A Combination of BRD4 and HDAC3 Inhibitors Synergistically Suppresses Glioma Stem Cell Growth by Blocking GLI1/IL6/STAT3 Signaling Axis

The incidence of glioma, the most common malignant brain tumor, is on the rise (1). Glioblastoma multiforme (GBM) is the most aggressive form of malignant astrocytoma (2, 3). Less than 5% of patients diagnosed with gliomas survive despite the currents standard treatment including resection, radiotherapy, and chemotherapy (3, 4). Studies have also suggested that the residual therapy-resistant and slow-dividing malignant cells, known as glioma stem cells (GSC; refs. 5–7). There is progress to reinitiate tumor formation and promote invasiveness and recurrence. However, current therapeutic approaches fail to target and eliminate GSCs. Thus, there is a critical need for developing new and effective therapeutics for treating GSCs.

Acetylation and deacetylation of histone are critical determinants in the control of gene expression. The level of histone acetylation is dependent on the recruitment of histone acetyltransferases (HAT) and histone deacetylases (HDAC; refs. 8, 9). Many studies have shown that HDACs are generally considered suppressors of gene expression and HDAC inhibitors exert potent anticancer effects (8, 9). Among members of the HDAC family, HDAC3 plays a key role in cell-cycle regulation, DNA damage control, and genomic stability maintenance (10, 11). Meanwhile, our data showed that inhibition of HDAC3 could lead to suppress GSC proliferation using global epigenetic screening with a shRNA library capable of targeting 243 known regulators of chromatin remodeling. However, we have noted that HDAC3 inhibition promoted tyrosyl phosphorylation of STAT3 (pY⁷⁰⁵), which potentially limited the inhibitory effects of HDAC3i on GSC growth.

Some studies showed that the combination of pan-HDACi, SAHA, and a BRD4i, JQ1, augmented cell death and more potently than the individual drugs, and suppressed the growth of advanced pancreatic ductal adenocarcinoma (12). In addition, SAHA poorly inhibited breast cancer cell growth owing to increased STAT3 pY⁷⁰⁵ and owing to its ability to upregulate the expression of receptor for the leukemia-inhibitory factor (LIFR; ref. 13). BRD4 was essential for LIFR transcription and the downstream activation of STAT3, via JAK1 by LIFR. Thus, inhibition of HDAC in association with JAK1 or BRD4 inhibitors potentially affords therapeutic benefit for patients with breast cancer (13). In the current study, we found that a HDAC3-specific inhibitor, RGFP966, was unable to suppress growth due to an elevated activation of STAT3-dependent signaling in GBM. However, a BRD4 inhibitor, JQ1, strongly blocked STAT3 signaling. These observations prompted us that BRD4 inhibitor may enhance GSC sensitivity to HDAC3 inhibitor. Our study demonstrated that upon cotreatment of cells with a HDAC3i, RGFP966 (14), and a BRD4i, JQ1 (15), strongly blocked STAT3 signaling and promoted GSC growth suppression, indicating that combination inhibition of HDAC3 with BRD4 as potential therapies for glioma progression.

The human GSCs lines TS543 and TS576 were provided by Cameron Brenner, Holland, and Mellinghoff laboratories, Memorial Sloan-Kettering Cancer Center (New York, NY). Murine GSCs, CSC2078, CSC1534, and CSC1589 were isolated from hGFAP⁻Cre⁺ p53^L/^L Pten^L/⁺ mice as described previously (16). They were cultured in neurobasal media (Thermo Fisher Scientific, Inc.) supplemented with EGF (20 ng/mL) and basic FGF (bFGF; 10 ng/mL) or differentiation medium (neurobasal media with 1% FBS). BRD4i and HDAC3i (MedChemExpress) were dissolved in DMSO.

CSC2078, CSC1534, and CSC1589 were three independent GSC lines from hGFAP⁻Cre⁺ p53^L/^L Pten^L/⁺ mice, which displayed high self-renewal, partial differentiation, and strong tumor initiation capacities. A shRNA library was established and blocked the expression of 243 known chromatin regulators (17) in these cells and a screening score standard for the candidate genes was as followed: a negative control shRNA (shRluc.713) was set as “Score 4” and those with positive control (shRpa3.455) was set as “Score 1”. Notably, the presence of GFP fluorescence in the vector could enable distinguishing of shRNA-expressing cells from their surrounding noninfected control cells. If the size of GFP-positive colonies is smaller than the negative control, the knockdown of this gene can suppress glioma cell growth. So, according to the size and number of GFP⁺ clones, target genes whose inactivation by the cognate shRNA that scored in the range of “Score 1–2” in at least two of the three glioma lines used were found out.

RNAi was used for knocking down the expression of specific genes. The sequences of HDAC3 shRNA, STAT3 shRNA, and other siRNAs were shown in Tables 1 and 2. shRNAs capable of targeting STAT3 (18) and HDAC3 (17) were used as published previously. The siRNAs specific for BRD4, GLI1, and IL6 (Genepharma) were transfected using the reagent (Thermo Fisher Scientific, Inc.) following the manufacturer's protocol. Cells were collected after transfection for 48 to 72 hours.

Cell viability was assessed after various treatments using a Cell Counting Kit-8 (MedChemExpress) per the manufacturer's instructions. Cells were seeded in 96-well plates (5,000 cells/well) in 100 μL media. After allowing them to adhere to the plates overnight, various treatments were given.

Combination index (CI) scores were calculated as previously described using CompuSyn software, which was published by the ComboSyn, Inc. Cell proliferation data obtained in the presence of DMSO (i.e., vehicle) single drugs and their combination were used for calculating CI values for each combination point. Quantitatively, the following define the metrics used for drug impact on cells: additivity (CI = 1), synergy (CI < 1), and antagonism (CI > 1). The resulting values were used to construct a plot of CI values over a range of fractions affected. On the basis of the CI values that exhibited best synergy, we chose 500 nmol/L of HDAC3i and 100 nmol/L of BRD4i for treating CSC2078, CSC1589, and TS576 cells in all further experiments. The effective drug concentrations of HDAC3i and BRD4i for TS543 cells were 500 and 200 nmol/L, respectively.

Cells were seeded on 6-well plates after treatment at a concentration of 5 × 10⁴ cells/well and number of cells. Cell numbers were counted under a microscope at 24, 48, and 72 hours posttreatment with the indicated agents. Each condition had 3 replicates.

After a 0.6% soft agar (prepared in growth media) feeder layer was solidified, cells (2,500 cells/well) were plated in an upper layer consisting of 0.4% soft agar, and then treated with BRD4i (100 nmol/L) or HDAC3i (500 nmol/L). After 15 days, cells were stained with 0.5% crystal violet and photographed.

Self-renewal capacity of GSCs was measured by neural sphere formation assays. Cells (500/well) were seeded into 6-well with proliferation media. After treatment with the indicated agents for 10 days, the number of neural spheres formed in each condition was measured by counting of five random fields under IX71 inverted fluorescence microscope (Olympus).

Cells were collected and washed twice with cold PBS. They were suspended in Cell Cycle Detection Kit (BestBio), and the numbers of cells in each cycle phase were determined using a C6 flow cytometer (Becton Dickinson). Each condition had 3 replicates.

Cells (1 × 10⁵) were plated into 6-well dishes and harvested for analyses after various treatments. They were stained with a FITC-Annexin V using a commercially available kit (BD Biosciences) and the FITC⁺ cells were scored using a Becton Dickinson C6 flow cytometer. Each condition had 3 replicates.

Equal quantities of total cellular proteins were subjected to Western blot analyses as previously described earlier (19). Antibodies used were as follows: pSTAT3 (Tyr705; cat no. #9145), cleaved caspase-3 (cat no. #9664), and acetylated-lysine (cat no. #9814s) from Cell Signaling Technology; BCL-2 (cat no. 26593–1-AP), GLI1 (cat no. 66905–1-AP), and β-actin (cat no. 66009–1-Ig) were from Proteintech Group; STAT3 (cat no. sc-482), HDAC3 (cat no. sc-11417), and Mcl-1 (cat no. sc-12756) were from Santa Cruz Biotechnology and H3K27ac (cat no. ab4729) was from Abcam. Each condition had 3 replicates.

RNA was isolated using the TRIzol reagent (Thermo Fisher Scientific, Inc.) and was reverse transcribed using the Thermo Scientific RevertAid First Strand cDNA Synthesis Kit (cat no. #K1622). qPCR was performed using the SYBR Premix Ex TaqTM II (Tli RNaseH Plus; cat no. #RR820A, TaKaRa) by Bio-Rad CFX96 PCR machines (Bio-Rad) The 2^−ΔΔC_t method was used for quantifying specific transcripts. β-Actin expression was used as internal control for normalization. The primers used were shown in Table 3. Each condition had 5 replicates.

Cells grown on glass coverslips or tissue sections were treated with 4% paraformaldehyde for 30 minutes at room temperature and were then stained using a One Step TUNEL Apoptosis Assay Kit (Beyotime Biotechnology Inc.) per the manufacturer's instructions. FITC-labeled TUNEL-positive cells were observed and enumerated using the IX71 fluorescence microscope (Olympus), and images were captured. Each condition had 3 replicates.

Total RNA was isolated using the TRIzol Reagent (Thermo Fisher Scientific, Inc.). Sequencing libraries were generated using the TruSeq RNA Sample Preparation Kit (Illumina) and then were purified according to AMPure XP system (Beckman Coulter). Purified products were subjected to paired-end (PE) sequencing based on the Illumina HiSeq sequencing platform using next-generation sequencing (NGS). The raw data (Raw Data) were filtered out based on 200 bp cDNA fragments, and the resultant high-quality sequence data (Clean Data) was aligned to the reference genome of the species. On the basis of the alignment results, the expression level of each gene was calculated. On this basis, the samples were further analyzed for differential analysis, enrichment analysis, and cluster analysis on a Hiseq platform (Illumina, Inc.) by Shanghai Personal Biotechnology Co. Ltd. Sequence Read Archive (SRA)/NCBI: Coordinates have been deposited with accession code SUB7884448.

Chromatin immunoprecipitation (ChIP) assay was performed using the EZ-Magna ChIP A/G Chromatin Immunoprecipitation Kit (cat no. 17–10086, Millipore) according to the procedures provided by the manufacturer. Following antibodies were used for immunoprecipitation: IgG (cat no.17–10086, Millipore), H3K27ac (cat no. ab4729, Abcam), BRD4 (cat no. #13440, Cell Signaling Technology), and GLI1 (cat no. NB600–600, Novus Biologicals). Quantitative PCR analyses of the DNA fragments recovered from the immunoprecipitated products were performed using a Bio-Rad CFX96 real-time PCR machine (Bio-Rad), using primers that encompass GLI1 promoter. The primers of GLI1 transcriptional start site and IL6 promoter were listed in Table 4. Each condition had 3 replicates.

Cells were fixed in 4% paraformaldehyde and then permeabilized with 0.3% Triton X-100 for 15 minutes. After blocking with 5% BSA, they were incubated overnight with select primary antibodies at 4°C in dark. After washing the sections with PBS, samples were stained with diluted green Alexa Fluor 488–conjugated secondary antibody (AS035; ABclonal Biotech Co., Ltd.) and diluted red Alexa Fluor 647–conjugated secondary antibody for 30 minutes in the dark at room temperature. Finally, cells were treated with DAPI for 2 minutes, and images were acquired using a BX53 fluorescence microscope (Olympus).

Mouse tumor xenografts were immersed in 4% paraformaldehyde for at least 4 hours at room temperature, followed by dehydration, dipping in wax, and paraffin embedding for the preparation of paraffin sections. The sections were incubated with 2% BSA for 30 minutes first and then with the select primary antibodies and subsequently secondary antibodies. Images were captured using an Olympus BX53 microscope fitted with a digital camera.

ELISA was used to measure HDAC3 activity and IL6 level (MEIMIAN) in cell culture supernatants following the manufacturer's protocol. ELISA data were collected using a microplate reader (FLUO star Omega) at absorbance 450 nm. The assay sensitivity ranges of ELISA for HDAC3 activity and IL6 were 1.6 to 65U/L and 10 to 240 pg/mL, respectively.

BALB/c nude mice were purchased from Beijing Vital River Laboratory Animal Technology. Mice were housed in an air-conditioned room at 22°C ± 2°C with ample supply of food and water. All animal experiments were conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals, with the approval of the Scientific Investigation Board of the College of Basic Medicine, Jilin University (Changchun, China). Mice were subcutaneously transplanted with TS543 cells (1 × 10⁷/mouse) and tumors were allowed to establish. Tumor-bearing mice were randomized into four groups (n = 6 each) and treatment was initiated when average tumor size reached 4 mm³. HDAC3i (25 mg/kg once every other day) via subcutaneous injection and BRD4i (50 mg/kg once daily) via intraperitoneal injection were given for the number of days indicated. Both drugs were given concurrently if a drug combination was used. The size of tumor was measured every 3 days and tumor volume was calculated with the following formula: V = (length) × (width)²/2. After 21 days, mice were euthanized, and tumor tissues were collected and prepared for Western blot or IHC staining.

All analyses were performed using the Prism GraphPad 6.00 software. One-way ANOVA with a Newman–Keuls multiple comparison test was performed to discern the significance of the difference between groups. The comparisons of two sets of data were analyzed using Student t test. A P value <0.05 was considered statistically significant. Data are presented as mean ± SD in each case.

In an effort to identify the critical chromatin regulators required for stemness of CSCs, we employed RNAi with a shRNA library capable of knocking down 243 known genes and identified as HDAC3 as important factor required for self-renewal and proliferation of GSCs (Supplementary Fig. S1A and S1B). We found that inhibition of HDAC3 suppressed GSCs from hGFAP-Cre⁺ p53^L/L Pten^L/+ mice (16). To further verify the role of HDAC3 in GSCs growth, we treated CSC2078 cells with various doses of a HDAC3i, RGFP966 (up to 500 nmol/L). Cell viability analyses showed that HDAC3i did not strongly block cell viability. Its effects were marginal at 1,000 nmol/L, the highest dose tested in this study (Fig. 1A). The drug was active, because it was able to block HDAC3 activity as shown by an ELISA (Fig. 1B). The ineffectiveness of HDAC3i on CSC2078 cell growth at least did not due to a failure of HDAC3 inhibition. Puzzlingly, Western blot analyses showed that STAT3 phosphorylation was increased in a time- and dose-dependent manner in response to HDAC3i treatment. Therefore, we measured levels of antiapoptotic proteins such as Bcl-2 and Mcl-1, the downstream targets of STAT3. Indeed, Bcl-2 and Mcl-1 were also increased in the presence of HDAC3i (Fig. 1C). To determine the importance of STAT3 in this process, we knocked it down using a specific shRNA. Indeed, depletion of STAT3 significantly restored the growth-inhibitory effects of HDAC3 inhibitor (Fig. 1D). STAT3 shRNA itself only slightly elevated the proapoptotic activated caspase-3, which was further increased robustly after HDAC3i treatment (Fig. 1E). Consistent with this, in the presence of STAT3 shRNA, HDAC3i strongly inhibited cell proliferation (Fig. 1F) and self-renewal capacity of GSCs (Fig. 1G). HDAC3i was incapable of blocking growth strongly and the study confirmed that JQ1 can effectively inhibit the STAT3 activation and synergistically enhance the sensitivity of breast cancer to HDAC inhibitor SAHA to achieve a better therapeutic effect (13), so we tested another inhibitor that blocks the acetylation reader BRD4. Interestingly, we found that BRD4i significantly reduced STAT3 phosphorylation (Fig. 1H). On the basis of the above studies, we selected BRD4i to overcome the poor efficacy of HDAC3i and evaluated whether a combination of HDAC3i and BRD4i could cause a more durable suppression of GSC growth.

Cotreatment with HDAC3i and BRD4i synergistically suppressed the viability of CSC2078, CSC1589, TS543, and TS576 cells (Fig. 2A; Supplementary Fig. S2A–S2C), as demonstrated by a marked CI values (Fig. 2B; Supplementary Fig. S2D–S2F). Similarly, cell growth was significantly suppressed by HDAC3i in the presence of a BRD4-siRNA (Supplementary Fig. S2G). Kinetic analyses following treatment of cells with HDAC3i in the presence of either JQ1 or siBRD4 showed similar robust effects on CSC2078 cell growth (Fig. 2C and D). In addition, similar patterns of growth inhibition were noted with other GSCs, CSC1589 (Supplementary Fig. S2H), TS543 (Supplementary Fig. S2J), and TS576 cells (Supplementary Fig. S2I). GSCs form undifferentiated stem-like clones in soft agar growth media. Therefore, we next tested whether silencing of HDAC3 and BRD4 also synergized to suppress colony formation. As shown in Supplementary Fig. S2K, cotreatment with HDAC3i/BRD4i caused a marked suppression of colony formation. Figure 2E and Supplementary Fig. S2L–S2N showed quantification of these data. A very similar effect on self-renewal capacity was noted when siBRD4-transfected cells were treated with HDAC3i (Fig. 2F and G; Supplementary Fig. S2O).

To further examine the mechanisms of growth suppression, we evaluated the cell-cycle status and cell death using flow cytometry. The combination of HDAC3i with either BRD4i or BRD4 siRNA (Supplementary Fig. S3A–S3G) forced more cells into G₁ arrest (Fig. 2H) and caused increased apoptosis in cells (Fig. 2I), than those treated with either HDAC3i or BRD4i alone. Consistent with these observations, qPCR analyses showed that the expression of c-Myc, and Cyclin D1 (cell-cycle progression), and Bcl-2 and Bcl-XL (anti-apoptotic) was significantly inhibited, while those of p21 (a CDK-inhibitor) and Bim and Bax (proapoptotic) were significantly upregulated after treatment with HDAC3i and BRD4i in CSC2078 cells (Fig. 2J). A similar effect was noted in human TS543 cells (Supplementary Fig. S3H). To measure apoptosis in these cells, we used TUNEL assays. HDAC3i/BRD4i caused significantly more TUNEL-positive (high fluorescence) in CSC2078 and TS543 cells (Supplementary Fig. S3I and S3J), compared with either the HDAC3i or BRD4i alone. Collectively, these observations indicated that the HDAC3i/BRD4i combination strongly suppresses GSCs proliferation via cycle arrest and promotes apoptosis.

To investigate whether the HDAC3i/BRD4i combination blocks tumor growth in vivo, we used a glioma xenograft mouse model (Fig. 3A). HDAC3i and BRD4i alone produced some significant growth inhibition compared with the untreated control in vivo. However, their combination dramatically reduced tumor volume and growth rate (Fig. 3B and C), compared with either drug alone. Importantly, there was no loss of weight (Supplementary Fig. S4A) and no organ tissue damage (Supplementary Fig. S4B) in mice following any of the treatments, indicating no significant side effects. Tumors resected from the treatment groups were used for IHC analyses to determine the changes in the expression of certain growth-associated gene products. The HDAC3i/BRD4i combination strongly suppressed the expression of growth-promoting proteins PCNA, BCL-2, c-Myc, and STAT3 pY⁷⁰⁵. These tumors also expressed significantly more BAX (Fig. 3D). TUNEL staining of tumor sections revealed that the HDAC3i/BRD4i combination induced more apoptosis when compared with the controls treated with either drug alone (Fig. 3D). Indeed, qPCR analyses of tumor tissues revealed that corresponding alterations in RNAs coding for Bim, Bax, Bcl-2, Bcl-XL, c-Myc, Cyclin D1, and p21 (Fig. 3E). In addition, Western blot analyses of tumor showed that the combination strongly blocked the pY⁷⁰⁵ of STAT3 and increased caspase-3 activation (Fig. 3F). Thus, the HDACi/BRD4i combination robustly suppresses tumor growth via an induction of apoptosis.

Given the positive effects of HDAC3 on differentiation (20, 21), we next examined whether HDAC3i/BRD4i also promotes GSC differentiation, which typically limits proliferation. Data from qPCR analyses showed that the inhibition of HDAC3 and BRD4 suppressed the expression of NESTIN (a stem cell marker) and increased that of S-100β (a mature astrocyte marker) in CSC2078 cell growth in a differentiation medium (Fig. 4A) and in vivo (Fig. 4B). Because the levels of Olig2 (an oligodendrocyte marker) and Tuj1 (a neuronal precursor marker) were inconsistent between in vitro and in vivo studies (Fig. 4A and B), we did not pursue them in further studies. Immunofluorescence imaging for cell differentiation makers in HDAC3i/BRD4i–treated cells showed a pattern similar to that of qPCR. Nestin levels were decreased, while those of S-100β were increased (Fig. 4C and D; Supplementary Fig. S5A and S5B) after treatment with HDAC3i/BRD4i. These data were also confirmed in tumor tissues using IHC analyses (Fig. 4E and F). Treatment with HDAC3i or BRD4i alone did not cause such obvious changes.

These results suggest that inhibition of HDAC3 and BRD4 could suppress the “stemness” of GSCs into and could promote their differentiation into astrocytic lineage.

To investigate a molecular basis for the induction of cell death, we analyzed gene expression profiles using RNA sequencing (RNA-seq) in CSC2078 cells treated for 48 hours with either HDAC3i or BRD4i alone, or their combination. These studies revealed alterations in the expression of 2,811 genes (down: 1,484; up: 1327) in BRD4i-treated cells; 2,562 genes (down: 880; up: 1,682) in HDAC3i-treated cells; 2,463 (down: 1,272; up: 1,191) in HDAC3i/BRD4i–treated cells (Supplementary Fig. S6A) and also revealed that there were 1,294 overlapping genes between HDAC3i- and HDAC3i/BRD4i–treated cells (Supplementary Fig. S6B), compared with the sham control. Among the 1,294 genes, approximately 35 genes were increased by HDAC3i, which then were significantly inhibited in the presence of BRD4i (Fig. 5A). The synergistic upregulation indices of such genes were shown in Fig. 5B. Among the highly induced 35 genes, GLI1 was scored at the highest synergy index. On the basis of these observations, we hypothesized that GLI1 may be a key promoter of GSC growth.

Therefore, we next verified whether alterations in the expression of GLI1 occurred in GSCs treated with HDAC3i/BRD4i. GLI1 expression and STAT3 pY⁷⁰⁵ at the protein levels were blocked in GSCs treated with HDAC3i/BRD4i (Fig. 5C; Supplementary Fig. S6C), compared with those GSCs treated with HDAC3i alone. Similarly, in the presence of siBRD4/HDAC3i and BRD4i/shHDAC3, GLI1 and STAT3 activation induced by HDAC3i was strongly suppressed (Supplementary Fig. 6D and E). Furthermore, the expression of antiapoptotic proteins Mcl-1 and Bcl-2, two known downstream targets of activated STAT3, was also suppressed in the presence of HDAC3i/BRD4i. However, it was interesting that BRD4i, on its own, did not affect GLI1 expression compared with DMSO or siControl (Fig. 5D), consistent with RNA-seq results.

To verify whether GLI1 inhibition was sufficient to promote apoptosis of GSCs, we depleted the endogenous GLI1 in cells using RNAi and then treated with HDAC3i. Western blot analyses showed that GLI1-knockdown suppressed STAT3 activation and antiapoptotic protein expression (Fig. 5E; Supplementary Fig. S6F). Silencing of GLI1 alone significantly inhibited self-renewal (Fig. 5F), cell proliferation (Fig. 5G), and induced apoptosis (Fig. 5H) in GSCs. Strikingly, depletion of GLI1 was sufficient to promote sensitivity of GSCs to HDAC3i.

Taken together, these results showed that HDAC3i/BRD4i strongly blocked STAT3 activation and cell proliferation, and promoted apoptosis, owing to a suppression of GLI1 expression.

To provide evidence that GLI1 is regulated by histone acetylation and BRD4 binding, we performed a ChIP assay in CSC2078 cells. One of the positive regulatory chromatin makers is the acetylation of histone H3 at K²⁷ position (H3K27ac; ref. 22). Using antibodies specific for this marker, we performed a ChIP assay to detect its binding at the GLI1 promoter. The experiment process of ChIP was shown in Fig. 6A. Two distinct sets of primers were used for this assay. DNA recovered in the ChIP products was used as a template for qPCR with GLI1-specific primers. HDAC3i treatment significantly increased the deposition of H3K27ac at the GLI1 enhancer (Fig. 6B). We therefore suspected that BRD4 may be involved in GLI1 transcription in HDAC3i-treated cells. Therefore, we performed a similar ChIP assay with a BRD4-specific antibody. ChIP-qPCR showed a recruitment of BRD4 to GLI1 promoter in the presence of HDAC3i (Fig. 6C).

GLI-1 is known to regulate STAT3 (23). Therefore, we next investigated the mechanism by which GLI-1 activates STAT3. We hypothesized that GLI-1 induces the expression of a cytokine that induces STAT3. IL6 is a positive regulator of Y⁷⁰⁵ phosphorylation of STAT3 (23). To examine the possible connections between GLI1, IL6, and STAT3 activation, we first examined IL6 expression in GSCs transfected with siGLI1. qPCR analyses demonstrated that IL6 transcript level was significantly downregulated upon silencing of GLI1 (Fig. 6D). ChIP assays revealed that GLI1 binding to the IL6 gene promoter was significantly reduced in the presence of siGLI1, compared withg the siControl (Fig. 6E). Consistent with these observations, silencing of GLI1 also suppressed IL6 protein expression as measured by Western blots (Fig. 6F and G). Similarly, secretion of IL6 into the culture media was downregulated as measured by an ELISA (Fig. 6H). Thus, IL6 appeared to be a direct target of GLI1. Using Western blots, we further demonstrated that depletion of IL6 using siRNA blocked STAT3 activation (Fig. 6I), compared with the control siRNA. In GLI1-depleted cells, IL6 loss accompanied with a loss of STAT3 activation, which culminated in the inhibition of the expression of antiapoptotic proteins. Conversely, treatment with IL6 not only reactivated STAT3 but also induced the antiapoptotic proteins in GLI1-depleted cells (Fig. 6J).

The poor survival rate of patients with GBM is a major clinical problem. Most therapeutics developed to date do not provide a long-lasting benefit to these patients (1, 3, 4). GSCs pose a major hurdle to the currently available therapeutic glioma maintenance, invasiveness, and recurrence. Hence, there is an urgent need for developing new and effective therapies of malignant gliomas from the perspective of GSCs.

Although the pan-HDAC inhibitors have shown to afford some antitumor effects in some solid tumors, their effectiveness is limited because of STAT3 activation or the drug-associated side effects caused by global inhibition of all HDACs affecting host processes in pancreatic ductal adenocarcinoma (12) and breast cancer, thus limiting their use (13). Hence, the expression of specific HDACs and their association with disease must be empirically determined in each disease before targeting them for inhibition in a therapeutic setting. shRNA screening library identified that HDAC3 is critical for maintaining stemness in GSCs in cell culture models. Consistent with this, among the HDACs, HDAC3 is overexpressed in gliomas and its expression is associated with growth and differentiation of GSCs. However, we have found that HDAC3i was not highly effective at inhibiting GSC growth due to its ability to activate STAT3. HDAC3 inhibitors are known to open chromatin structure and recruit an “acetyl-lysine reader” to the acetylation site to regulate downstream target gene expression (24). BRD4, as a histone acetylation reader, is required for transcriptional induction (25), and BRD4i inhibited STAT3 pY⁷⁰⁵ of GSCs in our study. Hence, we hypothesize that histone acetylation level alters STAT3 signaling and the simultaneous inhibition of HDAC3 and BRD4 will effectively block the aberrant growth by inhibiting STAT3 signaling. Indeed, this combination had robustly quelled proliferation and stemness, enforced differentiation and apoptosis in GSCs (Figs. 2 and 4), thereby increasing possibility of clinical promotion for glioma therapy. Mechanistically, HDAC3 inhibition alone sustained acetylation of H3K27, which allowed the recruitment of BRD4 to the GLI1 gene promoter and its transcription. GLI1 then induced IL6 expression. Such IL6 in an autocrine manner appears to activate STAT3 in GSCs. Thus, the combination of HDAC3i and BRD4i exerts a potent synergistic tumor-suppressive effect through an inhibition of the GLI1/IL6/STAT3 signaling axis glioma.

Recently, HDAC3 has been reported to regulate differentiation in hepatic cells and certain leukemia (20, 21). It promoted inflammatory gene expression by deacetylating NF-κB p65, in chronic obstructive pulmonary disease (26). HDAC3 inhibitor can effectively prevent deacetylation of primarily chromatin proteins, which ultimately impact cell proliferation, differentiation, and apoptosis by altering gene transcription (8, 27). However, very little is known about the role of HDAC3 in GSCs. RGFP966, the HDAC3i used in this study can cross the blood–brain barrier (28) and has been used in preclinical testing in other models of neuronal degeneration (29). Some studies reported that Coptis chinensis downregulated STAT3 phosphorylation by reducing the expression of HDAC3, which ultimately inhibited the invasion, migration, and proliferation of glioma cells (30, 31). In mouse embryonic stem cells (mESC), STAT3 pY⁷⁰⁵ induces the transcription of various antiapoptotic genes such as Mcl-1 and Bcl-2. STAT3 pS⁷²⁷ induces neuronal differentiation of mESCs (32). Importantly, we confirmed that STAT3 pY⁷⁰⁵ was the main mechanism for a lack of sensitivity of GSCs to HDAC3i treatment. The resistance to HDAC3i could be overcome by using BRD4 inhibitor.

GLI1 proteins participate in the Hedgehog (Hh) signaling pathways. Hh signaling pathway is composed of Hh ligands, Hh receptors (Ptch), smoothened (Smo), and GLI proteins and participated in multifarious processes including human tumorigenesis, malignancy, and metastasis (33, 34). Activation of this pathway results induces the zinc-finger transcription factors GLI1, GLI2, and GLI3. GLI1 controls the genes downstream of the Hh pathway (33, 35), which is associated with a poor prognosis in some diseases such as pancreatic and breast cancer. Indeed, induction of GLI1 by HDAC3i was consistent with a higher level H3K27ac deposition in GSCs. Such induction of GLI1 appears occur via a noncanonical pathway (36, 37). Previous studies reported that the genomic footprint of BRD4 overlaps with that of GLI1 in medulloblastomas and Hh-driven tumors are inhibited by BRD4i (25, 38), although mechanisms were not understood. In this study, we provided a direct evidence that BRD4 was recruited to the GLI1 promoter by acetylated lysine residues (in the presence of HDAC3i) and transactivated its expression.

Further investigations will be necessary to determine how GLI1 and STAT3 pathways are linked. In the medulloblastoma, GLI1 directly associates with STAT3 and regulates STAT3 activity (39). Alternatively, GLI1 could maintain the levels of activated STAT3 through the modulation of IL6 signaling, which has been demonstrated in a mouse pancreatic cancer model (40). IL6 binding to the transmembrane receptor IL6R has been shown to be a critical factor in promoting malignancy (41). Conversely, Dsg2, a desmosomal cadherin, induces STAT3 phosphorylation in basal cell carcinoma, which further increased GLI1 levels in both an autocrine and paracrine manner (42). Even so, the cross-talk between Hh/GLI1 and STAT3 pathways is relatively unknown in GSCs. The increased IL6 levels in gliomas are associated with the grade of malignancy and contribute to the maintenance of tumor heterogeneity (43, 44). The IL6/STAT3 signaling pathway promotes survival and the stemness potential of GBM (44). Therefore, we focused on the IL6 alteration. Indeed, the downregulation of GLI1 significantly lowered the expression of IL6 and the stemness of GSCs. We also confirmed that GLI1 depletion inhibited IL6 and STAT3 activation, which suggested that GLI1 acted as upstream of IL6. In summary, our studies identified a GLI1/IL6/STAT3 signaling axis as a potential target for HDC3i/BRD4i to block GSCs (Fig. 6K). Thus, targeting the GLI1/IL6/STAT3 pathway may provide therapeutic benefit by inhibiting tumor malignancy in GSCs.

No potential conflicts of interest were disclosed.

Q. Wang: Conceptualization, data curation, formal analysis, methodology, writing-original draft. S. Jia: Data curation, formal analysis. D. Wang: Data curation, formal analysis. X. Chen: Data curation. D.V. Kalvakolanu: Writing-review and editing, administrative, technical, or material support. H. Zheng: Writing-review and editing, administrative, technical, or material support. X. Wei: Formal analysis. N. Wen: Formal analysis. H. Liang: Formal analysis. B. Guo: Conceptualization, supervision, funding acquisition, methodology, writing-review and editing, administrative, technical, or material support. L. Zhang: Conceptualization, supervision, funding acquisition, writing-review and editing, administrative, technical, or material support.

The current study was funded by the Fundamental Research Funds for the Central Universities, JLU and the National Natural Science Foundation of China (grant numbers 81773217 to L. Zhang), Research Fund of Jilin Provincial Science and Technology Department (grant numbers 20190701065GH and 20200404120YY to L. Zhang), Jilin Province Health Technology Innovation Project (grant numbers 2019J030 to L. Zhang), and Chunhui International Research Project of Ministry of Education. D.V. Kalvakolanu is supported by the Cigarette Restitution Funds of the University of Maryland Greenebaum NCI Comprehensive Cancer Center. We thank Dr. Cameron Brenner at the Holland and Mellinghoff laboratories (MSKCC) for providing the human GSC lines TS543 and TS576.

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.