mTOR target NDRG1 confers MGMT-dependent resistance to alkylating chemotherapy

Cell Culture, Reagents, Transfections, and Treatment Regimens.

Details are provided in SI Appendix, Methods.

Plasmid-Based Knockdown of NDRG1.

To silence NDRG1 gene expression, two short-hairpin RNA (shRNA) sequences targeting different sites were cloned into the pSUPER-puro vector (11, 12). The sequences are provided in SI Appendix.

Lentiviral Preparations.

Lentiviral particles for the knockdown experiments were produced by cotransfecting psPAX2, pMD2.G (both Addgene plasmid 12259), and pLKO.1 constructs (TRC1; Sigma-Aldrich) in HEK293T cells using TransIT LT1 (Mirus Bio). Details are given in SI Appendix.

Cloning of NDRG1 Variants.

NDRG1 dephospho-variants were generated by using site-directed mutagenesis PCR and changing the codons of the phospho-sites Thr and Ser to Val and Ala, respectively. Two different phosphorylation-deficient variants, NDRG1-T346V-T356V and NDRG1-T328V-S330A-T346V-T356V, were generated.

Immunoblot.

Preparation of cell lysates and immunoblots were performed as described before (12). Antibodies are given in SI Appendix, Table S8.

Proximity Ligation Assay.

T98G and LN-229 cells (n = 2 × 10⁴) were confluently grown at O₂ of 1% (vol/vol) on coverslips for 72 h. Fixation was done using 30 min Cytofixx Pump Spray cell path and 30 min 4% (vol/vol) paraformaldehyde. For detection the Red Duolink In Situ Proximity Ligation Assay (PLA) Kit was performed according to manufacturer´s instructions with anti-NDRG1 polyclonal rabbit (Sigma-Aldrich) and mouse anti-MGMT (Life Technologies) applied for 12 h. Mounting was done with VECTASHIELD HardSet Mounting Medium with DAPI.

Animal Experiments, Image Processing, and Histology.

All animal work was approved by the governmental authorities (Regierungspräsidium Karlsruhe) and supervised by institutional animal protection officials in accordance with the National Institutes of Health guidelines given in Guide for the Care and Use of Laboratory Animals. Details are provided in SI Appendix.

Clinical Data.

All clinically related research in this manuscript is covered by the Ethical Vote for the UKT-05 (13), NOA-04 (14), and NOA-08 (15) trials.

Statistical Analysis.

Quantitative in vitro data are expressed as mean ± SD (SD), as indicated. All in vitro experiments reported here represent at least three independent replications performed in triplicate if not otherwise stated. Statistical significance was assessed by two-sided Student’s t test or ANOVA (Microsoft Excel). Values of P < 0.05 were considered significant and asterisked without correction for multiple statistical tests. Mouse glioma volumes were corrected for outliers using Grubbs’ test. Survival data were plotted by the Kaplan–Meier method and analyzed by the log-rank test. SigmaPlot Software was used for all analyses.

Hypoxia-Induced Alkylator but Not Radiotherapy Resistance in Malignant Gliomas Depends on NDRG1.

To identify factors that mediate hypoxia-induced alkylator resistance (SI Appendix, Fig. S1A), we screened human glioma cell lines for their response to alkylating chemotherapy in hypoxic conditions and subjected LN-229 glioma cells to a proteome screen (SI Appendix, Fig. S1B). This screen revealed hypoxia-specific up-regulation of seven and down-regulation of two proteins (SI Appendix, Table S1), of which up-regulated NDRG1 was further analyzed because (i) its up-regulation in hypoxia was unequivocally confirmed in all conducted assays (SI Appendix, Table S1), (ii) it had been implicated as a target of hypoxia (16) and a prognostic factor in other types of tumors (17), and (iii) it was found down-regulated in a transcriptome analysis following pharmaceutical mTOR inhibition with RAD001 (at www.ebi.ac.uk/arrayexpress under the accession number E-MEXP-3802). Hypoxia induced NDRG1 in all tested glioma cell lines (Fig. 1A). This was specific for NDRG1 because NDRG2–4 were not differentially regulated (Fig. 1B). In human glioma specimens, NDRG1 was associated with the degree of malignancy, and in glioblastomas it was prominently expressed in putatively hypoxic, perinecrotic areas (Fig. 1C and SI Appendix, Fig. S2A).

Knockdown of NDRG1 resulted in sensitization of established glioma cells and naturally highly NDRG1-expressing T269 and T325 primary glioma cells to TMZ (Fig. 1D and SI Appendix, Fig. S3A), indicating that NDRG1 mediates hypoxia-induced resistance to alkylating agents. Conversely, NDRG1-overexpressing cells showed a reduction in the TMZ-induced G2/M arrest (Fig. 1D and SI Appendix, Fig. S3B), which corresponded to a reduction in proliferation in vitro (Fig. 1E) and tumor growth in vivo (Fig. 1F), whereas proliferation or clonogenicity of glioma cells exposed to radiotherapy at 2 or 4 Gy remained unaffected (SI Appendix, Fig. S3D). It is notable that NDRG1-overexpressing cells not exposed to TMZ proliferated slower than the controls (Fig. 1E and SI Appendix, Fig. S4 D and F).

NDRG1 Is Transcriptionally Activated by Radiotherapy and Phosphorylated in the Course of TMZ Treatment.

Next, we analyzed the influence of therapeutic measures altering the tumor microenvironment on NDRG1 expression and activity. In vitro, irradiation (SI Appendix, Fig. S5A) but not TMZ induced NDRG1 mRNA and protein expression. In contrast to hypoxia, irradiation-induced NDRG1 expression was dependent on p53 expression (SI Appendix, Figs. S5B and S4 A and B) but was not impaired by HIF-1α or HIF-2α gene silencing (SI Appendix, Fig. S5C), indicating that hypoxia and irradiation use diverse signaling pathways to induce chemoresistance via NDRG1. Long-term exposure to TMZ led to an increased phosphorylation of NDRG1 at position T346 in surviving cells (SI Appendix, Fig. S5D). NDRG1 phosphorylation at T346 is associated with increased activity (18). Collectively, these data indicate that hypoxia and irradiation but not alkylating chemotherapy activate NDRG1 via distinct pathways resulting in resistance toward alkylating chemotherapy (SI Appendix, Fig. S6).

NDRG1 Is a Predictive Marker for Response to Alkylating Chemotherapy.

Next, we interrogated patient tumor tissue to recapitulate the relevance of inducible NDRG1 for therapy resistance. NDRG1 is induced at tumor recurrence (Fig. 2 and SI Appendix, Fig. S2B). As opposed to tumor tissue at diagnosis, NDRG1 expression in the treated tissues was not predominantly seen in perinecrotic areas anymore, but NDRG1 was widely expressed in glioblastoma cells even perivascularly opposed to the situation in the untreated tumors (Fig. 2B). NDRG1 expression in patients with low-grade gliomas progressing without interim genotoxic treatment also increased (SI Appendix, Fig. S2B). High NDRG1 levels at recurrence suggest a poor response to alkylating chemotherapy, but not to the antiangiogenic agent bevacizumab, in a small group of patients (SI Appendix, Table S2). A predictive role of NDRG1 for poor response to radiochemotherapy was suggested by post hoc NDRG1 expression analyses, which revealed that progression-free survival (PFS) and overall survival (OS) of glioblastoma patients from the UKT-05 trial (13) with moderate or high expression of NDRG1 was reduced compared with patients with low NDRG1-expressing tumors (Fig. 2C). This was supported by an analysis of the Repository for Molecular Brain Neoplasia Data (REMBRANDT) database, which revealed that the OS of glioblastoma patients with intratumoral up-regulation of NDRG1 was reduced compared with patients with intermediate or down-regulated expression of the NDRG1 transcript (Fig. 2D). To determine whether the prognostic impact of NDRG1 is specifically related to alkylating chemotherapy, tissue samples of the NOA-04 trial comparing primary radiotherapy with primary alkylating chemotherapy (14) were analyzed. NDRG1 expression was associated with reduced PFS in TMZ-treated patients but not with radiotherapy in this not preplanned subgroup analysis (Fig. 2E). Collectively, these data from several study patient populations suggest that NDRG1 expression in glioma tissue is associated with a poor response specifically to alkylating chemotherapy.

NDRG1 Is an Effector of the mTORC2/SGK1 Pathway.

As opposed to transcriptional regulation of NDRG1 by hypoxia and radiation, the signaling cascade that mediates NDRG1 phosphorylation at the T346 residue in TMZ-resistant glioma cells is unclear but most likely downstream of mTOR. Knockdown of the mTORC2 subunit rapamycin-insensitive companion of mTOR (RICTOR) but not the mTORC1 subunit regulatory associated protein of mTOR (RAPTOR) resulted in a reduction of NDRG1 phosphorylation and expression. Control of NDRG1 phosphorylation by RICTOR is independent of its transcriptional regulation of NDRG1 as knockdown of RICTOR in cells with exogenous NDRG1 expression, resulting in a reduction of NDRG1 phosphorylation and subsequent sensitization toward TMZ (Fig. 3A).

Studies in pancreatic cancer indicated that phosphorylation of NDRG1 at this specific site is mediated by the putative mTOR downstream effector, serum glucocorticoid-induced protein kinase 1 (SGK1) (19). Dexamethasone (DEX), which is an integral part in the treatment of malignant gliomas as a means to control edema, induced SGK1 transcription and increased the phosphorylation of NDRG1 at T346 (Fig. 3B). To test the hypothesis that treatment with DEX blunts the efficacy of alkylating chemotherapy in patients with glioblastoma, we performed a subgroup analysis of the NOA-08 trial. In this trial, elderly patients with malignant astrocytoma received radiotherapy or TMZ until progression and were treated with steroids at clinical discretion to treat or prevent vasogenic cerebral edema (15). Subgroup analyses allowed us to generate the hypothesis that steroid administration was associated with reduced PFS of patients treated with TMZ but not radiotherapy (Fig. 3C and SI Appendix, Table S3). An animal experiment with U87MG cells supports a negative impact of DEX on the efficacy of TMZ (Fig. 3D and SI Appendix, Fig. S6B). Pharmacological inhibition of SGK1 by EMD638683 resulted in decreased phosphorylation of T346 and overcame the NDRG1-mediated protection from TMZ (Fig. 3E). This is specific neither for TMZ nor for glioma cells as EMD638683 treatment also decreased constitutive NDRG1 phosphorylation in pancreatic, breast, colon, and ovarian cancer cells (SI Appendix, Fig. S7A). EMD638683-mediated sensitization toward chemotherapy was specific for alkylating agents as a sensitization was seen for lomustine in breast and ovarian cell lines (SI Appendix, Fig. S7B), but resistance neither to 5-fluorouracil nor to cisplatin was decreased (SI Appendix, Fig. S7 C and D).

NDRG1 Interacts with Three DNA Repair Enzymes and Promotes Protein Stability/Activity of MGMT.

Although described for the mediation of cisplatin resistance in glioma via activation of the mTORC2-mediated cascade, in the present paradigm, nuclear factor (NF)κB is not influenced by the NDRG status (SI Appendix, Fig. S8C). To unravel the molecular mechanisms by which NDRG1 prevents TMZ-induced cytotoxicity, a yeast two-hybrid screen for potential interaction partners was performed. Of 119 possible interaction partners (SI Appendix, Table S4), three proteins were involved in DNA repair: polynucleotide 3′-phosphatase, polynucleotide 5′-hydroxyl-kinase (PNKP), DNA-(apurinic or apyrimidinic site) lyase (APEX1), and MGMT (Fig. 4A). Interaction of these three proteins with NDRG1 was verified in a pull-down assay using HEK293 cells (Fig. 4B). The interaction of MGMT and NDRG1 was studied further because (i) in particular, MGMT has been implicated in mediating the resistance of gliomas to alkylating agents (15, 20); (ii) only MGMT expression correlated with TMZ resistance in gliomas cells; and (iii) knockdown of MGMT but not PNKP or APEX1 rendered glioma cells more susceptible toward the antiproliferative effects of TMZ (Fig. 4 C and D). Bimolecular fluorescence complementation (BiFC) assays confirmed a direct interaction of MGMT and NDRG1 in both HEK293T (Fig. 4E, Upper) and T98G glioma cells (SI Appendix, Fig. S10A) exogenously overexpressing the two proteins. PLAs also revealed this interaction for native T98G cells exposed to hypoxia in the nucleus (Fig. 4E, Lower). This interaction critically depended on SGK1 activity (SI Appendix, Fig. S11) and phospho-threonine or -serine SGK1 target sites in the C terminus of NDRG1 because mutation of T328, 330, 346, and 356 significantly reduced the NDRG1/MGMT interaction in BiFC assays (Fig. 4F). Forced expression of NDRG1 did not render cells resistant to TMZ when MGMT was knocked down, indicating that TMZ resistance mediated by NDRG1 is dependent on MGMT (Fig. 4D). Next, the hypothesis was tested that NDRG1 increases MGMT activity resulting in enhanced repair of DNA damage mediated by alkylating agents. Exogenous expression of NDRG1 resulted in augmented MGMT levels at 8 h (Fig. 5A, Upper) when RNA synthesis was blocked. Exposure of these cells to TMZ resulted in an expected decrease via depletion of MGMT expression at 4 h independent from NDRG1 but a faster recovery of MGMT in NDRG1-overexpressing cells (Fig. 5A, Lower). Indeed, exogenous expression of NDRG1 resulted in enhanced demethylation activity of MGMT in glioma cells exposed to TMZ [−0.25, 95% CI (−0.41 to −0.09), P = 0.0219] (Fig. 5B). To test whether this is relevant for treatment outcome in patients with glioblastoma we again performed post hoc subgroup analyses of the NOA-08 trial. Stratification of the TMZ-treated group of NOA-08 despite all limitations of this approach revealed that steroid administration resulted in a reduced PFS only in patients with a methylated MGMT promoter and hence inactive MGMT (Fig. 5C and SI Appendix, Table S5), suggesting that treatment of patients with DEX by inducing NDRG1 activity compromises the efficacy of alkylating chemotherapy independent of DNA repair activity, although there may be confounding factors triggering a decision for or against DEX treatment. In addition, in patients who failed radiotherapy in this trial and who received salvage chemotherapy with TMZ (SI Appendix, Table S5), MGMT activity was no longer predictive of response to chemotherapy, as opposed to the primary situation (Fig. 5C). Collectively, these data indicate that multiple treatment measures including treatment with steroids and radiotherapy by inducing NDRG1 can impair inherent susceptibility of MGMT-methylated glioma cells to the therapeutic effects of alkylating chemotherapy (Fig. 5D).