BRD4 Regulates Transcription Factor ΔNp63α to Drive a Cancer Stem Cell Phenotype in Squamous Cell Carcinomas

Squamous cell carcinomas (SCC) are among the most frequently occurring solid cancers in humans, and develop in response to mutagens (1). SCC is classified by the location in which they appear, frequently occurring in the skin, head and neck, esophagus, lung, and cervix, as well as more rarely in the thyroid, prostate, bladder, and pancreas (2). Treatment can include surgical removal of the primary tumor, as well as radiation and chemotherapy (3). Recently, targeted therapies against EGFR have been used with modest effect (3). However, many patients experience local disease recurrence and metastatic disease following treatment (1). Increasing evidence suggests that cancer stem cells (CSC) are to blame for recurrence, drug resistance, and metastasis in multiple cancer types (4). An important goal in this context is to identify key proteins that are essential for CSC survival that can be targeted therapeutically.

Chromatin modifying proteins have been under intense focus for their potential as therapeutic targets using small-molecule inhibitors. One particular family of chromatin modifying proteins under scrutiny is the Bromodomain and extraterminal (BET) domain family of proteins (5). BET proteins modulate genome processes by binding to acetylated lysine residues of both histone and nonhistone proteins, including transcription factors (5). These proteins function as key regulators of transcriptional control in development and differentiation. Additionally, BET proteins are frequently critical to the deregulated transcription characteristic of transformed cells, making them ideal targets for small-molecule intervention (6).

ΔNp63α is an important member of the p53 family of transcription factors that regulates epithelial cell differentiation, stem cell status, and cell fate (7, 8). ΔNp63α is a key controller of epithelial differentiation and maintains the undifferentiated state of the basal layer of epithelial tissues (9). The function of p63 in epithelia was first observed in p63-deficient mice, which die shortly after birth due to defects in epidermal barrier formation, revealing that ΔNp63α is essential for epidermal morphogenesis (8). Furthermore, ΔNp63α overexpression is frequently seen in human cancers, including SCCs (10–12). Overexpression of ΔNp63α is generally associated with poor prognosis (13). While its function and importance has been clearly demonstrated, knowledge of the mechanisms that control ΔNp63α expression is limited.

In this study, we characterize a subpopulation of SCC cells with properties of CSCs from varying tissue types, including skin, pharynx, and tongue. We define this subpopulation in 3 separate cell lines and show that these CSC-like cells display elevated levels of canonical and epithelial stem cell markers. Furthermore, these cells are endowed with the features of enhanced migration, invasion, and tumor growth. We show that the chromatin-modifying proteins BRD4 and EZH2 are elevated in this aggressive subpopulation of cells. These chromatin-modifying proteins work together to regulate the expression of a critical transcription factor, ΔNp63α, which is not only required for spheroid formation, invasion, migration, and tumor growth, but is elevated in patient-derived SCC. We propose that targeting this network of CSC survival proteins offers a therapeutic strategy for preventing metastasis and recurrence.

DMEM (10–013-CV), Trypsin (25–054-CI), and 6-well Ultra low attachment plates (3471) were purchased from Corning. DMEM F-12 (11320–033) was purchased from Thermo Fisher. Cell lysis buffer (9803), and antibodies specific for C-MYC (5605), BRD4 (13440), STAT3 (9139), and STAT3-P (9145) were purchased from Cell Signaling Technology. Anti-EZH2 (612667) was purchased from BD Transduction Laboratories. Anti-p63 (SC-8431) was purchased from Santa Cruz. Peroxidase-conjugated anti-mouse IgG and anti-rabbit IgG were obtained from GE Healthcare. Matrigel (354234) and BD Biocoat cell inserts (353097) were from BD Biosciences. DAPI (D9542) and Anti–β-actin (A5441) were purchased from Sigma. EF.STAT3C.Ubc.GFP was a gift from Linzhao Cheng (Addgene plasmid # 24983; http://n2t.net/addgene:24983; RRID:Addgene_24983; ref. 14). DeltaNp63alpha-FLAG was a gift from David Sidransky (Addgene plasmid # 26979; http://n2t.net/addgene:26979; RRID:Addgene_26979; ref. 15). pcDNA3-cmyc was a gift from Wafik El-Deiry (Addgene plasmid # 16011; http://n2t.net/addgene:16011; RRID:Addgene_16011; ref. 16).

Human HSC-5 cells were obtained from Sekisui Xenotech, LLC. Human FaDu and Cal-33 cell lines were a generous gift from Leif Ellison. HSC-5, FaDu, and Cal-33 cells were cultured in DMEM supplemented with 10% FBS. All cell lines tested negative for Mycoplasma contamination using MycoAlert (LT07–218) from Lonza.

Lentivirus was produced using 293T cells maintained in DMEM with 1 mmol/L L-glutamine, 1 mmol/L sodium pyruvate, and 10% FBS. 293T cells were harvested and plated in 100-mm dishes at 5% confluence for 24 hours prior to transfection. Media were removed and plates were washed with Hank balanced Salt Solution before serum-free media were added containing 7.5 μg pPax2, 5 μg VSVG, and 5 μg single-guide RNA (sgRNA) and polyethyleneimine (PEI) for cotransfection. After 8 hours, 10% FCS was added, and 48 hours after transfection, the medium was collected, centrifuged for 15 minutes at 1,500 rpm, sterile filtered (22 μm), and stored at −80°C in aliquots.

Wild-type cancer cells in monolayer culture were infected with 1 mL of lentivirus-containing serum-free media with PEI for 5 hours. After 5 hours, media was removed, and serum-containing media was added. After 24 hours, fresh media was added containing 300 μg/mL G418. CRISPR-mediated depletion was assessed via Western blot after 2 weeks of selection with G418.

Monolayer cultures were harvested with trypsin and gently pipetted to form a single-cell suspension. Trypsin was inactivated by addition of serum-containing medium and cells were collected by centrifugation at 2,000 rpm for 5 minutes. Cell pellets were resuspended in spheroid medium consisting of DMEM/F12 (1:1) containing 2% B27 serum-free supplement (17504–044, Invitrogen), 20 ng/mL EGF (AF-100–15, PeproTech), and plated at 4 × 10⁴ cells per 9.5 cm² well in 6-well ultralow attachment plates (3471, Corning; #3471).

For immunoblot analyses, equivalent amounts of protein were electrophoresed on denaturing and reducing 10% polyacrylamide gels and transferred to nitrocellulose membrane. Membranes were blocked in 5% nonfat dry milk and then incubated with the appropriate primary (1:1000) and secondary antibody (1:5000). Secondary antibody binding was visualized using chemiluminescence detection technology.

Matrigel was diluted in 0.01 Tris-HCL/0.7% NaCl, filter sterilized and 0.1 mL was used to coat individual BD BioCoat inserts (Millicell-PCF, 0.4 mm, 12 mm, PIHP01250). Cells (2.5 × 10⁴) were plated in 100 μl growth medium supplemented with 1% FBS into the upper chambers of the inserts. The lower chambers contained growth medium supplemented with 10% FBS. After invasion, membranes were harvested and the surface of upper membranes were rinsed with PBS to remove unattached cells. Membranes were fixed in 4% paraformaldehyde, stained with 1 mg/mL DAPI, and the underside of the membrane was photographed using an inverted fluorescent microscope and cells were quantitated.

Cells (2 × 10⁶) were plated in 10-cm dishes and grown as monolayer cultures in spheroid medium until confluent. A 10 mL pipette tip was used to prepare areas void of cells and the dishes were washed to remove the dislodged cells. Images were collected at 0 to 18 h after the scratches were made using a 10× objective, and the width of the openings was measured as a function of time as an index of cell migration potential.

Total RNA was isolated with the RNeasy Mini Kit (Qiagen) and reverse transcribed using the Superscript III First-Strand Synthesis System (Invitrogen). RNA (1 mg) was used for cDNA preparation. The Power SYBR^TM GREEN PCR Master Mix (Thermo Fisher) was used to quantitate expression levels according to manufactures protocol. Signals were normalized to the levels of cyclophilin A mRNA to determine expression.

Animal experiments were approved by the Institutional Animal Care and Use Committee. Spheroid-derived cancer cells were dispersed with trypsin to produce single-cell suspensions, and cells were resuspended in 100 μl of spheroid medium and then mixed with Matrigel at a 1:1 ratio. The mixture, containing 1 × 10⁶ cells, was injected subcutaneously into the flanks of nude mice using 26.5-gauge needles attached to 1 cc syringes. Cells, and syringes were kept on ice throughout the procedure to prevent Matrigel from solidifying. Four mice were used per data point with 2 tumors per mouse. For drug treatments, drugs were administered Monday/Wednesday/Friday via intraperitoneal injection.

The tissue array of SCC samples was obtained from US Biomax (HN803f). IHC slides were stained in Discovery Ultra automatic IHC stainer (Roche) following recommended protocols. Briefly, after deparaffinization and rehydration, slides were subjected to antigen retrieval (Benchmark Ultra CC1, Roche) at 96°C for 1 hour; primary Ab incubation was performed at 37°C for 1 hour; and Discovery multimer detection system (Discovery OmniMap HRP, Discovery DAB; Roche) was used to detect and amplify signals.

All data are presented as mean values ± SEM of at least three biological replicates unless otherwise indicated. Two-tailed Student t test or ANOVA (one-way or two-way) were used to determine effects of treatments. P < 0.05 was considered significant.

Growth as nonattached multicellular spheroids can be used to enrich for CSCs with enhanced tumor-forming potential (17–20). Utilizing this method, we set out to compare spheroid-derived human SCC cells to their monolayer counterparts (Fig. 1A; Supplementary Fig. S1A and S1B). We first examined whether cells grown as spheroids had elevated expression of epithelial and canonical stem cell markers. Monolayer- and spheroid-derived cells were analyzed via immunoblot. We found that a host of stem cell markers, including canonical stem cell markers OCT4 and SOX2 (21), as well as epithelial stem cell markers BRD4, EZH2, the mark of EZH2 activity (H3K27me3), and ΔNp63α (22–24) were increased under spheroid conditions compared with monolayer growth (Fig. 1B; Supplementary Fig. S1A and S1B). Because expression of stem cell markers had previously been associated with invasion and metastasis (25), we examined whether spheroid-derived cells displayed enhanced invasion and migration compared with their monolayer counterparts. This revealed that cells grown as spheroids invaded through Matrigel-coated membranes at higher rates than monolayer-derived cells, with more cells having invaded to the high serum-containing underside of the membranes after 24 hours (Fig. 1C; Supplementary Fig. S1C and S1D). In line with this result, spheroid-derived cells migrated into scratch wounds faster than monolayer-derived cells (Fig. 1D; Supplementary Fig. S1E). Because CSCs are associated with the ability to seed new tumors, we assessed whether these cells were more aggressive in vivo. Therefore, we injected 1 × 10⁶ monolayer- or spheroid-derived cells subcutaneously into the flanks of nude mice and monitored tumor growth. Spheroid-derived cells formed markedly larger, more invasive tumors than monolayer-derived cells (Fig. 1E). Additionally, these tumors maintained elevated levels of both canonical and epithelial stem cell markers (Fig. 1F), suggesting this tumorigenic phenotype was driven by the CSC population. Furthermore, tumor-derived single-cell suspensions resulted in increased spheroid growth compared with cells derived from monolayer tumors (Fig. 1G), suggesting these cells were maintaining stem like properties in vivo. To assess whether these CSC markers were clinically relevant, we analyzed patient-derived tumor arrays of 80 normal epithelial and SCC samples (Fig. 1H), which showed that each of these markers were elevated in SCC compared with normal control tissues. Thus, SCC spheroids are enriched for CSC-like properties and enhanced expression of CSC markers is observed in human SCC.

The above studies suggest that a subpopulation of cells with stem cell–like characteristics is responsible for driving an aggressive phenotype in SCC. BRD4 is involved in stem cell regulation, and is frequently overexpressed in a number of cancer types (26–28). We found that BRD4 expression was highly elevated in spheroid cultures (see Fig. 1B), suggesting a role in mediating the CSC-like phenotype. To understand the necessity of BRD4 in this context, we studied how its deficiency impacted spheroid formation, migration, and invasion using CRISPR-mediated depletion. As shown in Fig. 2A and Supplementary Fig. S2A and S2B, BRD4-depleted cells formed fewer spheroids than their control counterparts. Furthermore, invasion through Matrigel (Fig. 2B; Supplementary Fig. S2C and S2D) and migration into a scratch wound (Fig. 2C) were also impaired in BRD4-depleted cells. Using the BRD4 small molecule inhibitor JQ1, we found that treatment of preformed spheroids with JQ1 reduced the number of spheroids, and led to the accumulation of cell debris (Fig. 2D). We next examined the impact of JQ1 treatment prior to spheroid formation, and found that JQ1 treatment at the time of seeding (prevention conditions) resulted in a roughly 50% decrease in the number of spheroids that formed over 9 days of growth (Fig. 2E; Supplementary Fig. S2E and S2F). We further assessed the effect of JQ1 on invasion and migration of spheroid-derived cells, revealing that JQ1 treatment slowed the rate of both of these CSC phenotypes as well (Fig. 2F and G; Supplementary Fig. S2G and S2H). To further validate that these effects were due to BRD4, we used a second BRD4 inhibitor, MS436. Indeed, we found that like JQ1 treatment, inhibition of BRD4 using MS436 impaired the same CSC phenotypes; it reduced spheroid growth, invasion, and migration (Fig. 2G–K; Supplementary Fig. S2I–S2L). An important goal was to assess whether BRD4 was a valid target in vivo. To determine this, we injected 2.5 × 10⁶ spheroid-derived cells subcutaneously into the flanks of nude mice and monitored tumor growth in control- and MS436-treated recipients. Tumors of MS436-treated mice were markedly smaller than controls (Fig. 2L). Additionally, single-cell suspensions of these tumors cultured under spheroid conditions indicated that MS436 treatment reduced spheroid number (Fig. 2M). These findings indicate that BRD4 is essential to SCC CSCs.

To elucidate the mechanism behind the BRD4-driven aggressive phenotype of SCC CSC enrichment, we screened for potential BRD4 targets, and identified numerous changes in expression of epithelial stem cell markers. BRD4 depletion or treatment with BRD4 inhibitors JQ1 and MS436 resulted in the reduction of several key stem cell markers at the protein level, including EZH2, H3K27me3, ΔNp63α, and C-MYC (Fig. 3A–D; Supplementary Fig. S2M–S2P). This strongly suggested that BRD4 promoted the undifferentiated state. BRD4 is associated with enhanced C-MYC expression, a known stem cell regulator (29, 30). Therefore, we first focused on C-MYC to determine if the compromised stem cell marker expression caused by BRD4 depletion was being driven by C-MYC in the context of SCC. Consistent with this idea, we found that inhibition of BRD4 with MS436 treatment reduced C-MYC mRNA levels (Fig. 3E). To further investigate the role of C-MYC in BRD4-driven stemness, we performed genetic rescue experiments in which C-MYC was overexpressed in the context of BRD4 inhibition (i.e., in the presence vs. absence of JQ1), and monitored the CSC phenotype. This revealed that whereas both spheroid formation and invasion were impaired upon JQ1 treatment, this effect was mitigated in the presence of forced C-MYC expression (Fig. 3F and G). We verified this finding using a second BRD4 inhibitor, MS436, revealing that C-MYC effectively rescued the impaired spheroid formation and invasion caused by inhibiting BRD4 with this compound as well (Fig. 3H and I; Supplementary Fig. S3A and S3B). In addition to rescuing the biological effects of BRD4 inhibition, we found that forced C-MYC expression also mitigated MS436-induced changes in epithelial stem-cell marker expression, with C-MYC effectively rescuing both EZH2 and ΔNp63α expression despite BRD4 being inhibited (Fig. 3J; Supplementary Fig. S3C and S3D). This evidence indicates that BRD4 is essential for CSC phenotypes and that it induces the epithelial stem cell markers EZH2 and ΔNp63α through C-MYC.

Understanding the mechanisms governing CSC signaling and the aggressive phenotype is a critical goal. To determine whether the compromise in EZH2 we found following BRD4 depletion was required for the CSC phenotype, we performed depletion and inhibitor studies for EZH2. Our finding that EZH2 expression was reduced when BRD4 was targeted prompted us to ask whether EZH2 loss had a similar effect as BRD4 loss in suppressing CSC phenotypes. Polycomb repressive complex 1 (PRC1) and polycomb repressive complex 2 (PRC2) suppress gene expression via covalent modification of select histones, leading to reduced expression of tumor suppressors (31, 32). EZH2 is a lysine methyltransferase and is the main catalytic subunit of the PRC2 complex. EZH2 catalyzes trimethylation of lysine 27 of histone 3 (H3K27me3), an important step in polycomb-mediated silencing of gene expression. We assessed the impact of EZH2 depletion on the CSC phenotype, and found that CRISPR-mediated EZH2 loss compromised spheroid formation, invasion, and migration (Fig. 4A–C; Supplementary Fig. S4A–S4D). Furthermore, these cellular outcomes were accompanied by reduced expression of stem cell markers (Fig. 4D). Whereas EZH2 depletion resulted in reduced ΔNp63α and compromised H3K27me3, no appreciable changes in levels of BRD4 or C-MYC were observed, indicating that EZH2 functioned downstream of BRD4 and C-MYC, but upstream of ΔNp63α. To further validate this signaling and determine whether the effect of C-MYC on ΔNp63α expression is dependent on EZH2, we forced expression of C-MYC in BRD4- or EZH2-deficient cells and monitored for ΔNp63α expression. Expression of C-MYC in BRD4-depleted cells rescues ΔNp63α protein expression, but not in EZH2 knockout cells (Fig. 4E; Supplementary Fig. S4E and S4F), suggesting EZH2 is critical to C-MYC–driven ΔNp63α. Having identified this BRD4–EZH2–ΔNp63α axis, we asked whether the methyltransferase activity of EZH2 was essential for the CSC phenotype. We found that inhibition of EZH2′s methyltransferase activity with GSK126—a selective, small-molecule inhibitor of EZH2 that competes for the substrate-binding site of EZH2 and thereby inhibits its methyltransferase activity—reduced spheroid formation, invasion, and migration (Fig. 4F–H; Supplementary Fig. 4G–S4J). Whereas GSK126 reduced H3K27me3 levels (hence indicating that EZH2 methyltransferase activity was being impaired) it did not alter overall levels of EZH2 (Fig. 4I). Importantly, GSK126 treatment effectively compromised ΔNp63α expression, indicating that EZH2-mediated ΔNp63α induction was dependent on the methyltransferase activity of EZH2. To assess whether EZH2 functioned to maintain ΔNp63α expression in vivo, we injected 2.5 × 10⁶ spheroid cells per each front flank of nude mice and treated animals with 25 mg/kg body weight GSK126. Treatment with GSK126 in vivo reduced overall tumor volume, while simultaneously reducing levels of H3K27me3 and ΔNp63α (Fig. 4J and K). We found that GSK126 targeted the cell population responsible for the aggressive phenotype, as growth of dissociated tumor cells as spheroids produced significantly fewer spheroids when mice had been treated with GSK126 (Fig. 4L). These findings indicate that the methyltransferase activity of EZH2 is required for inducing the ΔNp63α−driven CSC phenotype.

EZH2 methyltransferase activity is associated with transcriptional repression and reduced expression of target genes (33). The fact that we had found reduced ΔNp63α expression in response to EZH2 depletion or inhibition suggested the possibility that EZH2 was regulating ΔNp63α in a noncanonical fashion, even though its methyltransferase activity responsible for its canonical function of placing H3K27me3 was required. To investigate how EZH2 was regulating ΔNp63α, we screened for EZH2-interacting partners that have also been linked to ΔNp63α regulation. This led us to focus on STAT3. STAT3 activation by tyrosine phosphorylation promotes cell proliferation, angiogenesis, immune evasion, and metastasis (34–36). Additionally, STAT3 activity is elevated in CSCs in a number of cancer types (37). Furthermore, STAT3 is capable of binding to the promoter of ΔNp63α, and transcriptionally activating it (37). Therefore, we hypothesized that STAT3 was part of the noncanonical mechanism by which EZH2 regulated ΔNp63α. To explore this possibility, we asked whether EZH2 was required for STAT3 expression. Depletion (Fig. 5A) and inhibition (Fig. 5B; Supplementary Fig. S4K and S4L) studies showed that the phosphorylated active form of STAT3 (STAT3-P) was reduced in response to both types of EZH2 perturbation. To better understand how EZH2 was activating STAT3, we pulled down EZH2 using coimmunoprecipitation; this revealed that EZH2 and STAT3 interact (Fig. 5C). Because GSK126 specifically reduced STAT3-P, we utilized coimmunoprecipitation of methyl K and STAT3 in the presence or absence of GSK126. Whereas STAT3 was methylated under control conditions, inhibition of EZH2 by GSK126 reduced STAT3 methylation (Fig. 5D). To assess whether STAT3 activity was required for the CSC phenotype, we utilized a STAT3 inhibitor, STATTIC. STATTIC is a vinyl-sulfone compound that targets the STAT3–SH2 domain, preventing its activation by phosphorylation (38). We found that STATTIC effectively targeted the CSC phenotype; it reduced spheroid maintenance and formation, as well as invasion and migration (Supplementary Fig. S5A–S5H). Furthermore, these changes in the CSC phenotype appeared to be dependent on the ability of STAT3-P to regulate ΔNp63α expression, as treatment of preformed spheroids with STATTIC reduced ΔNp63α expression at both the transcript and protein levels (Supplementary Fig. S5I–S5L). To further investigate if STAT3 activity was sufficient for the EZH2-driven phenotype and ΔNp63α expression, we treated with STAT3C—a constitutively active form of STAT3—in the context of EZH2 inhibition with GSK126. This revealed that STAT3C rescued the compromised spheroid formation and invasion caused by interfering with the methyltransferase activity of EZH2 (Fig. 5E and F), and further shows that rescue of the CSC phenotype coincided with restoration of ΔNp63α expression (Fig. 5G). To further demonstrate that methylation of STAT3 by EZH2 was critical to driving the aggressive phenotype and ΔNp63α expression, we used CRISPR to deplete STAT3, then rescued with either wildtype STAT3 or STAT3 with the EZH2 methylation site mutated (K180A). Figure 5H,–J and Supplementary Figure S5L–S5Q show that STAT3 depletion reduces spheroid formation, invasion, and ΔNp63α expression that is rescued by wildtype STAT3, but not the methylation mutant STAT3 K180A. These findings demonstrate the importance of STAT3 methylation in driving the aggressive phenotype and ΔNp63α. Figure 5K and Supplementary Figure S5R show that when the EZH2 methylation site on STAT3 is mutated, EZH2 does not coimmunoprecipitate with STAT3. To determine if STAT3 was a potential therapeutic target to reduce ΔNp63α in vivo, tumor-bearing mice were treated with STATTIC. Tumors of STATTIC-treated mice were markedly smaller than controls (Fig. 5L), and had reduced ΔNp63α expression (Fig. 5M). Additionally, single-cell suspensions of these tumors cultured under spheroid conditions indicated that STATTIC treatment reduced spheroid number, suggesting that STATTIC was targeting the population of tumor cells with stem-like properties (Fig. 5N). These findings elucidate a novel mechanism whereby EZH2 uses its noncanonical function to activate STAT3 by methylating it, serving as an oncogenic switch linking a BRD4–ΔNp63α cascade that drives the CSC population.

ΔNp63α is an important member of the p53 family of transcription factors involved in epithelial cell differentiation, stem cell status and fate (39). ΔNp63α is a key bottleneck of epithelial differentiation that maintains the undifferentiated stem cell state of proliferating basal layer cells in a variety of epithelia (39). We reasoned that the BRD4/EZH2-regulated pathway was critical for the CSC phenotype because of its ability to regulate ΔNp63α expression. To test this idea, we created pan-p63–depleted cells lacking all p63 isoforms using gene editing, and found that the predominant p63 isoform in epithelial tissues and in SCC, ΔNp63α, was dramatically reduced (Fig. 6A). CSC phenotypes such as spheroid formation (Fig. 6B; Supplementary Fig. S6A and S6B), as well as invasion and migration (Fig. 6C and D) were dramatically reduced in pan-p63–depleted cells. To further validate ΔNp63α as the isoform responsible for the phenotype, we reexpressed ΔNp63α in pan-p63–depleted cells (Fig. 6E), and assessed its ability to rescue the effect of p63 loss. We found that whereas spheroid formation and invasion were severely compromised in cells deficient for all p63 isoforms, ΔNp63α expression alone was sufficient for restoring CSC phenotypes (Fig. 6F and G; Supplementary Fig. S6C and S6D). To determine if ΔNp63α was capable of driving the phenotype independent of BRD4 and EZH2, we reexpressed ΔNp63α into BRD4-deficient and EZH2-deficient cells (Fig. 6H). Importantly, expression of ΔNp63α was sufficient for rescuing spheroid formation and invasion despite the lack of BRD4 or EZH2 in these cells (Fig. 6I and J; Supplementary Fig. S6E and S6F). These findings indicate that the BRD4–EZH2–STAT3 oncogenic network culminates with ΔNp63α to drive CSCs in SCC.

The CSC model suggests that a fraction of cancer cells possess the ability to initiate and to sustain tumor growth, thereby contributing to tumor relapse (40). CSCs share many features with their normal stem cell counterparts, including their slow-cycling nature, and the ability to give rise to a hierarchy of daughter cells with varying degrees of differentiation (41). These CSCs undergo aberrant differentiation, creating cellular heterogeneity within the tumor (42). Furthermore, CSCs display enhanced antiapoptotic signaling, increased membrane transporter activity, and enhanced migratory, invasive, and metastatic potential (41). Identification and characterization of CSCs in various cancer types can lead to the development of therapeutic strategies that target the CSC population, thus preventing metastasis and recurrence.

An important strategy for targeting CSCs is identifying proteins that are critical for driving their aggressive phenotype. In order to do this, we screened a number of known epithelial and canonical stem cell markers in SCC CSCs compared with their bulk monolayer counterparts. We identified a subpopulation of cells (0.14%) that survive when selected under spheroid conditions. These spheroid-derived cells express elevated levels of a number of stem cell markers including BRD4, EZH2, ΔNp63α, SOX2, and OCT4. BRD4, EZH2, and ΔNp63α are expressed robustly in epithelial stem cells and are critical regulators of self-renewal and the undifferentiated state (9, 23, 26, 43). SOX2 and OCT4 are canonical stem cell markers that are overexpressed in some CSC types including prostate, breast, and epidermal SCC (44, 45). Overexpression of these critical stem cell markers in spheroid culture suggests that spheroids are enriched for cells with similar properties to that of embryonic stem cells and may be evidence that they are transformed. In addition to the elevated stem cell marker levels in spheroid culture, spheroid-derived cells behave more aggressively, migrating to close a scratch wound more rapidly, and invading through Matrigel at higher rates than the bulk monolayer counterparts. Spheroid-derived cells also form large and invasive tumors, further evidence of the aggressive phenotype of cells containing these CSC-like properties. We also demonstrate that stem cell marker expression is maintained in the tumors in vivo, indicating that the CSC population is enriched within the tumor. Furthermore, these markers are elevated in patient-derived SCC specimens compared with normal tissue, suggesting these targets are clinically relevant.

In these studies, we identify a BRD4-driven signaling cascade that mediates the ΔNp63α-dependent aggressive phenotype of CSCs in SCC. BRD4 is an important member of the BET domain family of proteins, characterized by two N-terminal bromodomains and an extraterminal (ET) domain. Previous studies have shown that BRD4 binds directly to acetylated lysine residues of histones as well as those of transcription factors, and can recruit transcriptional coregulators (46). Importantly, BRD4 is significantly overexpressed in a number of cancer types where it acts as an oncogene, including melanoma, colon, and bladder cancer (47–49).

Gene editing identified BRD4 as critical to the CSC phenotype, with BRD4 depletion impairing spheroid formation, migration, and invasion. To further explore BRD4 as a potential therapeutic target of SCC CSCs, we utilized JQ1 and MS436, two small-molecule inhibitors of BRD4. JQ1 competitively binds to the acetyl-lysine recognition site of BRD4's bromodomain, displacing BRD4 from acetylated chromatin and repressing transcription of target genes (50). MS436 inhibits BRD4 transcriptional activity and has a 100-fold higher affinity for the first bromodomain of BRD4 over the second (51). Treatment with either inhibitor results in the destruction of preformed spheroids, reduces the number of spheroids forming when treated at initiation of growth, and impairs invasion and migration. Importantly, tumor growth in vivo is greatly impaired by both of these BRD4 inhibitors. This data has important translational implications, as it demonstrates that pharmacologic inhibition of BRD4 can be used to target SCC CSCs.

Understanding how BRD4 regulates this phenotype was an important goal, and the changes in the CSC phenotype following perturbation of BRD4 prompted us to look at other stem cell markers elevated within the spheroid population. We show that depletion or inhibition of BRD4 causes a dramatic reduction of C-MYC expression at both the protein and the mRNA level. Additionally, other stem cell markers including EZH2 and ΔNp63α are reduced at the protein level following BRD4 depletion and inhibition. Forced expression of C-MYC rescues spheroid formation and invasion in the presence of BRD4 inhibitors, demonstrating that C-MYC is a critical mediator of the BRD4-driven aggressive CSC phenotype. Additionally, we show that EZH2 and ΔNp63α protein expression are restored by C-MYC as well. This led us to look at whether EZH2 was also an essential component of the signaling cascade regulating the aggressive phenotype.

EZH2 depletion experiments produce similar changes to the aggressive phenotype as BRD4 depletion, implicating EZH2 as a key component of the BRD4-mediated axis. GSK126 is a selective competitive inhibitor of S-adenosyl-methionine–dependent EZH2 methyltransferase activity (32). Treatment with GSK126 is associated with reduced catalytic activity as demonstrated by compromised H3K27me3 (32). An important finding is that GSK126 suppresses the aggressive phenotype, including spheroid formation and maintenance, invasion, migration, and tumor growth. EZH2-depleted cells demonstrate not only a reduction in spheroid formation, migration, and invasion, but reduced ΔNp63α expression. However, BRD4 and C-MYC protein levels remain unchanged, indicating that BRD4 and C-MYC are upstream of EZH2, and that EZH2 is upstream of ΔNp63α.

Immunoprecipitation studies show that EZH2 interacts with STAT3 in spheroids, and this interaction results in the methylation and subsequent activation (phosphorylation) of STAT3 via EZH2. These findings are consistent with previous reports showing that EZH2 can activate STAT3 in glioblastoma stem cells (52). This was further shown to be dependent on methylation of K180 on STAT3 by EZH2 in glioblastoma CSCs (52). Our studies show that STAT3 is not only important for the BRD4/EZH2-driven phenotype, but that it is a critical regulator of ΔNp63α mRNA and protein. Although more studies are required to determine the exact mechanism behind increased STAT3 activation following methylation, our findings highlight EZH2 and STAT3 as potential therapeutic vulnerabilities for SCC CSCs.

The observation that ΔNp63α is reduced in each of the depletion and inhibitor studies that produce a biological response prompted us to examine whether ΔNp63α is the key driver of this phenotype. We show that pan-p63 depletion severely impairs spheroid formation, invasion, and migration, and that these phenotypes are completely rescued by reexpression of the ΔNp63α isoform. Importantly, forced expression of ΔNp63α in BRD4- and EZH2-depleted cells restores the aggressive CSC phenotype as well. This demonstrates that BRD4 and EZH2 drive the CSC phenotype via regulation of ΔNp63α, and that ΔNp63α is the critical regulator of this phenotype.

The above studies indicate that ΔNp63α is a critical regulator of the aggressive phenotype of CSCs. Furthermore, two chromatin modifying proteins that are highly elevated in epithelial stem cells—BRD4 and EZH2—are essential for maintaining ΔNp63α expression. These findings highlight BRD4 and EZH2 as potential therapeutic targets in SCC to effectively reduce ΔNp63α and in turn, mitigate the CSC phenotype. This work has important therapeutic implications for preventing recurrence, drug resistance, and metastasis.

J.E. Wilkinson reports personal fees from Cold Spring Harbor during the conduct of the study. No disclosures were reported by the other authors.

M.L. Fisher: Conceptualization, funding acquisition, investigation, writing–original draft, writing–review and editing. S. Balinth: Investigation, writing–review and editing. Y. Hwangbo: Conceptualization, investigation. C. Wu: Conceptualization, investigation. C. Ballon: Investigation. J.E. Wilkinson: Data curation. G.L. Goldberg: Clinical relevance. A.A. Mills: Conceptualization, supervision, funding acquisition, writing–review and editing.

This work was supported by the Office of the Director, NIH through award numbers 5P30CA045508 (Cancer Center Support Grant), CA225134 (to M.L. Fisher), CA247400 (to S. Balinth), as well as R01CA190997 and R21OD018332 (to A.A. Mills). This project was also supported through the Cold Spring Harbor Laboratory and Northwell Health Affiliation.

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.