Bidirectional Regulation between NDRG1 and GSK3β Controls Tumor Growth and Is Targeted by Differentiation Inducing Factor-1 in Glioblastoma

In 2016, the World Health Organization (WHO) revised the classification of glioma based on the presence of mutations in the isocitrate dehydrogenase (IDH) 1 and 2 genes and chromosome 1p/19q status (1). Glioblastoma (GBM), WHO grade IV and the most common and aggressive primary brain tumor, exhibits a high recurrence rate, and poor prognosis attributable to its invasive nature and resistance to therapy (2). Although abnormal genomic alterations underlying GBM pathogenesis have been discovered (3, 4), genomic changes that specifically drive the growth and survival of GBM cells remain unclear. Combination of standard therapy, irradiation, and temozolomide, with molecular-targeted drugs, such as EGFR, PI3K, or mTOR inhibitors, failed to improve the overall or progression-free survival in phase III clinical trials (2). The development of effective therapeutic strategies with improved benefits for patients with GBM requires identification of signaling pathways that critically regulate the survival and proliferation of GBM.

N-myc downstream regulated gene 1 (NDRG1) is a member of the NDRG protein family consisting of NDRG1–4, which are evolutionarily well conserved. NDRG1 is a 43 kDa, 394 amino acid protein and predominantly localized in the cytoplasm (5). Initially identified as a differentiation-related gene during embryogenesis, NDRG1 is inversely correlated with N-myc expression (6) and its mutation is responsible for hereditary motor and sensory neuropathy (7). NDRG1 is intimately involved in multiple stages of differentiation, including placentation (8), trophoblast formation (9), and maturation of mast cells (10) and macrophage lineage cells (11), as well as in the morphogenesis of various organs (11–13).

NDRG1 suppresses metastasis and oncogenesis in cancers of the brain, breast, colon, esophagus, pancreas, and prostate (5), whereas it promotes tumor growth and metastasis in cancers of the cervix, liver, lung, and stomach (5). We have previously reported that NDRG1 overexpression in pancreatic cancer cells suppresses tumor growth and angiogenesis via attenuation of NF-κB signaling pathway (14–16). In contrast, NDRG1 overexpression in gastric cancer cells promotes tumor growth and angiogenesis through enhanced JNK/AP-1 activation (17). Therefore, NDRG1 is a double-edged sword in human malignancies, and its role and effect could differ depending on tumor types. The precise mechanisms of how NDRG1 regulates tumor malignancy remain unclear.

NDRG1 expression in the nervous system is considerably enhanced during regeneration of injured nerves (18). In neurogenic tumors, higher NDRG1 expression is a favorable prognostic biomarker for patients with neuroblastoma (19). In addition, studies consistently reported that higher NDRG1 expression is correlated with favorable prognosis of patients with gliomas (20, 21). On the other hand, it was reported that NDRG1 bound and stabilized O⁶-methylguanine-DNA methyltransferase (MGMT), which inhibits the function of alkylating agents, and conferred resistance to temozolomide in GBMs (22). Thus, the functional significance of NDRG1 in GBM pathogenesis and patient outcomes remains unclear.

Here, we demonstrate that NDRG1 overexpression induces cell growth suppression and G₀–G₁ arrest accompanied by reduction of GSK3β expression and inactivation of cell growth signaling pathways. Furthermore, we demonstrate that the differentiation inducing factor-1 (DIF-1) induces a marked enhancement of NDRG1 expression, resulting in tumor growth suppression in vitro and in vivo. We discuss whether enhancing NDRG1 expression could be a promising approach to the development of potent and novel anti-GBM therapeutics.

We analyzed 28 patients who were newly diagnosed as GBM at Saga University Hospital (Saga, Japan) between 2009 and 2016. Clinical status including IDH1 R132H mutation and methylation of MGMT promoter of the patients is described in Supplementary Table S1. Written informed consent was obtained from all patients or their guardians, and all specimens were collected with the approval of the ethics committee and the Institutional Review Board at the Saga University Hospital (Saga, Japan; approval number: 2017-08-01). This study conforms to the principles of the Declaration of Helsinki. Surgically removed or biopsied GBM specimens were routinely fixed in 20% neutralized formalin and embedded in paraffin. For IHC analysis, 4-μm thick sections of a representative block for each tumor were deparaffinized. Heat-induced epitope retrieval was performed with citrate buffer (pH 6.0). The slides were treated with Peroxidase Blocking Solution (S2023; Dako) for 10 minutes to block endogenous peroxidase activity. The slides were then incubated with a primary antibody against NDRG1, which was generated as described previously (1:1,000; ref. 14) for 60 minutes at room temperature. Antibody labeling was achieved using EnVision + Dual Link System HRP (K4061, Dako), and visualized with diaminobenzidine, followed by counterstaining with hematoxylin. NDRG1-positive cells were counted from five high power fields (400×) for each IHC specimen in all patients. All specimens were independently scored for NDRG1 immunopositivity by two observers (H. Ito and Y. Nakahara) blinded to clinical information and the other observer's score.

Human GBM cell lines, U87MG, U251, and T98G were purchased from ATCC and U343 was kindly provided by B. Wetermark (Institute of Pathology, Uppsala, Sweden). 293TN was from System Bioscience. These cells were cultured in DMEM supplemented with 10% FBS. MGG8 and MGG23, human glioma stem–like cell (GSC) lines, were established and characterized at Massachusetts General Hospital (Boston, MA; refs. 23, 24), and cultured in neurobasal medium (Invitrogen) supplemented with l-glutamine (3 mmol/L), B27 supplement (Invitrogen), N2 supplement (Invitrogen), heparin (5 μg/mL; Sigma-Aldrich), EGF (20 ng/mL; R&D Systems), and FGF2 (20 ng/mL; PeproTech). All cell lines were obtained between 1992 and 2015. All cell cultures were maintained at 37°C in a humidified incubator with an atmosphere of 5% CO₂. All cell lines were passaged for <6 months and were not further tested or authenticated by the authors.

Anti-NDRG1 antibody was generated as described previously (1:5,000; ref. 14). Anti-phosphorylated-NDRG1 (Ser330) (1:1,000; 3506), anti-phosphorylated-NDRG1 (Thr346) (1:1,000; D98G11; 5482), anti-phosphorylated-AKT (Thr308) (1:1,000; D25E6; 13038), anti-phosphorylated-AKT (Ser473) (1:2,000; D9E; 4060), anti-AKT (1:1,000; 9272), anti-phosphorylated-S6 kinase (Thr389) (1:1,000; 108D2; 9234), anti-S6 kinase (1:1,000; 9202), anti-phosphorylated-S6 (Ser235/236) (1:2,000; D57.2.2E; 4858), anti-S6 (1:1,000; 5G10; 2217), anti-phosphorylated-GSK3β (Ser9) (1:1,000; D85E12; 5558), anti-GSK3β (1:1,000; 27C10; 9315), anti-phosphorylated-ERK (Thr202/Tyr204) (1:2,000; D13.14.4E; 4370), anti-ERK (1:1,000; 9102), anti-phosphorylated-β-catenin (Ser33/37/Thr41) (1:1,000; 9561), anti-β-Catenin (1:1,000; 9587), anti-cyclinD1 (1:1,000; 2922), anti-CDK4 (1:1,000; 2906), anti-cyclinE1 (1:1,000; 4129), anti-CDK2 (1:1,000; 2546), and anti-N-myc (1:1,000; 9405) antibodies were purchased from Cell Signaling Technology. Anti-α-tubulin (1:5,000; B-5-1-2; T6074) and anti-FLAG (1:1,000; M2; F1804) antibodies were from Sigma-Aldrich. Anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (1:5,000; 2275-PC-100) antibody was from Trevigen. Anti-β-actin (1:5,000; ab8226) antibody was from Abcam. CT99021 (Axon1386) was from Axon Medchem, Tideglusib (S2823) was from Selleck, GSK650394 (3572) was from Tocris Bioscience, and DIF-1 (046-320921) was from FUJIFILM Wako Pure Chemical Corporation. MG132 was from Calbiochem.

Preparation of the FLAG-tagged NDRG1 and NDRG1 deletion mutants, Δ3–Δ6, expression plasmids was as described previously (16). Cells were transfected with pcDNA3-FLAG-NDRG1, pcDNA3-FLAG-NDRG1 Δ3–Δ6, or pcDNA3 expression plasmids using Lipofectamine LTX (Invitrogen) following the manufacturer's protocol. A FLAG-tagged NDRG1 cDNA was cloned into a lentiviral vector (pCDH-EF1-MCS-BGH-PGK-GFP-T2A-Puro cDNA Cloning and Expression Vector, System Bioscience) for constitutive gene expression. To establish cell lines stably overexpressing NDRG1, U87MG cells were infected with FLAG-tagged NDRG1 cDNA or control lentivirus particles (System Biosciences) for 24 hours. Puromycin (400 ng/mL) was added to select stably transduced cells. Transduction was verified using Western blots.

U343 (3.5 × 10⁴ cells), U251 (3 × 10⁴ cells), U87MG (2.5 × 10⁴ cells), and T98G (1.75 × 10⁴ cells) cells were seeded in 24-well plates. The following day, cells were transfected with NDRG1 or control siRNA (Invitrogen). Cell numbers in each dish were counted by a Z2 Coulter Particle Count and Size Analyzer (Beckman Coulter Inc.) at indicated time points. Triplicate dishes were tested at each day, and results were expressed as the mean ± SD of triplicate dishes.

U87MG and U251 (3.5 × 10⁴ cells) were seeded in 35-mm dishes. The following day, cells were transfected with FLAG-tagged NDRG1, FLAG-tagged NDRG1 deletion mutants, Δ3–Δ6, or vector. U87/mock and U87/NDRG1 sublines (#1, #2, and #3; 3.5 × 10⁴ cells) were seeded in 35-mm dishes. The cell numbers in each dish were counted by a Z2 Coulter Particle Count and Size Analyzer at indicated time points. Triplicate dishes were tested at each day, and results were expressed as the mean ± SD of triplicate dishes.

293TN (2 × 10⁵ cells) or U87MG (2 × 10⁵ cells) were seeded in 35-mm dishes. After 48 hours incubation, each dish was treated with DMSO or DIF-1 at 20 μmol/L for 24 hours, or after 24 hours incubation, each dish was transfected with mock or NDRG1 vector for 48 hours. Cycloheximide (06741; Nacalai Tesque) was then added to cells at a concentration of 10 μg/mL, and cells were cultured for the indicated times. Cells were harvested and protein levels were visualized by Western blotting with specific antibodies.

GSCs (3.5 × 10⁴ cells) were seeded into 24-well plates, incubated for 3 days, and treated with DIF-1 at indicated doses and incubated for another 3 days. Five fields of each well were captured under a 40× microscope objective and the diameter of each sphere measured, then the number of spheres that were more than 50 μm were counted. Furthermore, from each well, spheres were trypsinized and dyed with trypan blue, and the number of viable cells that excluded trypan blue was counted using TC20 (Bio-Rad).

Male BALB/c nu/nu athymic nude mice (5–7 weeks old) were purchased from CLEA. For subcutaneous xenograft model, 5 × 10⁶ U87MG cells in 200 μL of 50% Matrigel were implanted into the subcutaneous tissue of the bilateral abdominal wall of the mice. Tumor sizes were measured, and tumor volumes (mm³) were calculated as follows: length × width² × 0.5. When tumors reached 100–200 mm³, 4 mice each were randomly allocated into two groups (n of evaluable tumors = 6/control group, and n = 7/DIF-1 treatment group). DIF-1 (300 mg/kg in the morning and 150 mg/kg in the evening, daily) or soybean oil was administered orally twice a day, 6 days/week (25). The tumors were harvested after 11 days, stored at −80°C, or fixed immediately in 10% formalin overnight at 4°C. For orthotopic xenograft model, 5 × 10⁵ U87MG cells in 10 μL of PBS and 2 × 10⁵ MGG8 cells in 5 μL of culture medium were stereotactically implanted into the right forebrain of the mice (2 mm lateral from bregma and 2 mm deep). Ten days after implantation, mice were randomly allocated into two groups (U87MG; n of evaluable tumors; n = 3/control group, and n = 4/DIF-1 treatment group: MGG8; n of evaluable tumors; n = 5/control group, and n = 4/DIF-1 treatment group). DIF-1 (300 mg/kg in the morning and 150 mg/kg in the evening, daily) or soybean oil was administered orally twice a day, 6 days/week (25). The right forebrain was harvested after 12 days, fixed immediately in 10% formalin overnight. Formalin-fixed, paraffin-embedded sections were stained with hematoxylin and eosin. To evaluate tumor volumes (mm³), we measured the maximal area of each tumor, and tumor volume was calculated as: long diameter × short diameter² × 0.5. All mice were housed in microisolator cages maintained under a 12 hours light/dark cycle. Water and food were supplied ad libitum. Animals were observed for signs of tumor growth, activity, feeding, and pain in accordance with the guidelines of the Harvard Medical Area Standing Committee on Animals. All animal experimental procedures were reviewed and approved by the Animal Ethics Committee of Kyushu University (Fukuoka, Japan; registration number: A30-097-2).

Male BALB/c nu/nu athymic nude mice (5–7 weeks old) were purchased from CLEA. All mice were randomly allocated into two groups [n of evaluable plasma and brains for high-performance liquid chromatography (HPLC) and gas chromatography–mass spectrometry (GC–MS) analysis; n = 3: n of evaluable brains for Western blot analysis; n = 7/control group, and n = 8/DIF-1 treatment group]. For HPLC and GC–MS analysis, DIF-1 (300 mg/kg) or soybean oil was administrated orally for 1 hour. For mouse plasma samples, blood samples were collected. Mouse plasma (300 μL) was isolated from blood sample by centrifugation at 500 × g for 15 minutes and was mixed with 300 μL chloroform. After mixing, the solution was centrifuged at 13,000 × g for 5 minutes, and then the organic phase was separated. To prepare brain sample, transcardial perfusion was performed with ice-cold PBS, brain tissues were harvested, and were lysed by sonication in 300 μL chloroform. A plasma or brain sample (10 μL) was then applied to a column for separation (5C18 Cosmosil AR-II 4.6 I.D. × 250 mm; Nacalai Tesque). The samples were eluted by a linear gradient of acetonitrile (30%–100%) in the presence of 0.1% formic acid over 40 minutes at a flow rate of 1.0 mL/minute. A UV detector was operated at 277 nm. A calibration curve was prepared by plotting the area ratios of DIF-1 normalized to the internal standard. A plasma or brain sample (10 μL) was also subjected to GC–MS [Shimadzu QP-2010SE with INERTCAP 5MS/SIL (0.25 mm inner diameter, × 30 m), GL Science Inc., column temperature 100°C–280°C, rate of temperature increase: 10°C/minute]. For Western blot analysis, DIF-1 (300 mg/kg in the morning and 150 mg/kg in the evening, daily) or soybean oil was administered orally twice a day, 6 days/week. After treatment of DIF-1 for 12 days, the brains were harvested and processed for Western blot analysis.

Data from The Cancer Genome Atlas (TCGA) Research Network (TCGA Glioblastoma Multiforme Provisional complete sample set), including mRNA expressions, mutations, putative copy-number alterations, and clinical data were obtained and analyzed using cBioPortal for Cancer Genomics (http://www.cbioportal.org). Data from REMBRANDT datasets and survival information for patients with GBM were obtained and analyzed using Betastasis (http://www.betastasis.com/glioma/). Patient prognoses were evaluated by Kaplan–Meier survival curves of patients with GBM with low or high expression of NDRG1 mRNA expression. In TCGA datasets, patients (n = 591) were divided into two groups by z-score over or less than ± 1.5, namely, NDRG1 low (n = 30) and NDRG1 high (n = 22). In REMBRANDT datasets, patients (n = 178) were divided into two groups by 50 percentile score of NDRG1 mRNA expression level, namely, NDRG1 low (n = 89) and NDRG1 high (n = 89).

The association of NDRG1-positive cell rates and overall survival, which was defined by the duration from diagnosis to death, was analyzed by Wilcoxon test and Student t test. The association of NDRG1 mRNA level and overall survival was analyzed by Wilcoxon test. Experimental results were expressed as mean ± SD (Figs. 2B, D, E, G, 3A and B, 4B, 5C, and H, 6A, C, and E, and 7B, and C; Supplementary Figs. S1A, S1B, S1D, S1E, S1G, S1H, S1I, S2A, S3A, S4A, and S5A) or SEM (Figs. 7A, E, F, and G; Supplementary Fig. S7A) of n observations. Statistical differences between groups were assessed by two-tailed Student t test. A P value of less than 0.05 was considered significant.

It was previously reported that higher NDRG1 expression in tumors, as assessed by IHC analysis, was predictive of improved prognosis in patients with glioma (20). First, we examined whether NDRG1 expression in tumors correlated with the survival duration of patients with GBM. Surgical samples of 28 newly diagnosed GBMs were IHC analyzed. Representative images of IHC staining of NDRG1 in four clinical samples are shown in Fig. 1A. The tumor samples of patients #26 and #27 showed abundant NDRG1 expression, while patients #1 and #8 showed NDRG1 staining in a small subset of cells (Fig. 1A), revealing heterogeneity between patients. Kaplan–Meier analysis of overall survival showed that patients with higher NDRG1 expression survived significantly longer than those with lower NDRG1 expression (P = 0.004; Fig. 1B). Furthermore, the NDRG1-positive cell rates of tumors from all 28 patients positively correlated with longer overall survival of the patients (Fig. 1C). These results suggest that NDRG1 expression levels correlate with favorable prognosis in patients with GBM.

We examined whether cell growth was affected by altered NDRG1 expression in four GBM cell lines. The cell lines U87MG and U251 harbor loss-of-function PTEN mutation, but U343 and T98G do not (26). These GBM cell lines expressed various levels of NDRG1 and cell growth–related signaling molecules; U87MG and U251 had a lower level of NDRG1 and a higher level of pAKT, as compared with U343 and T98 (Fig. 2A). It was previously reported that NDRG1 silencing enhanced GBM cell growth accompanied by upregulation of AKT phosphorylation (27). In accord with this study, siRNA-mediated NDRG1 silencing enhanced cell growth of the four GBM cell lines by 1.3- to 2.5-fold (Fig. 2B). In particular, U87MG and U251 cells showed more growth enhancement than T98G and U343 cells by NDRG1 silencing (Fig. 2B). NDRG1 siRNA markedly suppressed NDRG1 expression in all four GBM cell lines (Fig. 2C). In T98G, NDRG1 siRNA did not completely suppress NDRG1 expression, and residual NDRG1 might attenuate the effect of NDRG1 siRNA (Fig. 2C). In contrast, in U343 cells, NDRG1 expression was almost completely suppressed by NDRG1 siRNA (Fig. 2C). However the effect of siNDRG1 on cell growth was smaller than that of U87MG and U251 cells. This may be due to the difference of dependency on NDRG1 for cell growth among the cell lines. Interestingly, we found that expression of GSK3β and pGSK3β, as well as pAKT and pS6 was increased upon NDRG1 silencing across these GBM cells (Fig. 2C; Supplementary Fig. S1A). siNDRG1#3 was selected as siNDRG1 in the following studies. We further examined the effect of NDRG1 silencing on colony formation of U251, T98G, and U343 cell lines, and found that NDRG1 silencing significantly enhanced colony formation of all three GBM cell lines (Fig. 2D). U87MG could not form colonies. These data suggest that NDRG1 silencing–mediated enhancement of GSK3β, pAKT, and pS6 expression may underlie the observed increases in growth and colony formation.

Next, we examined whether GSK3β activation contributed to NDRG1 silencing–mediated enhancement of cell growth. Treatment with CT99021, a selective ATP competitive GSK3 inhibitor, specifically abrogated NDRG1 silencing–mediated growth enhancement to the control level (Fig. 2E; Supplementary Fig. S1B). In contrast, CT99021 did not significantly alter the growth of control cells (Fig. 2E; Supplementary Fig. S1B). As expected, treatment with CT99021 inhibited β-catenin phosphorylation, which is catalyzed by GSK3β (28). CT99021 also inhibited phosphorylation of AKT (T308 and S473) in both control and NDRG1-silenced cells (Fig. 2F; Supplementary Fig. S1C). Furthermore, CT99021 inhibited phosphorylation of S6K and S6 in both control and NDRG1-silenced cells, but the inhibitory effect of CT99021 was greater in NDRG1-silenced GBM cells than siControl-treated GBM cells (Fig. 2F; Supplementary Fig. S1C and S1D). Moreover, another GSK3β inhibitor, tideglusib, a selective, irreversible, and non-ATP–competitive inhibitor, consistently annihilated cell growth enhancement and suppressed S6K and S6 phosphorylation more in NDRG1-silenced cells than siControl-treated GBM cells (Fig. 2G and H; Supplementary Fig. S1E–S1G). There was no significant effect of CT99021 or tideglusib on total S6K and S6 protein expressions in both control and NDRG1-silenced GBM cells (Supplementary Fig. S1H). These data suggest that GSK3β activation contributes to cell growth enhancement by NDRG1 silencing.

The C-terminal domain of the NDRG1 protein contains several sites for GSK3β and serum and glucocorticoid-regulated kinase (SGK)-mediated phosphorylation (29). We further examined whether SGK activity was also involved in enhanced cell growth after NDRG1 silencing. A competitive SGK inhibitor, GSK650394, suppressed enhanced cell growth by NDRG1 silencing at 1 μmol/L (Supplementary Fig. S1I). However, GSK650394 suppressed cell growth and AKT, S6K, and S6 phosphorylation at similar levels in both control and NDRG1-silenced cells (Supplementary Fig. S1I and S1J). Although both GSK3β and SGK inhibitors suppressed cell growth enhanced by NDRG1 silencing, effects of GSK3β inhibitor were more specific in NDRG1-silenced cells. These results suggest the selective involvement of GSK3β, rather than SGK, in NDRG1-dependent cell growth alteration.

We next examined the effect of NDRG1 overexpression on cell growth. Transient transfection with exogenous NDRG1 cDNA induced growth suppression of U87MG and U251 cells (Fig. 3A; Supplementary Fig. S2A). Furthermore, we established three U87/NDRG1 sublines (#1, #2, and #3) via stable overexpression with lentivirus vector transduction, and observed that their growth rates were significantly slower than that of empty vector–transduced U87MG (U87/mock; Fig. 3B). The NDRG1-overexpressing sublines showed reduced levels of GSK3β, pβ-catenin, pAKT, pS6K, and pS6 expression (Fig. 3C), indicating that NDRG1 overexpression resulted in inactivation of GSK3β and AKT/S6 cell growth signaling pathways. We further found that transient overexpression of NDRG1 in another GBM cell line, U251, also induced inactivation of these signaling molecules (Supplementary Fig. S2B). U87/NDRG1 cells showed an increased cell population at G₀–G₁-phase and a decreased cell population at G₂–M-phase compared with U87/mock cells (Fig. 3D), indicative of cell-cycle arrest at G₀–G₁. NDRG1 overexpression decreased expression of cyclin D1 and E and cyclin-dependent kinase (CDK) 2 and 4 (Fig. 3E). These cyclins (D1 and E) and CDKs (2 and 4) are involved in cell-cycle transition from G₁ to S-phase (30). In addition, we examined the effect of NDRG1 overexpression on cell apoptosis and DNA synthesis by using Annexin V-FITC/propidium iodide (PI) staining assay and bromodeoxyuridine (BrdU) incorporation assay, respectively. NDRG1 overexpression did not alter apoptotic cell (Annexin V–positive) fractions in both U87MG and U251 cells (Supplementary Fig. S2C). In contrast, NDRG1 overexpression inhibited BrdU incorporation in both GBM cell lines, suggesting that NDRG1 overexpression inhibits DNA synthesis (Supplementary Fig. S2D). Together, NDRG1 suppresses GBM cell growth by concomitant downregulation of GSK3β, AKT/S6, and cell-cycle signaling pathways.

Furthermore, we determined the mechanism underlying the reduction of GSK3β expression mediated by NDRG1 overexpression. We examined the effect of NDRG1 overexpression on the stability of the GSK3β protein. In the presence of cycloheximide, NDRG1 overexpression destabilized GSK3β with a half-life of approximately 3 hours compared with control cells wherein the GSK3β protein exhibited a half-life of >24 hours, showing its high stability (Fig. 3F and G; Supplementary Fig. S2E). It has previously been reported that a proteasomal inhibitor, MG132, inhibited GSK3β degradation induced by HSP90 inhibitor or dexamethasone (31, 32). Consistent with these studies, the accelerated GSK3β degradation by NDRG1 overexpression was completely blocked by MG132 (Fig. 3H). These data suggested that NDRG1 reduced GSK3β levels by proteasomal degradation pathway.

We examined whether a specific domain of NDRG1 was responsible for NDRG1-induced cell growth suppression. Analysis of the amino acid sequence of NDRG1 indicates the presence of a phosphopantetheine attachment, a prominent α/β hydrolase fold, and NDRG1 phosphorylation sites by GSK3β and SGK at its C-terminal domain. We previously established several deletion mutants of NDRG1; Δ3 (234–304AA) and Δ4 (180–294AA) harbor deletions of the partial α/β hydrolase fold, while Δ5 and Δ6 harbor deletions of the partial (326–350AA) and total (326–394AA) phosphorylation sites on the C-terminal domain (Fig. 4A; refs. 16, 29, 33, 34). Transfection of wild-type (WT) and all mutant constructs induced marked cell growth suppression compared with empty vector–transfected control (Fig. 4B). Among the constructs, Δ5 and Δ6 NDRG1 mutants induced the most prominent cell growth suppression (Fig. 4B; Supplementary Fig. S3A), along with reduced GSK3β expression and AKT, S6K, and S6 phosphorylation (Fig. 4C; Supplementary Fig. S3B). Following transfection, Δ5 and Δ6 mutants showed increased NDRG1 protein stability compared with the WT and other mutants (Fig. 4D and E; Supplementary Fig. S3B), suggesting that the C-terminal domain containing GSK3β and SGK phosphorylation sites might be involved in the stability of NDRG1. Indeed, NDRG1 degradation was almost completely blocked by the selective GSK3 inhibitor CT99021 (Fig. 4F and G). Collectively, GSK3β-mediated phosphorylation of NDRG1 negatively regulated the stability of NDRG1. NDRG1 overexpression and its dephosphorylation by GSK3β inhibition effectively suppressed GBM cell growth.

DIF-1 was identified in Dictyostelium discoideum as a putative morphogen required for differentiation of stalk cells (35). The antitumor effects of DIF-1 including in vivo antimetastatic effects have been reported in close context with Wnt/β-catenin signaling pathway (25, 36). Given the physiologic roles that NDRG1 plays in differentiation (8–11), we hypothesized that DIF-1 might alter NDRG1 signaling.

DIF-1 markedly increased NDRG1 expression in a time-dependent manner in U87MG cells (Fig. 5A), accompanied with morphologic transition from fibroblastic to a round shape and cell–cell adherent characteristic (Fig. 5B). Furthermore, DIF-1 inhibited cell growth (Fig. 5C), and flow cytometric analysis demonstrated that DIF-1 induced cell growth arrest at G₀–G₁-phase (Fig. 5D). DIF-1 suppressed cyclin D1, cyclin E, CDK4, and CDK2 expression in a time-dependent manner (Fig. 5E). Consistent with these data, DIF-1 increased NDRG1 expression and inhibited cell growth in another GBM cell line, U251 (Supplementary Fig. S4A and S4B). Furthermore, DIF-1 decreased GSK3β levels and suppressed phosphorylation of AKT, S6K, S6, and β-catenin in a dose- and time-dependent manner (Fig. 5F and G; Supplementary Fig. S4B). We next examined the effect of DIF-1 on cell apoptosis and DNA synthesis by using Annexin V-FITC/PI staining assay and BrdU incorporation assay, respectively. Treatment with DIF-1 did not alter apoptotic cell (Annexin V–positive) fractions in both U87MG and U251 cells (Supplementary Fig. S4C). In contrast, DIF-1 markedly decreased BrdU incorporation in both GBM cell lines, suggesting that DIF-1 inhibits DNA synthesis (Supplementary Fig. S4D). DIF-1 induced an increase in NDRG1 and a decrease in GSK3β levels without affecting the mRNA levels of these genes (Fig. 5H). We next examined the effect of DIF-1 on the stability of NDRG1 and GSK3β proteins. Compared with DMSO control, DIF-1 consistently enhanced NDRG1 expression levels and decreased pNDRG1 and GSK3β expression levels (Fig. 5I; Supplementary Fig. S4E). NDRG1 degraded with a half-life of >24 hours and approximately 6 hours in DIF-1–treated cells and control cells, respectively (Fig. 5I and J; Supplementary Fig. S4E). In contrast, DIF-1 considerably destabilized GSK3β with a half-life of approximately 3 hours compared with control cells, wherein the GSK3β protein exhibited a half-life of >24 hours (Fig. 5I and J; Supplementary Fig. S4E). The accelerated GSK3β degradation by DIF-1 was abrogated by MG132 (Fig. 5K), suggesting the involvement of proteasomal degradation in DIF-1–induced downregulation of GSK3β. However, NDRG1 degradation in untreated control cells was not recovered by MG132 (Fig. 5K).

Furthermore, we examined whether NDRG1 silencing affected the effects of DIF-1 on cell growth and downstream signaling. NDRG1 silencing counteracted the cell growth suppression caused by DIF-1 (Fig. 6A; Supplementary Fig. S5A). DIF-1 did not upregulate NDRG1 in the presence of siNDGR1 (Fig. 6B; Supplementary Fig. S5B). DIF-1 suppressed GSK3β and pAKT expression, which was reversed when NDRG1 was silenced (Fig. 6B; Supplementary Fig. S5B). Decreased phosphorylation of S6K and S6, however, was observed in both control and NDRG1-silenced cells exposed to DIF-1 (Fig. 6B; Supplementary Fig. S5B). We further examined whether GSK3β overexpression specifically rescued DIF-1–induced cell growth suppression of GBM cells. Transient transfection of exogenous GSK3β cDNA significantly annihilated DIF-1–induced cell growth suppression (Fig. 6C). Collectively, these results indicate that DIF-1–induced upregulation of NDRG1 and downregulation of GSK3β drive modulation in downstream signaling, leading to cell growth suppression in GBM.

Next, we examined whether DIF-1 could suppress cell growth in patient-derived GSCs (MGG8 and MGG23), which were cultured under glioma sphere culture conditions to maintain the original phenotypes and genotypes (24, 37, 38). DIF-1 suppressed both sphere formation and cell growth of MGG8 and MGG23 in a dose-dependent manner under sphere culture conditions (Fig. 6D and E). Moreover, DIF-1 enhanced NDRG1 expression, and suppressed expression of pNDRG1 (S330), GSK3β, pβ-catenin, pAKT, pS6K, S6K, and pS6, as well as several tested cell-cycle–related proteins (Fig. 6F). Furthermore, DIF-1 suppressed N-myc expression in MGG8 cells that harbor MYCN amplification (Fig. 6F).

Finally, we determined the antitumor effects of DIF-1 in xenograft therapeutic models in vivo. Oral administration of DIF-1 suppressed tumor growth in both volume (P < 0.05) and weight (P = 0.069) in a U87MG subcutaneous xenograft model (Fig. 7A and B). No difference in body weight was observed between control and DIF-1–treated mice (Fig. 7C). NDRG1 expression levels were higher and pNDRG1 expression levels were lower in DIF-1–treated tumors (n = 7) when compared with control tumors (n = 6; Fig. 7D). GSK3β expression levels were lower in DIF-1–treated tumors than in control tumors; moreover, phosphorylation of β-catenin, AKT, and S6 was decreased following treatment with DIF-1 (Fig. 7D).

We next examined the ability of DIF-1 to penetrate the blood–brain barrier (BBB) by HPLC and GC–MS analyses. It was previously reported that HPLC analysis determined the plasma levels of DIF-1 at 1 hour after single oral administration of DIF-1 (25). Our HPLC analysis demonstrated that DIF-1 was detected at 8.58 ± 2.15 μg/mL in plasma (n = 3), and at 1.48 ± 0.66 μg/mL (n = 3) in the brain at 1 hour after single oral administration of DIF-1 (300 mg/kg; Fig. 7E; Supplementary S6A). GC–MS analysis also detected the base peak (m/z = 201, C₉H₁₀³⁵ClO₃) corresponding to DIF-1 in the brain as well as in plasma after oral administration of DIF-1 (Supplementary Fig. S6B). We further found that NDRG1 expression levels were relatively higher in DIF-1–treated brain tissues (n = 8) than untreated control brain tissues (n = 7; Fig. 7F). These data strongly suggested that DIF-1 was able to cross the BBB and distribute into the brain. Next, we examined the antitumor effects of DIF-1 in orthotopic MGG8 and U87MG GBM xenograft models. In these orthotopic therapeutic models, oral administration of DIF-1 induced tumor growth suppression of MGG8 cells (P < 0.05) and U87MG cells (P = 0.098; Fig. 7G; Supplementary Fig. S7A).

In this study, we studied the effects of NDRG1 on GBM cell growth using NDRG1 silencing and overexpression, and showed that NDRG1 suppresses GBM cell growth through decreased GSK3β expression together with AKT and S6 inactivation. Furthermore, NDRG1 induced cell growth arrest at G₀–G₁ cell-cycle phase.

This work demonstrated that GSK3β is critically involved in NDRG1-dependent growth regulation in GBM. NDRG1 is an intrinsic cell growth suppressor of GBM through suppression of GSK3β expression via accelerated degradation, followed by suppression of AKT/S6 cell growth signaling and cell-cycle pathways (Fig. 7H). It is well known that GSK3β phosphorylation at its N-terminal serine residue by AKT inactivates GSK3β (39), and historically, GSK3β has been considered as a tumor suppressor due to its ability to phosphorylate prooncogenic molecules, c-Myc, cyclinD1, and β-catenin, targeting them for proteasomal degradation (40). On the other hand, there is increasing evidence that GSK3β acts as a positive regulator of cancer cell proliferation and contributes to unfavorable prognosis in various cancers (41). Indeed, it was reported that GSK3β positively regulated the PI3K/AKT/mTOR signaling pathway by phosphorylating AKT and inhibiting PTEN and TSC2 (39). In GBM, GSK3β is now considered to promote invasion, tumorigenesis, and resistance to therapy (42, 43). These observations on protumorigenic functions of GSK3β are in accord with our finding that suppression of GSK3β is the main molecular mechanism of NDRG1 suppression of GBM.

NDRG1 contains Thr/Ser sites at its C-terminal domain that are targeted by GSK3β and SGK for phosphorylation (see Fig. 4A; refs. 16, 29, 33, 34). In this study, we revealed that deletion mutants lacking the phosphorylation domain exhibited extended protein stability and mediated more prominent suppression of cell growth, GSK3β expression, and cell growth signaling. Consistent with a previous study (44), the GSK3β inhibitor, CT99021, inhibited NDRG1 phosphorylation and stabilized NDRG1. These results indicated that NDRG1 and GSK3β negatively regulated each other's protein expression by promoting degradation or instability. Although our work supports a vital role of GSK3β in phosphorylation and destabilization of NDRG1, proteasome inhibitor MG132 did not rescue NDRG1 degradation, suggesting mechanisms independent of the proteasome. Further studies are required to understand the mechanisms underlying how the NDRG1 protein degrades in GBM cells and how NDRG1 phosphorylation connects to its degradation. Furthermore, we determined whether GSK3β and/or SGK are essential for GBM cell growth driven by NDRG1 loss. Treatment with GSK3β inhibitor, CT99021 or tideglusib, selectively suppressed the cell growth enhancement and cell growth signaling activated by NDRG1 silencing. The SGK inhibitor, GSK650394, on the other hand, suppressed cell growth and signaling activation regardless of NDRG1 status. These NDRG1-selective effects of GSK3β support a crucial role that GSK3β plays in GBM cell growth promoted by NDRG1 loss. However, further studies are necessary to understand the specific role of SGK and GSK3β in NDRG1 loss–accelerated cell growth in connection with NDRG1 phosphorylation status.

Concerning prognostic significance of NDRG1 in patients with GBM, current (Fig. 1) and previous (20, 21) studies demonstrated that NDRG1 protein expression levels consistently correlated with longer survival. However, microarray data obtained from TCGA and REMBRANDT datasets (45, 46) showed that there is no significant correlation between NDRG1 mRNA levels and overall survival in patients with GBM (Supplementary Fig. S8A). This apparent discrepancy may indicate that protein levels rather than transcript levels better reflect the biological functions of NDRG1 in GBM.

DIF-1 has previously been reported to exhibit an inhibitory effect on cancer cell growth through inactivation of the Wnt/β-catenin signaling pathway and resulting reduction of cyclin D1 and c-Myc expression levels (25, 36, 47). This study demonstrates a novel mechanism of DIF-1 and its antitumor effects in GBM (Fig. 7H). DIF-1 potently enhanced NDRG1 protein levels, suggesting protein stabilization as a mechanism of DIF-1–induced NDRG1 upregulation (Fig. 7H). DIF-1 also accelerated degradation of GSK3β in a NDRG1-dependent manner, followed by inactivation of the AKT/S6 cell growth signaling pathways (Fig. 7H). Of note, DIF-1–induced growth suppression was significantly overcome by NDRG1 silencing, followed by a considerable restoration of GSK3β and pAKT expression. These data strongly suggest that DIF-1–induced NDRG1 upregulation directly mediated cell growth suppression. However, DIF-1 suppression of S6K and S6 phosphorylation was not recovered by NDRG1 silencing, indicating the presence of DIF-1 effects that are NDRG1 independent. Therefore, multiple mechanisms likely underlie DIF-1 antitumor effects, including enhancement of NDRG1 expression, inhibition of S6K/S6 phosphorylation, and other previously reported mechanisms. Finally, in vivo tumor growth of GBM cells was suppressed by DIF-1. Our in vitro and in vivo data consistently demonstrated that enhancement of NDRG1 expression, inhibition of GSK3β and NDRG1 phosphorylation, and inactivation of cell growth signaling pathways characterized the cell signaling effects of DIF-1 (Fig. 7H).

In addition, DIF-1 inhibited N-myc expression in MGG8 cells that harbor MYCN amplification (24, 38). The myc family of genes including MYCN, drive the development of nervous system and hematologic tumors (48), and NDRG1 was initially identified as a gene downregulated by N-myc or c-Myc (49). The myc gene is reportedly amplified in 4% of GBMs (4) and contributes to the maintenance of GBM cancer stem cells (50). The unique effects of DIF-1 on N-myc suppression and enhanced NDRG1 expression may contribute to the elimination of GBM stem cells.

We found that DIF-1 could cross the BBB and distribute into brain (Fig. 7E; Supplementary Fig. S6A and S6B), and oral administration of DIF-1 suppressed GBM tumor growth in both subcutaneous and orthotopic therapeutic models (Fig. 7A, B and G; Supplementary Fig. S7A). Furthermore, oral administration of DIF-1 induced its penetration through the BBB, enhanced the expression of NDRG1 in brain tissues, and inhibited GBM growth in the brain. Despite such encouraging anti-GBM effects of DIF-1 we showed in vitro and in vivo, its development as a novel therapeutic agent for GBM, and potentially other cancer, requires pharmacologic optimization. The half-life of DIF-1 in plasma concentration is shorter than 12 hours (25). In addition, high dose administration of DIF-1 was required to suppress tumor growth of GBM. Using DIF-1 as a prototype, development of derivative agents that have improved pharmacokinetics properties such as stability in the blood will allow clinical application of DIF-1–related compounds for cancer therapy.

In conclusion, we show that NDRG1 is a potent tumor suppressor in GBM. The bidirectional regulation of NDRG1 and GSK3β that we discovered herein represents an attractive therapeutic target for GBM. Agents that upregulate NDRG1 like DIF-1 could contribute to the development of potent therapeutic strategies against GBM.

No potential conflicts of interest were disclosed.

Conception and design: H. Ito, K. Watari, M. Kuwano, M. Ono

Development of methodology: K. Watari

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Ito, K. Watari, T. Shibata, Y. Murakami, Y. Nakahara

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Ito, T. Miyamoto, H. Wakimoto

Writing, review, and/or revision of the manuscript: H. Ito, K. Watari, Y. Murakami, H. Wakimoto, M. Kuwano, M. Ono

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H. Izumi, M. Ono

Study supervision: M. Kuwano, T. Abe

This research was supported by JSPS KAKENHI grant number JP17K07169 (K. Watari) and by St. Mary's Institute of Health Sciences (M. Ono).

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.