Inhibition of EZH2 Enhances the Antitumor Efficacy of Metformin in Prostate Cancer

Prostate cancer is the second leading cause of cancer-related death in males in the United States, with 174,650 new cases and 31,620 deaths estimated in 2019 (1). Despite the benefit from local therapy and initial hormone therapy, the majority of patients will relapse with castration-resistant prostate cancer (CRPC), which might progress to metastatic disease. Some drugs, such as abiraterone acetate, enzalutamide, docetaxol, sepuleucel-T, have been approved by the FDA to treat advanced prostate cancer, but they can only prolong patients' life for 3 to 6 months (2–5). Therefore, discovery of new therapy strategies is urgently needed.

Metformin (N, N-dimethylbiguianide), the most commonly used oral drug to treat type II diabetes, has a good safety profile and limited side effects. Increasing observational and cohort studies have shown that patients with diabetes who were treated with metformin usually exhibited lower risk of cancer (6), indicating that it is feasible to repurpose metformin as an anti-cancer drug. In prostate cancer, metformin was observed to reduce prostate cancer incidence and slow down the development of CRPC (7, 8). Moreover, a phase II trial observed prostate-specific antigen (PSA) secretion was decreased by the use of a high-dose metformin in progressive metastatic CRPC (9). Recently, a population-based cohort study showed encouraging results that metformin use after diagnosis of prostate cancer might increase survival of patients (10). All of these suggest that metformin could be a useful medication for prostate cancer therapy, but more studies are still needed to further evaluate such a notion.

Enhancer of zeste homolog 2 (EZH2), the catalytic subunit of polycomb-repressive complex 2, suppresses expression of a number of genes via catalyzing histone-3 lysine 27 trimethylation (H3K27me3; ref. 11). Accumulating evidence has shown that EZH2 plays an important role in tumor oncogenesis and progression (12). GSK126 (GSK2816126; ref. 13) is an S-adenosyl-methionine–competitive inhibitor targeting EZH2, and it has recently been shown effective and well tolerated in lymphoma (14). However, EZH2 also acts independently of its methyltransferase activity. For instance, EZH2 can activate NK-κB by forming a complex with RelA and RelB, in which methylation is not involved (15). In prostate cancer, EZH2 can increase AR's expression by binding to its promoter, which is also methylation-independent (16). Therefore, although several EZH2 methyltransferase inhibitors, including GSK126, are proved effective, suppression of EZH2's protein level by drugs should be considered to enhance therapy targeting EZH2.

In our study, we found that the combination of metformin and GSK126 exerts a synergistic antiproliferative effect in prostate cancer cell lines and in human prostate tumor explants. Furthermore, we demonstrate that metformin can induce downregulation of EZH2 via upregulating miR-26a-5p in LNCaP. Although metformin's role was strongly hampered by the interaction between androgen receptor (AR) and EZH2 in 22Rv1, it can be restored when combined with GSK126. All of these suggest that the combination of metformin and GSK126 is effective, and acts, at least partially, through inhibition on both EZH2's expression and methyltransferase activity.

LNCaP, 22Rv1, PC3-Neo, and PC3-AR were used in this study. LNCaP and 22Rv1 were purchased from the ATCC. PC3-Neo and PC3-AR were kindly provided by Dr. Kerry Burnstein (University of Miami). All cells were grown in RPMI-1640 medium supplemented with 10% FBS in a humidified incubator at 37°C with 5% CO₂. All cells were within 50 passages and Mycoplasma were detected every 3 months using MycoAlert PLUS Mycoplasma Detection Kit (Lonza, LT07-705). R1881 was purchased from Sigma, metformin and GSK126 were obtained from Selleckchem.

Whereas antibodies against caspase-9, cleaved caspase-9, caspase-3, cleaved caspase-3, PARP, cleaved PARP, H3, β-actin were purchased from Cell Signaling Technology, antibodies against AR, EZH2, and H3K27me3 were obtained from Active Motif. Antibody against Ki-67 was purchased from Abcam.

Upon harvest, cells were suspended with TBSN buffer (20 mmol/L Tris-HCl, pH 8.0, 0.5% NP-40, 5 mmol/L EGTA, 1.5 mmol/L EDTA, 0.5 mmol/L sodium vanadate, and 150 mmol/L NaCl) with protease inhibitors and phosphatase inhibitors, sonicated and then collected, followed by protein concentration measurement by Protein Assay Dye Reagent from Bio-Rad. Equal amounts of protein lysates from each sample were mixed with SDS loading buffer, resolved by SDS-PAGE, transferred to PVDF membranes, followed by incubations with appropriate primary and secondary antibodies.

Total RNA was extracted from tissues or cells using RNeasy mini kit (Qiagen) and reverse transcribed into cDNA using miScript II RT kit (Qiagen). FastStart Universal SYBR Green Master (Roche) was used to measure the expression level of mRNA. The primers used are: EZH2, tccctagtcccgcgcaatgagc (forward), ttgttggcggaagcgtgtaaaatc (reverse); β-actin, agaactggcccttcttggagg (forward), gtttttatgttctatggg (reverse). For the detection of microRNAs, specific primers (MS0029239, MS00008372, MS00031220, MS00003122, MS00003129) were purchased from Qiagen, and cDNA was amplified using miScript SYBR Green PCR Master Mix (Qiagen). The relative expression level of miRNA or mRNA was normalized to RNU6-2 or β-actin, respectively.

Cells (0.5 × 10³) were seeded in 6-well plates with 2 mL RPMI-1640 supplemented with 10% FBS. One day later, the medium was replaced with new medium containing different drugs. After 12 days, the colonies were fixed by 10% formalin and stained with 5% crystal violet. Colony numbers were counted by using ImageJ software.

A total of 5,000 prostate cancer cells were seeded into 96-well plates, and treated with different drugs with indicated concentrations for 72 hours, followed by incubation with the tetrazolium dye MTT 3-(4, 5-dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide for 4 hours. After the purple formazan was dissolved by DMSO, absorbance at 570 nm was measured by a plate reader.

Cytotoxicity of the drugs was evaluated via determining viability by MTT assay. Combination indices (CI) were calculated by the multiple drug effect equation of Chou and colleagues (17). CI = (D)₁/(D_x)₁ + (D)₂/(D_x)₂, where (D_x)₁ and (D_x)₂ in the denominators are the doses for metformin and GSK126 alone that gives x% inhibition, whereas (D)₁ and (D)₂ in the numerators are the doses of metformin and GSK126 in combination that also inhibited x%. Antagonism is indicated when CI > 1, CI = 1 indicates an additive effect, and CI < 1 means synergy.

Chromatin immunoprecipitation (ChIP) assay was performed by using a commercial kit (Millipore, #17-10085) following the manufacturer's instruction. AR-binding sites were predicted by PROMO (18, 19), and the PCR generated approximately 200 bp products from the miR-26a-5p proximal (<2,000 bp) promoter-containing sites. Antibodies against AR (#39781) and EZH2 (#39901) were purchased from Active Motif. For Re-ChIP, the immunoprecipitated protein–DNA complexes were eluted with Re-ChIP elution buffer (1 × TE, 2% SDS, 15 mmol/L DTT) at 37°C for 30 minutes, and the elutes were diluted 20-fold with ChIP dilution buffer for further incubation with the secondary antibodies and beads. The primers used are: P1, gttgtgggtccaagtacaaatagttttcc (forward), caatatcacctgcctggcctcaa (reverse); P2, gaatttcagaagtttccgtatcccccac (forward), cttttggggtgggtatttgctaaagat (reverse); P3, aattaaaatgaaaattccagtctcctgcctcc (forward), gatggcttttaaaagcatgaagtgtgga (reverse); P4, gcaatagaatgcagaccgatggg (forward), ctatgggagctttctgtccttggc (reverse).

HEK293T or PC3 cells were transfected with the indicated plasmids using lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. Briefly, 0.1 μg of reporter plasmids containing the sequence of interest, together with 5 ng of an internal control plasmid pRL-TK, were transfected into the cells cultured in a 24-well plate. For the expression of AR, EZH2 (S21D) and EZH2 (S21A), the indicated amounts of plasmids were co-transfected into the cells. Total plasmid DNA was normalized to 0.6 μg per well by using an empty plasmid. Luciferase activity was assayed after 24 hours of transfection using a Dual Luciferase reporter assay system (Promega). The firefly luciferase activities were corrected by the corresponding Renilla luciferase activities and presented as means ± S.D.

All the animal experiments were approved by the Purdue University Animal Care and Use Committee. 22Rv1 cells (2.5 × 10⁵/mouse) were mixed with Matrigel (Collaborative Biomedical Products), and the mixture was injected subcutaneously into right flanks of castrated nude mice (Harlan Laboratories). After 2 weeks, the tumor-bearing mice were randomized into control and treatment groups (four mice/group). Metformin was dissolved in water and administered to mice via oral gavage (30 mg/kg body weight/d). GSK126 in 20% captisol with PH adjusted to 4–4.5 was injected intraperitoneally into mice (50 mg/kg body weight/d; refs. 20, 21). Tumor volumes were calculated from the formula V = L × W²/2 [where V is volume (cubic millimeters), L is length (millimeters), and W is width (millimeters)].

Mice bearing LuCaP35CR tumors were obtained from Dr. Robert Vessella at the University of Washington. Tumors were implanted and amplified in pre-castrated NSG mice. When tumor size was big enough, tumors were harvested and cut into approximately 25 mm³ pieces, followed by implantation into 16 pre-castrated NSG mice. After tumor size reached approximately 200 mm³, mice were randomized into four groups, followed by similar treatment and measurement as described above.

Xenograft tumors were fixed in 10% neutral-buffered formalin, paraffin-embedded, sectioned to 5 mm, and stained using conventional hematoxylin and eosin (H&E) staining. Immunofluorescent chemistry staining was accomplished with the M.O.M.TM kit from Vector Laboratories.

Blood was collected from mice by retro-orbital bleeding, followed by centrifuge to collect serum. PSA levels were determined using a PSA ELISA kit (Abnova KA0208) as manufacturer instructed.

All numerical data are presented as mean ± SD. The statistical significance of the results was analyzed by using unpaired two-tailed Student t test. The P values of <0.05 indicate statistical significance.

To investigate whether metformin and GSK126 act synergistically to inhibit the growth of prostate cancer cells, colony formation assay was conducted with LNCaP, 22Rv1, and RWPE-1. Compared with treatment of metformin or GSK126 alone, combination of the two drugs exerted a stronger inhibitory effect on colony formation by LNCaP and 22Rv1 (Fig. 1A and B), but not the colony formation by RWPE-1 (Fig. 1C), a non-transformed prostate epithelial cell line. In comparison with mono-treatments, the combination of metformin and GSK126 also led to a greater inhibitory effect on cell survival of LNCaP and 22Rv1 (Fig. 1D and E), but RWPE1 cells were not affected (Fig. 1F). Moreover, the role of apoptosis was investigated upon different treatments. In LNCaP, the treatment with metformin or GSK126 as a single agent could induce slight apoptosis, but no combinational effect was observed (Fig. 1G; Supplementary Fig. S1A). In contrast, neither metformin nor GSK126 alone induced apoptosis in 22Rv1, but there was a dramatic increase of apoptosis induced by the combination (Fig. 1H; Supplementary Fig. S1B). Finally, CIs were measured to determine the types of drug interactions. As shown in Fig. 1I and J, the combinations exhibited slight to moderate synergy in LNCaP (CI range, 0.87–0.72), and moderate to strong synergy in 22Rv1 (CI range, 0.67–0.33). Altogether, these results demonstrate that the combination of metformin and GSK126 exerts synergistic inhibitory effect on prostate cancer cell growth.

Increasing evidence shows that EZH2 is usually upregulated in prostate cancer, and closely associated with progression, invasion, and metastasis (22, 23). Therefore, we were prompted to investigate the role of EZH2 in the antiproliferative effect induced by the treatments. As shown in Fig. 2A, metformin alone significantly suppressed EZH2's expression in androgen-sensitive LNCaP cells, and notably, the combined treatment resulted in enhanced inhibition on EZH2's activity indicated by the level of H3K27me3. In contrast, 22Rv1 cells, which are androgen-independent, displayed a limited reduction of EZH2 when treated with metformin alone, but it was decreased significantly by co-treatment of metformin and GSK126 (Fig. 2B), indicating that GSK126 restored metformin's ability of downregulating EZH2. To further explore how EZH2 was degraded, the relative mRNA level was measured by qRT-PCR. Consistently, metformin alone significantly reduced EZH2 mRNA level in LNCaP (Fig. 2C), but the combination of metformin and GSK126, instead of mono-treatments, was required to decrease EZH2 mRNA level in 22Rv1 cells (Fig. 2D). It is worthy noticing that the EZH2 expression displayed a similar changing trend with the cell growth upon indicated treatments in both cell lines, so we wondered whether the growth inhibition of prostate cancer cells is indeed due to EZH2 reduction. To test this hypothesis, we expressed exogenous EZH2 in LNCaP and 22Rv1, and then treated them with indicated drugs. We found that overexpression of EZH2 partially rescued the cells from the growth inhibition induced by the treatments (Fig. 2E and F). In summary, these results indicate metformin-induced downregulation of EZH2 expression is antagonized in 22Rv1 cells, but such an ability can be restored by co-treatment with GSK126. Also, downregulation of EZH2 is one of the reasons contributing to the antiproliferative effect induced by the treatments.

Next, we aimed to further dissect the underlying mechanism for metformin-mediated EZH2 downregulation. Metformin has been reported to target a variety of microRNAs (24, 25), and some of them, including miR-26a, miR-101, let-7a, let-7b, and let-7c, directly regulate EZH2 expression in prostate cancer (26). To examine whether these microRNAs are responsible for metformin-induced inhibition on EZH2's expression, RT-PCR was used to determine their expression levels in LNCaP treated with metformin. As shown in Fig. 2G and H, metformin, instead of GSK126, enhanced the expression of miR-26a-5p and miR-101-3p. However, LNCaP cells treated with metformin exhibited decreased expression levels of let-7a-5p, let-7b-5p, and let-7c-5p (Fig. 2I), which could not explain metformin-induced EZH2 downregulation. To further verify these observations, LNCaP cells were transfected with miRNA inhibitors, which are small, double-stranded RNA molecules designed to inhibit specific mature miRNAs, and then treated with metformin. As shown in Fig. 2J, metformin-induced reduction of EZH2 was restored by miR-26a-5p inhibitor, but not by the inhibitors targeting miR-101-3p (Fig. 2K), let-7a-5p, let-7b-5p, and let-7c-5p (Fig. 2L), indicating that miR-26a-5p is the mediator of metformin-induced EZH2 downregulation. Moreover, the miR-26a-5p level was also assessed in 22Rv1 upon indicated treatments, and we found that although metformin alone failed to upregulate miR-26a-5p, the combined treatment enhanced it (Fig. 2M). To further validate this finding, we applied the miRNA inhibitor to block the function of miR-26a-5p. We found that the reduction of EZH2 expression level induced by the combination treatment was restored upon addition of miR-26a-5p inhibitor (Fig. 2N). Also, as EZH2 increased, more 22Rv1 cells survived upon the combination treatment (Fig. 2O). Collectively, we conclude that metformin-induced downregulation of EZH2 is through upregulating miR-26a-5p in LNCaP and 22Rv1.

To investigate which factors regulate metformin's effect on EZH2 level in prostate cancer cells, we compared the responses with metformin in several prostate cancer cell lines, including PC3, DU145, LNCaP, and 22Rv1. As indicated, the low-concentration metformin was capable to significantly decrease EZH2 protein levels in PC3 and DU145 (Fig. 3A and B), both of which are AR negative (Supplementary Fig. S1A). In LNCaP, which is AR positive and androgen sensitive, metformin inhibited EZH2 expression as well, but with a lower inhibition efficiency (Fig. 3C). However, metformin failed to reduce EZH2 of 22Rv1, which is AR positive and androgen-refractory (Fig. 3D). Because AR, as well as its cofactors, plays a key role in prostate cancer progression and acquisition of drug resistance, we hypothesized that AR might impede metformin's ability of downregulating EZH2. To test this hypothesis, we assessed the effect of synthetic androgen (R1881) stimulation of AR on EZH2 expression upon metformin treatment. As shown in Fig. 3E, EZH2 protein level was significantly decreased by metformin, accompanied by a reduction of AR activity in LNCaP. However, metformin-induced downregulation of EZH2 in LNCaP was partially restored by addition of R1881. Meanwhile, we also detected the level of miR-26a-5p, and found that treatment of LNCaP with R1881 significantly abolished metformin-induced re-expression of miR-26a-5p (Fig. 3F). To further confirm this, two engineered PC3 cell lines were used. PC3-AR contains the coding region of human AR and stably expresses it, whereas PC3-Neo contains the same vector without the AR cDNA sequence. PC3-AR and PC3-Neo cells were treated with metformin, followed by western blot to determine EZH2 protein levels. As shown in Fig. 3G, PC3-Neo, rather than PC3-AR, displayed a remarkable decreasing trend of EZH2 expression as metformin concentration increases. Accordingly, miR-26a-5p was upregulated upon the treatment of metformin in PC3-Neo, but not in PC3-AR (Fig. 3H). Finally, we constructed two 22Rv1 cell lines with stable knockdown of AR, and treated them with metformin. As expected, depletion of AR led to a significant decrease of EZH2 protein level (Fig. 3I) and an increase of miR-26a-5p in response to metformin treatment (Fig. 3J), further supporting the notion that AR affects metformin-induced EZH2 downregulation.

Next, we investigated whether AR could bind to the promoter of miR-26a-5p and directly regulate its expression in prostate cancer cells. After assessing the 2-kb region of genomic DNA upstream of miR-26a-5p using PROMO, we identified eight potential binding motifs for AR lying within −1,955 to −1,903, −1,136 to −1,128, −488 to −449 regions on chromosome 3, and −189 to −12 regions on chromosome 12 (Fig. 4A). To examine whether AR could physically bind to the promoter of miR-26a-5p, ChIP-PCR assays were performed in 22Rv1 and LNCaP. Two AR-binding sites, P1 and P2, of miR-26a-5p promoter regions exhibited significant enrichment upon immunoprecipitation with the AR antibody, but no bands were evident for other two sites, P3 and P4 (Fig. 4B). We then sub-cloned the promoter region, including both P1 and P2 upstream of luciferase gene into a reporter plasmid. The dual-luciferase assay showed that the transcriptional activity was reduced when AR bound to the sites, and treatment of R1881 further decreased it (Fig. 4C and D), indicating that AR binds to the promoter of miR-26a-5p and inhibits its expression as predicted. Furthermore, we asked whether metformin and GSK126 affected the AR binding to the promoter of miR-26a-5p. As shown in Fig. 4E and F, either metformin or GSK126 alone reduced the AR binding to the promoter within limited extent, but the combination of metformin and GSK126 almost completely removed AR from the regions.

It is well known that metformin can reduce AR recruitment to the promoters of its target genes (27), but we asked how GSK126 could also reduce it. Recently, EZH2 was shown to interact with AR to regulate its binding to its target genes, in a manner dependent on EZH2's methyltransferase activity (28). In addition, the interaction between AR and EZH2 is mediated by phosphorylation at S21 of EZH2 (28). Therefore, we further hypothesized that AR cooperates with EZH2 to suppress miR-26a-5p expression. To test this, the miR-26a promoter construct was co-transfected with empty vector, AR, EZH2-S21D and EZH2-S21A either alone or in combination with AR plus EZH2-S21D and AR plus EZH2-S21A in HEK293T cells. We found that EZH2-S21D, instead of EZH2-S21A, promoted AR's inhibition on the transcriptional activity (Fig. 4G), suggesting that EZH2 interacts with AR to regulate the expression of miR-26a-5p. To further validate this point, re-ChIP assay was performed, and revealed that AR, bound to the miR-26a-5p promoter region, was significantly associated with EZH2 in 22Rv1 cells (Fig. 4H), but not in LNCaP cells (Fig. 4I). Altogether, these results suggest that expression of miR-26a-5p was suppressed by AR in prostate cancer cells, and that the AR/EZH2 complex reinforced the suppression in advanced androgen-refractory prostate cancer cells.

To further validate our in vitro finding, we evaluated the effect of metformin and GSK126 alone or in combination with a 22Rv1-derived xenograft mouse model. After a 24-day treatment, we found that metformin alone could barely affect the tumor growth, and GSK126 alone only exerted a limited inhibitory effect on it (Fig. 5A–C). In contrast, the combination of metformin and GSK126 significantly blocked the tumor growth and decreased the tumor weight. Meanwhile, no significant body weight loss was observed among all four groups (Fig. 5D), implying that the combination of the two drugs with the indicated doses has little toxic side effect. To confirm the responses, we conducted histological analyses of these tumor samples. H&E staining showed that single-agent metformin, as well as GSK126, slightly reduced the tumor cell content (Fig. 6E); however, tumor cell content was markedly decreased after the combination therapy with metformin and GSK126 (Fig. 5E). Furthermore, the immunostaining of Ki67 and cleaved caspase-3 also showed that the combination of metformin and GSK126 led to a significant decrease in overall proliferation and a dramatic increase of apoptosis (Fig. 5F–I). Finally, proteins were extracted from the harvested tumors and subjected to western blot against EZH2. We also found that co-therapy of metformin and GSK126 significantly lowered EZH2 expression than the mono-therapies (Fig. 5J–L). These results are consistent with what we observed in the cell-based experiments, thus confirming the synergistic effect between metformin and GSK126.

To better mimic the growth and situation of CRPC, a patient-derived xenograft study was conducted with LuCaP35CR, which could recapitulate the major genomic and phenotypic features of the disease in humans. As predicted, the combination of GSK126 and metformin resulted in a stronger tumor-inhibitory effect than did monotherapies (Fig. 6A–C). Of note, the serum PSA, which is often elevated with prostate cancer progression, displayed significant reduction upon the combination treatment (Fig. 6D). Furthermore, histological analyses were performed with the tumor samples. As shown in Fig. 6E, GSK126-treated tumors exhibited some apoptotic bodies and morphological changes, including cytoplasm reduction, nuclear pyknosis, and karyorrhexis. Remarkably, the tumors treated with both metformin and GSK126 showed increasing number of apoptotic bodies with pyknotic or fragmented nuclei, as well as condensed cytoplasm (Fig. 6E). Immunostaining of Ki67 and cleaved caspase-3 also revealed that the combination of metformin and GSK126 led to a significant decrease of overall proliferation and a dramatic increase of apoptosis (Fig. 6F–I). Finally, we also analyzed EZH2 protein levels in all the tumor samples, and found that co-treatment of metformin and GSK126 led to a lower expression of EZH2 compared with either drug alone (Fig. 6J–L). In summary, these results are consistent with that of 22Rv1-derived xenograft study, further confirming that metformin and GSK126 can act synergistically in CRPC.

We have demonstrated that the combination of metformin and GSK126 synergistically inhibited proliferation of LNCaP and 22Rv1, but not RWPE1 whose EZH2 is not affected by metformin (Supplementary Fig. S2B). Downregulation of EZH2 is one of the reasons contributing to the antiproliferative effect induced by the treatments. However, the cell viability could not be completely rescued by overexpression of exogenous EZH2. This could be because (i) transduction efficiency of Myc-EZH2 is not high enough; (ii) other molecules besides EZH2 may also be involved in the process. Compared with LNCaP, the combination is more synergistic in 22Rv1; meanwhile, 22Rv1 is much more resistant to metformin, suggesting that GSK126 strongly enhances the potency of metformin and re-sensitizes 22Rv1 to it. Previous studies of our laboratory have showed that different prostate cancer cells exhibit different sensitivities to metformin, and advanced AR-positive prostate cancer cells are usually resistant (29, 30), which is concordant with our finding. Similar to most malignancies, prostate tumors are usually composed of multiple cell types, with complexed characteristics and biological features, resulting in intratumoral heterogeneity; therefore, it seems that metformin alone is not a promising anti-cancer therapy for CRPC, and combinations of metformin with other drugs potentiating its function should be more viable.

It has been well documented that EZH2 is critical for prostate cancer growth, development, and progression (12). Recent studies revealed that EZH2 could act independently of its methyltransferase activity (16), indicating that additional approaches inhibiting EZH2 expression should be considered. In our study, we have demonstrated that metformin downregulated EZH2 through upregulating miR-26a-5p by using of microRNA inhibitors. Because EZH1 contains the same miR-26a–targeting sequence (UACUUGA) in its 3′-UTR region, it is likely that the miR-26a-5p inhibitor can stabilize H3K27me3 through regulating EZH1, consequently resulting in discordant change at the levels of EZH2 and H3K27me3. Furthermore, we have found that metformin's effect on EZH2 was actually antagonized by AR, which impeded metformin-induced re-expression of miR-26a-5p, and modification of AR changed metformin's effect on EZH2. Metformin could decrease EZH2 by eliminating AR's inhibitory effect on miR-26a-5p in LNCaP, but it could also reduce EZH2 in PC3 and DU145 as well, indicating that additional molecules besides AR are involved in metformin's regulation on EZH2. However, AR is still considered to be the determinant of the combination synergy, as we did not observe significant synergy using PC3 or DU145 (Supplementary Fig. S2C–S2F).

It was reported that AR inhibited gene expression by directly binding to its promoter (31). Moreover, AR was showed to regulate miRNAs' expression by directly binding to their promoters (32, 33). In our study, we observed AR bound to the promoter of miR-26a and suppressed its expression. Furthermore, the gene inhibition induced by AR could be reinforced by EZH2-S21D, which is consistent with previous studies demonstrating that EZH2 promoted AR recruitment to its sites via directly methylating it (28, 34). Also, we found that GSK126 decreased AR's recruitment at the promoter of miR-26a, further supporting the interaction between AR and EZH2 we observed.

We also tested the combination of the two drugs by in vivo experiments, and found that it could inhibit tumor growth, induce apoptosis, and downregulate EZH2 in both 22Rv1-derived tumors and LuCaP35CR xenografts. Meanwhile, the dosage we used did not cause any toxic effect. In the phase I study, GSK126 was escalated to maximum dose of 3,000 mg twice a week for patients with no dose limiting toxicity observed (14). However, the dosage used in our experiments is 50 mg/kg body weight/d, which was slightly higher and more frequent than that in the clinical trial. Till now, the safety limit of GSK126 is still undetermined, so our results are valuable to define the appropriate dosage range of medication of GSk126 for clinical use. The safe dosage of metformin for patients should be below 3,000 mg per day (35, 36), and the dosage used in our mice experiments is 30 mg/kg body weight/d, which is within the safe medication range. Although plasma concentrations of metformin in the vein may only reach tens of micromoles in the mice (37), it is enough to induce a significant lower cell survival rate (Supplementary Fig. S3A), as well as a stronger inhibition on EZH2 (Supplementary Fig. S3B), when combined with GSK126.

In summary, as shown in Fig. 6M, miR-26a-5p is negatively regulated by AR in LNCaP, which can be easily eliminated by metformin. Moreover, EZH2 reinforces AR's inhibition on miR-26a-5p expression in 22Rv1, which results in the resistance. However, inhibition of EZH2's methyltransferase activity with GSK126 can inhibit the interaction between AR and EZH2, restoring metformin's effect in 22Rv1. Therefore, our results suggest that the combination of metformin and GSK126 would be an effective approach targeting EZH2 for future prostate cancer therapy, in particular, for patients with AR-positive CRPC.

No potential conflicts of interest were disclosed.

Y. Kong: Conceptualization, data curation, methodology, writing-original draft. Y. Zhang: Investigation. F. Mao: Methodology. Z. Zhang: Methodology. Z. Li: Investigation, visualization. R. Wang: Methodology. J. Liu: Methodology. X. Liu: Conceptualization, supervision, funding acquisition, investigation, visualization, methodology, project administration, writing-review and editing.

This work was funded by NIH R01 CA157429, R01 CA192894, R01 CA196835, and R01 CA196634 (to X. Liu) and the Chinese Scholarship Council (to Y. Kong). This research is also supported by the University of Kentucky Markey Cancer Center (P30CA177558).

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.