding with high levels of acetylated Histone H3 (34). Consistent with this role, AP-2α has been shown to prevent heterochromatin formation at sites of occupancy in human melanoma cell culture (35), though a mechanism remains elusive. To determine whether AP-2α facilitates transcription by aberrantly causing the epigenetic activation of E2F pathway promoters, we performed ChIP-qPCR by immunoprecipitating Histone H3K9/K14Ac and H3K27Ac from a TFAP2A KO clone or the parental A375 cell line. As shown in Fig. 4A, loss of AP-2α markedly reduced histone H3 acetylation at E2F pathway promoters corresponding to the observed transcriptional repression.

To uncover the biochemical mechanism underlying promoter hyperacetylation by AP-2α, we performed BioID (Supplementary Table S1), a proximity labelling approach to identify protein–protein interactions (23). BioID determined AP-2α interacts with nearly all members of the nucleosome remodeling and deacetylase (NuRD) multiprotein complex (Fig. 4B, top) and co-IP for HDAC2 and MTA2 confirmed the specificity of interactions (Fig. 4B, bottom). By analyzing ChIP-seq datasets for NuRD complex members, we observed co-occupancy of the deacetylation complex alongside AP-2α and acetylated histone H3 at E2F target promoters (Supplementary Fig. S5A), which we confirmed occurs in A375 human melanoma cells by ChIP-qPCR (Fig. 4C). ChIP-re-ChIP experiments further confirmed AP-2α and the NuRD complex co-occupy the same chromatin fragments at amplicons corresponding to E2F pathway promoters (Supplementary Fig. S5B). These data suggest that via co-occupancy, AP-2α prevents the deacetylation activity of the NuRD complex, driving hyperacetylation of promoter nucleosomes and facilitating transcription (Fig. 4D). Consistent with this model, siRNA-mediated knockdown of the NuRD complex histone deacetylases HDAC1 and HDAC2 restores expression of E2F pathway genes in a TFAP2A knockout clone, confirming a functional interaction between AP-2α and the NuRD complex at E2F pathway promoters (Supplementary Fig. S5C–S5E).

Within the E2F pathways and transcriptionally activated by AP-2α are EZH2 and EED, components of the polycomb repressive Ccmplex 2 (PRC2). EZH2 is the oncogenic and prometastatic methyltransferase within this complex that frequently becomes overexpressed during ocular and cutaneous melanoma transformation (36–38). PRC2 drives transcriptional silencing of tumor suppressor genes, such as p21/CDKN1A (37), via H3K27 trimethylation of associated nucleosomes (39), facilitating melanomagenesis and driving progression. High expression of EZH2, the catalytic enzyme within PRC2, has a strong positive association with metastasis in multiple cancers including melanoma (36, 39), breast cancer (40–43), bladder cancer (44), esophageal squamous cell carcinoma (45), and ovarian carcinoma (46). Because of this well-published prometastatic influence, we hypothesized that EZH2 is a principal effector within the E2F pathways that underlies AP-2α–driven metastasis.

To test this hypothesis, we treated AP-2α–positive melanoma and TFAP2A knockout clones with tazemetostat (EPZ-6438), an orally available, highly specific EZH2 inhibitor FDA approved for the treatment of metastatic epithelioid sarcoma and relapsed or refractory follicular lymphoma. Tazemetostat treatment mirrored the reduced colony formation following loss of TFAP2A, markedly restraining the ability of AP-2α–positive melanoma cell lines to form anchorage-independent colonies (Fig. 5A–D). All TFAP2A knockout clones were refractory to treatment (Fig. 5E; Supplementary Fig. S6A) but AP-2α–positive control A375 subclones retained tazemetostat responsiveness (Fig. 5F; Supplementary Fig. S6B), confirming AP-2α–mediated drug responsiveness. Paralleling the increase in cells residing in S-phase with knockout of TFAP2A (Fig. 2C) tazemetostat treatment increased the proportion of cells in S-phase, suggesting transcriptional activation of EZH2 contributes to the observed cell-cycle phenotype (Supplementary Fig. S6C and S6D). In addition, treatment of melanoma PDXs with tazemetostat inhibited colony formation in soft agar (Fig. 5G). AP-2α likewise occupied the EZH2 promoter and activated its transcription in these melanoma PDXs (Fig. 5H–K), suggesting this mechanism is common in clinical samples. These in vitro data support a model of EZH2 being a critical downstream target of AP-2α within the E2F pathways that drives metastasis.

To test the translational potential of EZH2 inhibitors as anti-metastatic adjuvant therapies, luciferized A375 cells were xenografted into the flanks of NOD/SCID mice. After allowing xenografts to establish, mice were divided and began daily oral treatment with tazemetostat or a vehicle control for 24 days. Bioluminescence imaging 4 weeks after xenograft showed substantial contralateral metastatic burden in vehicle-treated animals; however, tazemetostat-treated animals lacked this metastatic bioluminescence (Fig. 6A and B). Investigators blinded to the treatment regimen excised primary tumors in survival surgeries and withdrew treatment to allow outgrowth of micrometastases and monitor survival influence. Corresponding to the decreased metastatic bioluminescence, tazemetostat treatment reduced macroscopic abdominal metastases (Fig. 6C) and improved median survival by 52.6%, from 39 to 59.5 days (P = 0.0011; Fig. 6D). There were no differences in tumor-free survival (Supplementary Fig. S7A) or primary tumor caliper measurements (Supplementary Fig. S7B) comparing vehicle- and tazemetostat-treated mice, suggesting tazemetostat targets the metastatic phenotype rather than simply preventing primary tumor outgrowth. The observed slight reduction in flank luminescence in tazemetostat treated animals (Fig. 6A) resulted from decreased flank tumor peritoneal infiltration rather than an effect on primary tumor burden, raising the likelihood that tazemetostat inhibited the vertical growth phenotype. Furthermore, reduced histone H3K27 trimethylation in tazemetostat-treated primary tumors confirms treatment efficacy (Fig. 6E).

EZH2 inhibits differentiation during epidermal development (47), raising the possibility that tazemetostat could drive tumors into a well-differentiated state that would produce durable antimetastatic effects. To test this hypothesis, we generated single-cell suspensions from primary tumors of mice treated with vehicle or tazemetostat in Fig. 6A. We flank xenografted equal numbers of cells into naïve animals to generate secondary xenografts, again observing no alterations in tumor-free survival (Supplementary Fig. S7C) or primary tumor size (Supplementary Fig. S7D), suggesting no effects on primary tumor outgrowth. However, bioluminescence imaging of these naïve mice with secondary xenografts 4 weeks after implantation showed substantially reduced thoracic and abdominal metastases in the mice receiving xenografts from tazemetostat-treated animals (Fig. 6F and G). As the luciferized tumor cells bicistronically express GFP, we disaggregated livers from mice receiving secondary xenografts to perform flow cytometry. This approach verified a greater than 6-fold reduction of GFP-positive melanoma cells that metastasized to the livers of mice that received xenografts from tazemetostat-treated tumors (Fig. 6H and I). These data confirm EZH2 inhibition with tazemetostat produces durable anti-metastatic effects on melanoma tumors that persist long-term after treatment withdrawal.

We identified subpopulations of cells within the A375 cell line showing a high degree of E2F pathway activation (Fig. 2D). To visualize whether cellular subpopulations highly expressing Tfap2a and Ezh2 (Tfap2a^High/Ezh2^High) are unique to transformed cells, we analyzed scRNA-seq datasets examining the transcriptomes of untransformed melanocyte stem cells and melanomas, both derived from Tyr-CreER:Braf ^CA/+;Pten^fl/fl transgenic mice (48). These mice contain a Cre-inducible Braf ^V600E mutation and conditional knockout of the tumor suppressor Pten causing them to develop metastatic melanomas following topical administration of tamoxifen (49). We observed a dramatic expansion of melanoma subpopulations highly coexpressing Tfap2a, Ezh2, and E2f1 in these datasets (Fig. 7A–D). To test whether this pattern of coexpression of Tfap2a and E2F pathway targets includes AP-2α promoter occupancy that parallels our observations in human melanoma, we performed ChIP-seq in the TKLP mouse melanoma cell line. We observed AP-2α occupancy at E2F pathway promoters in this mouse melanoma model, suggesting the mechanism of direct E2F pathway activation by AP-2α is conserved in mice (Fig. 7E and F; expression of additional E2F pathway targets shown in Fig. 7G; Supplementary Fig. S8A; AP-2α occupancy in Supplementary Fig. S8B). The occupancy and coexpression of Tfap2a and Ezh2 directly parallels our observations in human melanoma (Supplementary Fig. S4). The expansion of Ezh2^High clusters in the melanoma scRNA-seq library is consistent with past publications showing significant overexpression of EZH2 in melanoma cells relative to healthy tissues. In stark contrast to traditional chemotherapies, this suggests EZH2 inhibition should elicit strong specific effects on melanoma metastasis while having limited effects on healthy tissues.

Like MITF, high expression of the transcription factor AP-2α maintains melanomas in a well-differentiated state. Paradoxically, loss of MITF drives melanoma cells to senesce, which led to the designation of MITF as a melanoma lineage addiction gene (50). We show loss of AP-2α is sufficient to nearly eliminate melanoma metastasis and anchorage-independent colony formation; hence, we conclude melanoma similarly relies on AP-2α for disease progression. This mechanistic parallel is consistent with reports of MITF and AP-2α exerting cooperative influence over the melanocytic lineage during development (5). It appears likely that AP-2α is sufficient to lower the threshold for survival to allow metastasizing cells to escape anchorage-independent cell death and form distant colonies. This observation mirrors the developmental role of AP-2α maintaining the viability of migrating neural crest cells, the progenitor tissue of melanocytes, and implies melanomas necessarily invoke this developmental program during metastasis.

Though AP-2α has been described as a pioneer factor (33), a transcription factor capable of holding chromatin in an accessible state, relatively little has been concluded surrounding a biochemical mechanism. However, genomic sites of occupancy by AP-2α are known to coincide with high levels of acetylated Histone H3 (34), implying a role of AP-2α in maintaining this marker of euchromatin. Our data led us to conclude that AP-2α inhibits the NuRD complex, leading to activation of genes in the E2F pathways and informing a potential mechanism for AP-2α–mediated epigenetic activation. This E2F pathway regulation is consistent with past literature showing AP-2α co-occupies target promoters alongside E2F1 (51). Though an interaction with the NuRD complex appears contributory to AP-2α–mediated transcriptional activation of E2F pathway genes, we cannot exclude the involvement of other AP-2α–interactive epigenetic modifiers (such as DNMT1 or HAT1; Fig. 4B), additional co-activators (such as STAT3; Supplementary Table S1), or that AP-2α could directly recruit the RNA polymerase II preinitiation complex. We speculate that AP-2α inhibits the deacetylation activities of the NuRD complex, though subsequent experiments will be necessary to determine whether this involves enzymatic or steric inhibition.

Within the E2F pathways and transcriptionally activated by AP-2α is EZH2, encoding a potently oncogenic and prometastatic histone H3K27 methyltransferase. EZH2 is active in embryonic stem and progenitor cells where it represses differentiation and promotes self-renewal (52). This normal physiologic role is coopted to drive an EMT-like transcriptional program in melanoma (53) promoting anchorage-independent colony formation (37), a marker of metastatic capability. In mice, Ezh2 is highly expressed in basal epidermal stem cells in developing embryos and wanes shortly after birth (47). Ezh2 expression decreases during cellular specification and mice with Ezh2-deficient basal epidermal cells show substantially accelerated differentiation (47).

Paralleling its role inhibiting differentiation of epidermal stem cells, EZH2 drives a dedifferentiation program in melanoma involving repression of MITF. This dedifferentiation program confers motility and invasiveness to cells (53, 54) and raises the likelihood of enhanced metastatic ability resulting from EZH2 overexpression. Indeed, Zingg and colleagues has recently shown conditional knockout of Ezh2 or treatment with the preclinical EZH2 inhibitor GSK503 inhibits metastasis of mouse melanomas (39). We show AP-2α activates transcription of this prometastatic oncoprotein, regulation that underlies AP-2α–driven metastasis.

Using the FDA-approved small molecule inhibitor tazemetostat (EPZ-6438), we determined EZH2 inhibition can halt the ability to form colonies in soft agar, an in vitro marker of metastatic ability, of multiple human melanoma cell lines and PDXs (Fig. 5). Currently, there are no targeted therapies approved for patients with tumors lacking BRAF mutations. As tazemetostat showed efficacy on a PDX (MB4667) that harbors a NRAS^Q61R-activating mutation, it appears EZH2 inhibition could be clinically beneficial for this population. Furthermore, tazemetostat has no effect on TFAP2A KO clones with minimal EZH2 expression, confirming its specificity. In vivo, tazemetostat shows promise as an antimetastatic adjuvant therapy by reducing the metastasis of xenograft bearing mice. Interestingly, the antimetastatic effect is heritable, as treatment-naïve mice receiving secondary xenografts from tazemetostat-treated tumors show a greater than 6-fold reduction in metastatic burden (Fig. 6).

Independent investigators have confirmed the antimetastatic potential of tazemetostat through studies in multiple cancer types. In melanoma, tazemetostat prevents in vitro spheroid formation, migration, and invasion (55), all hallmarks of metastatic capability. Furthermore, tazemetostat elicits differentiation of thyroid cancer cells (56) and inhibits growth of colorectal cancer tumors (57). In addition to inhibiting these prometastatic phenotypes, tazemetostat shows favorable pharmacokinetics to support its clinical use with limited adverse effects. Tazemetostat is efficacious at nanomolar concentrations in vitro and shows high specificity, showing 36-fold higher efficacy on EZH2 than EZH1 (58). Tazemetostat is quickly metabolized by the cytochrome P450 pathway (estimated elimination half-life of 3.1 hours; ref. 58), suggesting clinical adverse responses could be rapidly mitigated by drug withdrawal. Tazemetostat pharmacokinetics are unaltered by renal impairment, mild hepatic impairment, body weight, race, or sex (58) signifying its administration could be broadly effective and carry low risk to patients.

Multinational clinical trials testing the efficacy of tazemetostat have shown promising results in patients with cancer. A recent phase II trial in patients with sinonasal carcinoma and synovial sarcoma showed 41% of patients receiving tazemetostat achieved stable disease (clinical trial NCT02601950, preliminary data available as an abstract). NCT02601950 likewise showed promising and durable responses in patients with locally advanced or metastatic epithelioid sarcoma showing a disease control rate of 26% (defined as an objective confirmed response to treatment or stable disease lasting longer than 32 weeks). This response proved durable and carried a median duration of response not yet reached (ranging from 7.1 to 103.0+ weeks) confirming a lack of rapid acquired resistance. These clinical trials have shown promising effects on locally advanced and metastatic cancers that are traditionally chemoresistant and led to tazemetostat being awarded orphan drug status for treating epithelioid sarcoma. Our findings show AP-2α facilitates metastasis and induces tazemetostat responsiveness in advanced melanomas through transcriptional activation of EZH2. While immunotherapy and BRAF^V600E/MEK inhibitors are likely to continue to be used as first-line therapies, tazemetostat is a drug with immense potential to repurpose as an anti-metastatic therapy for melanoma.

R.J. Weigel reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

J.R. White: Conceptualization, resources, formal analysis, supervision, funding acquisition, validation, investigation, visualization, writing–original draft, project administration, writing–review and editing. D.T. Thompson: Conceptualization, formal analysis, supervision, validation, investigation, visualization, writing–original draft, writing–review and editing. K.E. Koch: Conceptualization, formal analysis, validation, investigation, writing–review and editing. B.S. Kiriazov: Conceptualization, validation, investigation, writing–review and editing. A.C. Beck: Conceptualization, investigation, writing–review and editing. D.M. van der Heide: Conceptualization, investigation, writing–review and editing. B.G. Grimm: Investigation, writing–review and editing. M.V. Kulak: Conceptualization, software, formal analysis, validation, investigation, writing–review and editing. R.J. Weigel: Conceptualization, resources, software, formal analysis, supervision, funding acquisition, validation, investigation, writing–original draft, project administration, writing–review and editing.

This work was supported by the NIH grants R01CA183702 (PI: R.J. Weigel) and T32CA148062 (PI: R.J. Weigel). D.T. Thompson, K.E. Koch, B.S. Kiriazov, A.C. Beck, and D.M. van der Heide were supported by the NIH grant T32CA148062. The authors thank Robert Cornell and Colin Kenny for their contribution to data interpretation and study design. The data presented herein were obtained at the Flow Cytometry Facility, which is a Carver College of Medicine/Holden Comprehensive Cancer Center core research facility at the University of Iowa. The facility is funded through user fees and the generous financial support of the Carver College of Medicine, Holden Comprehensive Cancer Center, and Iowa City Veteran's Administration Medical Center. Research reported in this article was supported by the NCI of the NIH under award number P30CA086862. This research was supported with computational resources from the University of Iowa High Performance Computing Facility and Galaxy. The TKLP cell line was a gift from Ned Sharpless facilitated by Stephen Pedroza (University of North Carolina at Chapel Hill). Melanoma PDX cell lines were provided by The Center for Rare Melanomas Biorepository at The University of Colorado. The M21 cell line was a gift from the laboratory of Robert Cornell (University of Iowa). pCas9D10A_GFP was a gift from Kiran Musunuru (Addgene plasmid, #44720; http://n2t.net/addgene:44720 RRID:Addgene_44720). MCS-BioID2-HA was a gift from Kyle Roux (Addgene plasmid #74224; http://n2t.net/addgene:74224; RRID:Addgene_74224).

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.