NDRG4 Is Required for Cell Cycle Progression and Survival in Glioblastoma Cells^*

Plasmids

pLKO.1 lentiviral nontargeting shRNA clone, NDRG1 targeting shRNA clone (sh-NDRG1-a: NM_006096.2–779s1c1), and NDRG4 targeting shRNA clones (sh-NDRG4-a: NM_020465.1–640s1c1; sh-NDRG4-b: NM_020465.1–769s1c1) were purchased from Sigma. pGIPZ lentiviral nontargeting shRNA clone and NDRG2 targeting shRNA clones (sh-NDRG2-a: RHS4430–99149037; sh-NDRG2-b: RHS4430–98851006) were purchased from Open Biosystems. pBabe-NDRG2 was generated by PCR amplification of NDRG2 from pCMV-HA-hNDRG2 (generously donated by Libo Yao and co-workers (15)) and subcloning into the BamHI and EcoRI sites of pBabe-puro. NDRG4 overexpression constructs were generated by PCR amplification of NDRG4(B) and NDRG4(H) from brain cDNA and subcloning into the BamHI site of pBabe-puro.

Generation of Overexpression and Knockdown Cells

Overexpression retrovirus was produced in 293T cells by co-transfection of pBabe constructs with the pCL10A-1 packaging vector. Knockdown lentivirus was produced in 293T cells by co-transfection with the psPAX2 and pVSVG packaging vectors. For both, medium was replenished 24 h after transfection, and virus was collected 48 h later. Supernatant was filtered, mixed with 4 μg/ml polybrene (Sigma), and added to target cells. For all experiments, cells were infected with virus for 48 h and selected with puromycin (1 μg/ml) for 72 h unless otherwise indicated.

Isolation of GBM CSCs and Nonstem Cells

Matched subpopulations of CD133⁺ cells (enriched for GBM CSCs) and CD133⁻ cells (nonstem cells) were isolated from human GBM xenografts or primary GBM tumors as previously described (19).

Cell Culture

GBM CSCs (CD133⁺) were cultured in Neurobasal Medium without Phenol Red (Invitrogen) containing the following additives: B-27 supplement without vitamin A (Invitrogen), 20 ng/ml human recombinant epidermal growth factor (Invitrogen), 20 ng/ml human recombinant basic fibroblast growth factor (Invitrogen), GlutaMAX (Invitrogen), MEM nonessential amino acids solution (Invitrogen), and sodium pyruvate (Invitrogen). GBM nonstem cells (CD133⁻) were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum. Normal primary human astrocytes were purchased from Lonza and cultured in the medium provided in the Astrocyte Medium BulletKit (CC-3186). All other GBM cell lines were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum.

Real-time Polymerase Chain Reaction

mRNA was isolated with the RNeasy Plus Mini Kit (Qiagen), and cDNA was synthesized using the SuperScript III First-Strand Synthesis System for RT-PCR (Invitrogen). Quantitative real-time PCR was performed using iQ SYBR Green Supermix (Bio-Rad), and results were normalized to β-2-microglobulin (B2M) levels. The following primers were used: NDRG4-F: GGAGGTTGTCTCTTTGGTCAAGGT, NDRG4-R: CTCATGACAGCAGCCACCAGAAT, B2M-F: GAGGTTTGAAGATGCCGCATT, B2M-R: TGTGGAGCAACCTGCTCAGATA.

Intracranial Tumor Formation Assay

Following infection and selection, 1,000 CD133⁺ GBM CSCs or 300,000 U251 cells were implanted into the right frontal lobes of 5-week-old, male, athymic BALB/c nu/nu mice under a Duke University Institutional Animal Care and Use Committee-approved protocol. Mice were maintained for 6 months or until the development of neurological symptoms of functional impairment that significantly impaired their quality of life (ataxia, lethargy, seizures, and inability to feed). Because prior xenograft studies have demonstrated that these signs develop shortly before animal death, mice were euthanized at this point. Brains of euthanized mice were collected, fixed in 4% paraformaldehyde, paraffin-embedded, and sectioned. H&E staining and Ki-67 staining were performed by the Duke Pathology Research Histology Laboratory. The logrank test was used for statistical analysis.

Cell Viability (MTS), Annexin V, and Caspase 3/7 Activity Assays

Following infection and selection, the following numbers of cells per well were plated: 1,000 cells in 96-well plates (cell viability), 200,000 cells in 24-well plates (annexin V), and 5,000 cells in 96-well plates (caspase 3/7 activity). Assays were then carried out with the following kits: CellTiter® 96 AQ_ueous Nonradioactive Cell Proliferation Assay (Promega), Annexin V-PE Apoptosis Detection Kit (BD Biosciences), and Apo-ONE Homogenous Caspase-3/7 Assay (Promega).

Neurosphere Formation Assay

Following infection and selection, spheres of CD133⁺ GBM CSCs were dissociated, and varying densities of cells were plated in 24-well plates. The percentage of wells with neurosphere formation and the average number of neurospheres per well were measured at the indicated times.

Flow Cytometry Cell Cycle Analysis

Prior to flow cytometry analysis, cells were fixed with 70% ethanol and then resuspended in 500 μl phosphate-buffered saline, 50 μl of PI (1 mg/ml), and 10 μl of RNase (20 mg/ml). For synchronization experiments, cells were first treated with 2 mm thymidine for 16 h. Cells were then washed, infected with knockdown lentivirus for 9 h, and treated with 2 mm thymidine again for 14 h before release. For spindle checkpoint experiments, cells were infected with knockdown lentivirus for 24 h and then treated for 48 h with 100 ng/ml nocodazole.

Western Analysis and Nucleocytoplasmic Fractionation

Nucleocytoplasmic fractionation was performed as described previously (20). For all other experiments, cells were lysed in ULB as described previously (21). Western analysis was performed using the following antibodies: phosphohistone H3-Ser-10 (Upstate, 06–570), γ-tubulin (Sigma, T6557), NDRG4 (Sigma, HPA015313), NDRG1 (Sigma, HPA006881), NDRG2 (Sigma, HPA002896), XIAP (Cell Signaling, 2045), caspase-3 (Cell Signaling, 9662), caspase-9 (Cell Signaling, 9508), cyclin D1 (Cell Signaling, 2926), cyclin E (Cell Signaling, 4129), Smad1 (Abcam, ab33902), α-tubulin (Abcam, ab6160), p27^Kip1 (Santa Cruz Biotechnology, sc-528), and lamin A/C (Santa Cruz Biotechnology, sc-7293).

Immunohistochemistry Analysis

Freshly frozen human glioma surgical resection samples from the Brain Tumor Center Tissue Bank at Duke University were processed, and 10 micron sections were mounted on glass slides in accordance with the Duke University Medical Center Institutional Review Board. Staining was then performed based on the protocol provided with R.T.U. VECTASTAIN Elite ABC Reagent (Vector Laboratories). Background Buster (Innovex Biosciences) was used for 30 min, NDRG4 antibody (1:25 dilution; Sigma, HPA015313) and biotinylated secondary antibody (10 μg/ml, Vector Laboratories) were used for 1 h each, and the Liquid DAB/Substrate System (Innovex Biosciences) was used for 3 min. Ki-67 staining was performed by the Duke Pathology Research Histology Laboratory.

Single Cell Gel Electrophoresis (Comet Assay)

Cells were infected with knockdown or control lentivirus for 48 h, and comet assays were performed as described previously (22).

NDRG4 Expression Is Increased in GBM

To understand the role of NDRG4 in astrocytes and GBM cells and to determine the functional similarities and differences between NDRG4 and NDRG2, we began by studying NDRG4 expression in human tumor samples. Immunohistochemical (IHC) analysis of normal human cortex and GBM samples revealed increased NDRG4 protein expression in GBM (Fig. 1A and supplemental Fig. S1). Both normal samples tested had undetectable levels of NDRG4, but five out of six GBM samples showed a subset of neoplastic cells with moderate to high intensity cytoplasmic staining (supplemental Fig. S1). In the sample with the greatest degree of positivity, ∼10–15% of cells showed strong staining for NDRG4 (Fig. 1A). This finding is in contrast to the reported lower expression level of NDRG2 in neoplastic glioma cells relative to normal nonneoplastic glial cells, an observation fitting with the presumed role of NDRG2 as a tumor suppressor in this context (11, 13, 14).

Our IHC analysis observations are further supported by an independent antibody-based proteomic analysis that showed increased NDRG4 expression in human glioma samples compared with nonneoplastic cortical brain tissue (11). Furthermore, a second NDRG4 antibody used in this independent IHC analysis of normal human tissues and cancer tissues yielded staining patterns similar to the those produced by the antibody used in Fig. 1A (11), suggesting that the staining pattern we observed is specifically due to increased NDRG4 expression. However, to confirm the increased expression of NDRG4 in GBM, we next examined NDRG4 mRNA levels in normal primary human astrocytes and cells derived from human GBM xenografts. Using primers specifically targeting an RNA sequence common to all known isoforms of NDRG4, we measured NDRG4 mRNA levels by real-time RT-PCR analysis in two independent lots of primary astrocytes and in cultured cells derived from three different GBM xenografts. Consistent with our IHC results, we found that NDRG4 expression was 5–20-fold higher in cultured GBM xenograft cells than in normal astrocytes (Fig. 1B).

Given that GBM tumors contain heterogeneous subpopulations of cells, we next tested whether the subset of cells showing strong NDRG4 staining represented the highly tumorigenic GBM CSC population. To test this, we examined the relative expression level of NDRG4 in cultured GBM CSC-enriched populations (CD133⁺) and corresponding nonstem cell populations (CD133⁻) directly derived from GBM xenograft samples. The expression between these cell populations was similar (Fig. 1B). Thus, NDRG4 expression in GBM cells is not CSC-restricted.

NDRG4 Expression in GBM Is Cytoplasmic and Peaks during the G₁ and S Phases

Neoplastic GBM cells that were positive for NDRG4 expression by IHC analysis appeared to show strong cytoplasmic NDRG4 staining (Fig. 1A and supplemental Fig. S1), which is consistent with what has been reported in other cell types and in other species (2, 3). To confirm this localization in GBM cells and to further characterize the expression pattern of NDRG4, we determined its subcellular localization in U251 GBM cells by fractionation and Western analysis. Similar to what we observed through IHC analysis, NDRG4 localized to the cytoplasmic fraction (Fig. 1C). Furthermore, it remained cytoplasmic even upon treatment with the nuclear export inhibitor leptomycin B.

To begin to understand the biological functions of NDRG4, we next partially synchronized U251 cells with a double thymidine block and measured NDRG4 expression during different stages of the cell cycle. We found that NDRG4 levels fluctuate during cell cycle progression. At 9–12 h after release from the block, when the highest percentage of G₂/M cells was observed (∼50%), NDRG4 mRNA levels were lowest (Fig. 1D). In contrast, NDRG4 mRNA levels peaked at 0–3 h and 18–24 h after release, and NDRG4 protein expression peaked slightly later in an expression pattern opposing that of the mitotic marker phosphohistone H3. These time points correlate with progression through G₁ phase and entry and progression through S phase. This is supported by our IHC analysis, which showed that similar percentages of GBM cells stain positive for NDRG4 and the proliferation marker Ki-67 (∼10–15% and 15–20%, respectively; Fig. 1A, supplemental Fig. S1, and supplemental Fig. S2). It is also consistent with our observation that NDRG4 mRNA levels decrease with passage number in primary human astrocytes as they stop growing: NDRG4 expression levels at passage 7 were only 20% of what they were at passage 1 (supplemental Fig. S3), presumably due to a decreased percentage of astrocytes progressing through the G₁/S transition. A similar decrease in NDRG2 expression was observed with increasing passage. Taken together, these results prompted the additional comparative studies of NDRG2 and NDRG4 described below and indicated that NDRG4 is expressed in a cell cycle-specific manner that implicates it in cell cycle progression.

NDRG4 Is Required for GBM Cell Viability

To test the hypothesis that NDRG4 is involved in GBM cell cycle progression, we knocked down NDRG4 expression with two lentiviral shRNA constructs (sh-NDRG4-a and sh-NDRG4-b) that target different regions of the NDRG4 transcript, and we then assessed cell viability by MTS assay in the resulting knockdown cell lines. Knockdown of NDRG4 expression in U251 GBM cells dramatically decreased their viability (Fig. 2A). The magnitude of the reduction in viability was proportional to the efficacy of the two shRNA constructs: sh-NDRG4-a reduced NDRG4 expression and cell viability by ∼60%, whereas sh-NDRG4-b reduced each by ∼95%.

These dose-dependent effects strongly suggest that the decreased viability we observed was a specific effect caused by loss of NDRG4 expression. This is further supported by our finding that cell viability was unaffected by control, nontargeting shRNAs (Fig. 2A). Similarly, shRNAs that target NDRG2 did not affect cell viability despite reducing NDRG2 protein levels by more than 90% (Fig. 2B). Moreover, NDRG1 and NDRG2 levels were not affected by NDRG4 knockdown (supplemental Fig. S4), further indicating that the effects on cell viability were specifically due to NDRG4 knockdown.

To determine if the decreased cell viability caused by NDRG4 knockdown is a robust response, we next knocked down NDRG4 expression in several additional GBM cell lines as well as primary astrocytes, the presumed cell of origination for GBM. In every GBM cell line tested, NDRG4 knockdown reduced cell viability by 40–99% (supplemental Fig. S5). Moreover, the NDRG4 knockdown-induced decrease in viability was not restricted to transformed or neoplastic cells, as 90–95% knockdown of NDRG4 decreased the viability of primary astrocytes by 60% (Fig. 2C). Thus, the reduction in cell viability caused by NDRG4 knockdown is a robust response that is not unique to the U251 cell line model system.

This robust requirement for the presence of NDRG4 is in contrast to the function of NDRG2 in GBM cells. While NDRG4 expression is essential for U251 cell viability (Fig. 2A), reduction of NDRG2 expression by ∼95% did not affect the viability of U251 cells (Fig. 2B). Severalfold overexpression of NDRG2, however, reduced U251 cell viability by 30% at day 4 (Fig. 3A), which is consistent with previous findings from NDRG2 overexpression studies in U373 GBM cells (15). In contrast, severalfold overexpression of two NDRG4 isoforms that have been demonstrated to be expressed in the brain, NDRG4(B) and NDRG4(H), did not affect U251 cell viability (Fig. 3B). Thus, despite the strong similarity between NDRG2 and NDRG4 in sequence and in function in other cell types, our results indicate that these two genes have divergent functions in GBM cells.

Loss of NDRG4 in GBM Cells Results in G₁ Cell Cycle Arrest and Subsequent Apoptosis

Loss of cell viability as determined by MTS assay can have multiple etiologies including reduced metabolic activity, cell cycle arrest, and apoptosis. To differentiate between the possible etiologies, we knocked down NDRG4 expression in U251 cells that had been partially synchronized in early S phase by a double thymidine block, and we then measured the percentage of cells in each phase of the cell cycle after release. While control cells completed two cell cycles in a 36 h time window, cells with reduced NDRG4 expression completed the first cycle but stopped cycling in the G₁ phase of the second cycle (Fig. 4A). This indicated that NDRG4 is required for progression through G₁. To further confirm this, we did cell cycle analysis over an NDRG4 knockdown time course. A comparison of U251 cells at day 1 and day 1.5 after infection with sh-NDRG4-b knockdown virus revealed an increase in the percentage of cells in G₁ at day 1.5 (from 69 to 84%) and a corresponding decrease in the percentage of cells in S and G₂/M (from 30 to 15%; Fig. 4B). Moreover, we detected a number of NDRG4 knockdown-induced molecular events associated with the regulation of G₁ phase progression: p27^Kip1 levels increased and cyclin D1 levels decreased starting at day 1 after infection with the sh-NDRG4-b knockdown virus (Fig. 4C).

NDRG4 knockdown-induced G₁ arrest was followed by several apoptotic events. The initial accumulation of cells in G₁ that was first observed at day 1.5 after infection with sh-NDRG4-b knockdown virus was followed by a slight decrease in the G₁ population of cells at day 4 (92% at day 3.5 and 85% at day 4; Fig. 4B). This corresponded with an increase in the sub-G₀/G₁ population (2% at day 3.5 and 9% at day 4), which is suggestive of increased apoptosis. In support of this, NDRG4 knockdown caused a 5-fold increase over control in the percentage of annexin V-positive cells at day 5 (Fig. 4D). Furthermore, a number of molecular events associated with apoptosis were observed in NDRG4 knockdown cells but not control cells: XIAP and survivin levels decreased starting at day 2 after infection with sh-NDRG4-b virus, and this was followed by an increase in caspase cleavage at day 3 (Fig. 4E). The elevation in caspase activity was confirmed by fluorescent caspase 3/7 activity assays, which revealed an 11-fold increase in caspase activity in response to knockdown with the sh-NDRG4-b virus (Fig. 4F). The sh-NDRG4-a construct produced similar, although attenuated, effects consistent with its less pronounced effect on NDRG4 expression.

Loss of NDRG4 Does Not Induce DNA Damage or Affect Mitotic Progression

During the G₁ phase of the cell cycle, safeguards exist to ensure that genomic DNA is intact and ready for replication. Consequently, progression through G₁ can be stopped by factors that cause a loss of genomic DNA integrity due to improper mitosis. Because another NDRG family member, NDRG1, is required for correct mitotic spindle formation and progression through mitosis in mammary epithelial cells (23), we next used a mitotic checkpoint assay to investigate whether NDRG4 functions in an analogous manner in GBM cells. Treatment of control cells with nocodazole, a chemotherapeutic inhibitor of microtubule formation that interferes with the formation of the mitotic spindle, resulted in an increased polyploid population of cells manifesting as 4N and 8N cells (46 and 16% of cells, respectively, as compared with 10 and 1% of untreated cells; Fig. 5B). The presence of polyploidy rather than mitotic arrest is likely due to the mutated TP53 gene present in U251 cells. As expected, NDRG1 knockdown exacerbated the detrimental effect of nocodazole. The combined result of nocodazole treatment and a loss of NDRG1 was a dramatic increase in apoptosis manifesting as a large sub-G₀/G₁ population of cells (31% of cells as compared with 12% in the control nocodazole-treated population; Fig. 5, A and B). In contrast, NDRG4 knockdown in combination with nocodazole treatment did not cause pronounced apoptosis or polyploidy, but rather led to the accumulation of 2N and 4N cells (50 and 32% of cells, respectively, as compared with 10 and 46% in the control nocodazole-treated population), presumably due to an inability of the cells to enter S phase (Fig. 5, A and B). The differing effects of NDRG1 and NDRG4 knockdown in this context suggest that NDRG1 and NDRG4 have different biological functions and that the G₁ block and subsequent apoptosis caused by NDRG4 knockdown are not due to chromosomal defects acquired during mitosis.

Another common cause of G₁ arrest is DNA damage. However, DNA damage is also unlikely to be the cause of the NDRG4 knockdown-induced G₁ arrest. While DNA double-strand break-inducing ionizing radiation caused formation of “comet tails” characteristic of DNA damage in U251 cells (Fig. 5C), NDRG4 knockdown did not induce any noticeable “comet tail” formation in these comet assays (Fig. 5C). These results, in combination with the exclusive cytoplasmic localization of NDRG4, strongly suggest that DNA damage and mitotic defects are not the cause of G₁ arrest and apoptosis in NDRG4 knockdown cells.

NDRG4 Is Required for the Viability and Self-renewal of GBM CSC-enriched CD133⁺ Cells and Nonstem Cell-enriched CD133⁻ Cells

GBM tumors are heterogeneous populations of cells that include GBM CSCs, which are characterized by the ability to self-renew and recapitulate their parental tumors in immunocompromised mice (24–26). To determine if NDRG4 behaves differently in GBM CSC-enriched populations (CD133⁺) and corresponding nonstem cell-enriched populations (CD133⁻), we isolated these subpopulations of cells from human GBM xenografts derived from primary GBMs. The GBM CSC-enriched cell lines were previously validated by the expression of stem cell markers and were characterized as capable of differentiating into multiple brain cell lineages and potently forming tumors that recapitulate the heterogeneity of their parental tumors (19, 27). Using these GBM CSC-enriched cells and corresponding nonstem cell-enriched cells, we found that the role of NDRG4 in cell cycle progression and cell viability in both populations is the same as in the immortalized GBM cell line model systems.

In GBM CSC-enriched cells, NDRG4 knockdown decreased viability in a dose-dependent manner. Consistent with the level of knockdown produced by each, sh-NDRG4-a decreased cell viability by 60% at day 4 and sh-NDRG4-b decreased cell viability by 80% (Fig. 6A). Similar results were seen with the neurosphere formation assay (Fig. 6, B and C). When equivalent numbers of cells were plated, the number of neurospheres formed per well decreased with NDRG4 targeting (65% decrease with sh-NDRG4-a and 90% decrease with sh-NDRG4-b at day 4 after plating cells; Fig. 6B). Depending on the extent of NDRG4 knockdown, there was also a delay in the time necessary to form neurospheres or a decrease in the percentage of wells with neurosphere formation (Fig. 6C).

Just as NDRG4 knockdown decreased the viability of CSC-enriched CD133⁺ GBM cells, it also decreased the viability of nonstem cell-enriched CD133⁻ GBM cells. A 40–60% decrease in viability was observed at day 4 with both NDRG4 knockdown constructs, which reduced NDRG4 expression by 65 and 90% (Fig. 6D). Together, these data indicate that NDRG4 is required for the survival and self-renewal potential of both GBM CSC-enriched populations and GBM nonstem cell populations.

Knockdown of NDRG4 Decreases Growth of GBM Tumor Xenografts in Immunocompromised Mice

To confirm that NDRG4 does indeed behave differently than NDRG2 in the context of GBM, we assessed whether the decreased viability resulting from NDRG4 knockdown in cell culture translates into similar effects on GBM cells in vivo. To address this, we used a mouse tumor model system that best approximates the microenvironment of a GBM tumor. Prior to NDRG4 knockdown-induced apoptosis, we implanted U251 control and NDRG4 knockdown cells intracranially into immunocompromised mice. Consistent with our cell culture data, there was a statistically significant decrease and delay in the development of functional impairment caused by brain tumor formation in mice injected with either population of NDRG4 knockdown cells compared with the mice injected with control cells (p < .001 for both sh-NDRG4-a and -b; Fig. 7A).

Comparable results were seen with GBM CSC-enriched cells. As with the U251 cells, control and NDRG4 knockdown GBM CSC-enriched cells were implanted intracranially into immunocompromised mice. We found a statistically significant delay in the development of functional impairment caused by tumor formation in the mice injected with sh-NDRG4-a knockdown cells compared with the mice injected with control cells (p = .002, Fig. 7B). Tumor formation was confirmed by H&E staining and Ki-67 staining of brains resected from sacrificed mice. Representative images are shown in Fig. 7C. Collectively, these data reveal that NDRG4 is required for the viability of GBM cell lines in vitro and in vivo. Thus, NDRG4 diverges in function from its family member NDRG2 and does not act as a tumor suppressor in GBM.

NDRG4 Is Required for Cell Cycle Progression and the Survival of GBM Cells

NDRG4 expression fluctuates throughout the cell cycle, with maximal expression occurring during the G₁ and S phases. Loss of NDRG4 through targeted knockdown leads to cell cycle arrest in G₁ with an associated increase in p27^Kip1 levels and a decrease in cyclin D1 but not cyclin E levels. Taken together, these findings suggest that NDRG4 plays a critical role in the transition between the G₁ and S phases of the cell cycle.

Because cyclin D-CDK complexes are important for progression through early G₁ and cyclin E-CDK complexes are important for progression through late G₁ (28), G₁ arrest caused by NDRG4 knockdown likely occurs in early G₁ phase due to reduced cyclin D1 levels. However, increased p27^Kip1 expression has been demonstrated to prevent activation of cyclin E-CDK2 complexes (29), so NDRG4 knockdown likely also affects progression through late G₁. Although the precise molecular function of NDRG4 at the G₁ to S transition remains to be determined, we detected no DNA damage or defects in mitosis following NDRG4 knockdown, so it is unlikely that DNA damage or chromosome disruption is causing NDRG4 knockdown-induced cell cycle arrest. Furthermore, loss of NDGR4 likely has broader effects than simply preventing the start of S phase through CDKI activation or the regulation of critical cell cycle proteins, as NDRG4 knockdown cells ultimately undergo apoptosis.

The apoptosis caused by NDRG4 knockdown is associated with a decrease in the levels of both XIAP and survivin. XIAP and survivin are required for glioma growth and tumor formation (30–32), and high expression of these proteins is associated with poor prognosis in GBM patients (33–35). Because of the importance of these antiapoptotic proteins, inhibitors or knockdown reagents targeting XIAP can directly induce apoptosis in GBM cell lines and can act synergistically with other chemical inducers of apoptosis to cause death in cells with significant intrinsic resistance to toxic agents (30, 32, 36, 37). Given the critical role of these proteins in glioma cell survival, the loss of XIAP and survivin protein expression in NDRG4 knockdown cells may underlie the observed apoptotic effects.

The Role of NDRG4 in GBM Diverges from Previously Characterized Functions of the NDRG Family

Although our findings are consistent with recent findings in zebrafish embryos, where NDRG4 expression corresponds with proliferative stages of heart development and NDRG4 function is important for myocyte proliferation (6), the requirement of NDRG4 for the viability of GBM tumor cells and astrocytes differs from the pro-differentiation and growth inhibitory roles of NDRG4 in neurons and pancreatic cells. In nerve growth factor (NGF)-treated PC12 neuronal cells, NDRG4 knockdown reduces AP-1 activation and inhibits neurite extension (17). Overexpression of NDRG4 in the same cells enhances NGF-induced phosphorylation of MEK and ERK, presumably also enhancing differentiation (16). Similarly, NDRG4 expression is required for glucagon-like-polypeptide-1-induced growth inhibition and endocrine differentiation in rat pancreatic cells (38).

The role of NDRG4 in GBM also differs from the function of NDRG2 in the same context. Within the NDRG family, NDRG4 is most similar to NDRG2, with ∼65% identity at the amino acid sequence level (10). Accordingly, these two NDRG family members have similar functions in some cell types within the brain. In GBM cells, however, the opposite is true. Although the mRNA expression levels of NDRG2 and NDRG4 are weakly positively correlated in GBM samples (supplemental Fig. S6) (39), NDRG2 and NDRG4 have opposing protein expression patterns when comparing normal brain tissue to GBM tissue, and NDRG4 appears to have a function that diverges significantly from NDRG2. While overexpression of NDRG2 reduces GBM cell viability, knockdown of NDRG2 expression in the same cells has no effect. In contrast, overexpression of NDRG4(B) and NDRG4(H) does not affect GBM cell viability, while NDRG4 knockdown decreases the viability of the same cells.

When these opposing functions of NDRG4 and NDRG2 in GBM are taken in the context of the NDRG family in general, it is the function of NDRG2 that is consistent with the prevailing theories regarding the roles of these genes as tumor suppressors. Just as NDRG2 expression levels are reduced in GBM, decreased NDRG2 expression is observed in colorectal adenocarcinoma (40, 41), and thyroid (42), hepatocellular (43), gastric (44), and clear cell renal (45) carcinomas. Consistent with these expression patterns, NDRG2 has been reported to suppress cellular proliferation (44, 46), suppress invasion and metastasis (43), and be required for apoptotic pathways including Fas-mediated cell death (44) and p53-mediated apoptosis (47). The expression patterns and functions of NDRG1 in cancer mostly mirror those of NDRG2. NDRG1 expression is reduced in many cancer types (48–52), and NDRG1 can inhibit cellular proliferation (53) and tumor metastasis (49) and is an effector for both p53-mediated apoptosis (54) and the tumor suppressive effects of PTEN (55). NDRG3 has not been studied as extensively, but recent findings in prostate cancer suggest that it may behave differently than NDRG1 and NDRG2, as it promotes cell growth in this context (56). Taken together, however, the available evidence indicates that a major function of NDRG family members is to impede cancer progression through the regulation of proliferation, apoptosis, and metastasis (57, 58). Thus, our discovery that NDRG4 is required for the viability of GBM cells and astrocytes challenges the view that NDRG family members are fundamentally tumor suppressive.

In summary, our study establishes a novel role for NDRG4 that is in contrast to the known functions of NDRG2 and its other family members. Specifically, we show that the presence of NDRG4 is required for the viability of GBM cell populations and primary astrocytes. In GBM cells, we demonstrate that the presence of NDRG4 is required for continued progression through the cell cycle and ultimately for survival. As we learn more about precisely how NDRG4 functions and how it is regulated in GBM cells and astrocytes, we will be able to further dissect apart the diverse and divergent functions of the NDRG family.