Epigenetic Activation of TUSC3 Sensitizes Glioblastoma to Temozolomide Independent of MGMT Promoter Methylation Status

Using 5-Aza as our DNMT1 inhibitor, we first tested 5-Aza inhibition efficiency by assessing DNMT1 protein expression and DNMT1 activity. The Western blot result indicated that 5-Aza successfully suppressed DNMT1 protein expression in both MGMT-M and MGMT-UM GSCs (Figure S1A). Similar to Western blot, the measurement of DNMT1 activity demonstrated that DNMT1 enzyme activity was significantly suppressed by 5-Aza in all GSCs (Figure S1B). Interestingly, we found that 5-Aza dramatically reduced cell viabilities in MGMT-M GSCs when treated in combination with TMZ, while there was no significant impact on MGMT-UM GSCs (Figure S1C). This result suggested that 5-Aza may sensitize MGMT-M but not MGMT-UM GSCs to TMZ treatment. We further tested GSCs IC-50 against TMZ to validate the above observation. The results indicated that 5-Aza significantly decreased TMZ IC-50 in MGMT-M GSCs but had no impact on MGMT-UM GSCs (Figure 1A). Our findings suggested that 5-Aza could sensitize GSCs to TMZ treatment in an MGMT-promoter-methylation-status-dependent manner. Indeed, we observed a strong synergistic effect between TMZ and 5-Aza in MGMT-M GSC (Figure 1B, Loewe synergy score = 17.154). Supporting these findings, dual treatment with TMZ and 5-Aza also significantly suppress GSCs proliferation in MGMT-M GSCs compared to TMZ treatment alone (Figure S2A). No significant difference in proliferation was found in MGMT-UM GSCs (Figure S2B). Meanwhile, 5-Aza showed no impact on cell migration and invasion in either MGMT-M or MGMT-UM GSCs (Figure S3). To further investigate and validate the in vitro therapeutic synergies between TMZ and 5-Aza, we took advantage of orthotopic xenograft by intracranial implantation of GSCs. The animals were treated with vehicle control, TMZ only, 5-Aza only and TMZ combining 5-Aza. The results showed that 5-Aza alone did not significantly impact animal overall survival in animals bearing either MGMT-M (Figure 1C) or MGMT-UM (Figure S4) GSCs. However, the TMZ and 5-Aza combination significantly prolonged animal survival when bearing MGMT-M GSCs (Figure 1C), while the combination had no impact on the survival of animals bearing MGMT-UM GSC (Figure S4). Through our observation, it does appear that 5-Aza impacts GSCs sensitivity to TMZ treatment relative to MGMT promoter methylation status. However, 5-Aza had no influence either on MGMT mRNA expression in all the GSCs (Figure S5A) or MGMT promoter methylation status (Figure S5B,C). Since 5-Aza exhibited differential TMZ sensitivity reprogramming effects in MGMT-M versus MGMT-UM GSCs but no differential plasticity to MGMT status, we hypothesized that this phenomenon was due to epigenetic activation of specific gene(s) in MGMT-M GSCs by 5-Aza, which were not activated in MGMT-UM GSCs. We then performed microarray in MGMT-M (01 and 02) and MGMT-UM (01 and 02) GSCs treated with either DMSO control or 5-Aza to determine which gene(s) drove this phenomenon. According to the analysis, there were 42 genes that were activated by 5-Aza in MGMT-M GSCs but were not activated in MGMT-UM GSCs (Figure 1D, Red dotted circle). Further, we functionally screened these 42 gene candidates by siRNA silencing and viability assessment to determine which genes were closely associated with TMZ sensitization. The MGMT-M GSCs were treated with TMZ and 5-Aza and together with indicated siRNAs for all these 42 genes. We looked for specific gene(s) that could reverse the effects of combinatorial treatments with TMZ and 5-Aza. Notably, we found that 5-Aza failed to sensitize MGMT-M GSC to TMZ treatment only when the Tumor Suppressor Candidate 3 (TUSC3) gene was silenced, suggesting that TUSC3 could be the critical gene that was activated during 5-Aza treatment and was responsible for TMZ sensitization (Figure 1E, Purple dotted rectangle). Evaluation of TUSC3 mRNA level in MGMT-M GSCs confirmed and further demonstrated that 5-Aza treatment could successfully activate TUSC3 expression in MGMT-M GSCs (Figure 1F).

We further looked into the clinical relevance of TUSC3 in GBM patients by analyzing the TCGA dataset. Full mRNA expression combined from microarray (Agilent) and RNA-seq (Illumina HiSeq RNASeq V2) was downloaded from https://gdc.cancer.gov/about-data/publications/lgggbm_2016↗ (accessed on 31 October 2022) and the TUSC3 expression level was extracted for the GBM samples [31]. Overall survival (OS) for the TCGA GBM samples was retrieved from the publication by Liu et al. [32]. A log-rank p-value was calculated, and median cut was applied in all survival analysis. The TCGA dataset analysis showed that high TUSC3 expression confers a significant survival benefit (Figure 2A, p = 0.0021). This survival benefit provides a strategy to efficaciously target GBM, taking advantage of TUSC3 epigenetic reactivation. Moreover, the TCGA GBM dataset analyses revealed a lower expression profile of TUSC3 in GBM patient tumors compared to normal brain tissues (Figure 2B). TCGA RNA-seq data were downloaded from https://gdc.cancer.gov/about-data/publications/pancanatlas↗ (accessed on 31 October 2022) and log2-transformed and the TUSC3 expression for the GBM samples was extracted. Interestingly, TUSC3 expression is negatively correlated with promoter methylation level, which confirmed that TUSC3 expression is epigenetically regulated in GBM (Figure 2B). To further confirm these observations, we have utilized 16 GBM-patient-derived samples to perform Illumina methylation 850 EPIC arrays (Illumina, San Diego, USA). TUSC3 had large variability in the methylation level across all samples and CpG-probes, but TUSC3 promoter was significantly highly methylated in the TSS1500 and TSS200 regions (Figure 2C).

To further investigate the role of TUSC3 in TMZ sensitization, we utilized wild-type (WT) and TUSC3-stable overexpressing GSCs (TUSC3-OE). TUSC3 overexpression in both MGMT-M and MGMT-UM GSCs was confirmed by Western blot for protein level and real-time qPCR for mRNA level (Figure 3A). We first evaluated IC-50 against TMZ with wild-type and TUSC3-OE GSCs. As shown, TUSC3 overexpression significantly decreased TMZ IC-50 values in both MGMT-M and MGMT-UM GSCs compared to wild-type (Figure 3B). Furthermore, TUSC3 overexpression significantly inhibited GSCs proliferation in both MGMT-M GSCs and MGMT-UM GSCs when treated with TMZ (Figure S6A,B) compared to wild-type. These observations demonstrated that reactivation of TUSC3 could synergize with TMZ treatment in GSCs. We further established wild-type control (WT, with control sgRNA) and TUSC3 knockout (TUSC3-KO, with TUSC3 sgRNA) GSCs by CRISPR/Cas9 system in both MGMT-M and MGMT-UM GSCs (Figure 3C). We then performed a TUSC3 rescue experiment (with empty vector-EV and TUSC3 expressing construct) in wild-type and TUSC3 knockout GSCs (TUSC3-KO). Conversely, TUSC3 knockout led to an increase in TMZ IC-50 (Figure 3D), which was significantly reversed/downregulated by TUSC3 rescue (Figure 3D). This strongly suggested that 5-Aza relies on epigenetic reactivation of TUSC3 in order to sensitize GSCs to TMZ treatment either in MGMT-M or MGMT-UM GSCs.

According to our observation, TUSC3 activation is critical for sensitizing GSCs to TMZ treatment. Since 5-Aza successfully activated TUSC3 in MGMT-M GSCs but not in MGMT-UM GSCs, we were interested in exploring strategies to activate TUSC3 in MGMT-UM GSCs. Our data suggested that 5-Aza could not sensitize MGMT-UM GSCs to TMZ due to high expression of MGMT secondary to hypomethylation of MGMT promoter. We therefore sought to determine whether 5-Aza could activate TUSC3 and sensitize MGMT-UM GSCs to TMZ if MGMT expression was suppressed. We hypothesized that 5-Aza and suppression of MGMT could be pharmacologically leveraged to reactivate TUSC3 expression in MGMT-UM GSCs and further improve TMZ treatment efficacy (Figure 4A). We first treated two different MGMT-UM GSCs (UM-01 and UM-02) with Lomeguatrib, an irreversible MGMT inhibitor that has no impact on MGMT promoter methylation status [33]. Lomeguatrib treatment inhibition efficiency was evaluated by Western blot for protein expression and qRT-PCR for mRNA expression. As shown, the inhibitor Lomeguatrib significantly decreased both MGMT protein (Figure 4B) and mRNA expression (Figure S7) in MGMT-UM GSCs. Importantly, assessment of TUSC3 mRNA expression indicated that combination of 5-Aza and Lomeguatrib successfully activated TUSC3 expression in MGMT-UM GSCs, while 5-Aza or Lomeguatrib alone had no impact on TUSC3 expression (Figure 4C). We then assessed whether the combination of 5-Aza and Lomeguatrib could further sensitize MGMT-UM to TMZ. Interestingly, the combination of 5-Aza and Lomeguatrib significantly decreased TMZ IC-50 in MGMT-UM GSCs compared to the insignificant impact of either inhibitor alone (Figure 4D), suggesting potential therapeutic synergy between 5-Aza and MGMT inhibitions in reactivating TUSC3. Surprisingly, not only in GSCs, we also made similar observations in GBM cell lines. Specifically, 5-Aza activated TUSC3 expression in U251 GBM cells (MGMT-hypermethylated) and further decreased TMZ IC-50 (Figure S8A,B), while 5-Aza and Lomeguatrib combination successfully activated TUSC3 in T98G GBM cells (MGMT-hypomethylated) and sensitized cells to TMZ treatment (Figure S8A,B).

To further quantify drug interactions between 5-Aza, Lomeguatrib and TMZ in MGMT-UM GSCs, we calculated drug combination index utilizing the three-way (5-Aza vs. Lomeguatrib vs. TMZ) drug interactions model. We selected five doses for each drug. The drugs were combined in a 3D grid, where each drug was linearly increased in one axis. All drug responses were determined by viability test. As shown in Figure 5A, three-drug combinations, TMZ, 5-Aza and Lomeguatrib, demonstrated potent synergistic effects in MGMT-UM GSC (ZIP synergy score = 22.562 ± 2.722). Strong synergistic effects of drug combination on GSCs should facilitate effective dosing and lower toxicity on normal neural cells (NSCs). We then further tested those three drug combination responses on GSCs and NSC with a serial of 2-fold dose escalation and found out that the effective dose (reached 90% response, green line) for GSC had a minor impact on NSC, which indicated low toxicity (Figure S9). The above observations provided further rationale for combinatorial synergy between 5-Aza and MGMT inhibitor with TMZ in the highly resistant MGMT-UM GSCs. Our studies have thus far shown that optimal combinatorial therapy with 5-Aza and MGMT inhibition (Lomeguatrib) had a significant therapeutic impact on MGMT-UM GSCs under TMZ treatment. Next, we wanted to validate the above findings in vivo, utilizing orthotopic xenograft models. Xenografts bearing MGMT-UM GSCs were established by intracranial injection. All the mice received TMZ treatment plus one of the following: (i) vehicle, (ii) 5-Aza, (iii) Lomeguatrib and (iv) 5-Aza combining Lomeguatrib treatment. The Kaplan–Meier survival curve indicated that the combination of 5-Aza and Lomeguatrib together with TMZ significantly prolonged mice survival, while 5-Aza alone with TMZ did not have significant influences (Figure 5B). MGMT and DNMT1 silencing efficiencies in solid tumors by 5-Aza and Lomeguatrib, respectively, were confirmed by Western blot (Figure 5C) and qRT-PCR (Figure 5D) with protein and mRNA extracted from tumor tissues. mRNA expression assessment also demonstrated that TUSC3 was successfully activated by 5-Aza and Lomeguatrib combination in vivo (Figure 5E). This result provided in vivo support that combining 5-Aza together with MGMT inhibitor and TMZ treatment could lead to significantly improved outcomes in MGMT-UM GBM.

GBM-patient-derived samples were collected from patients at Moffitt Cancer Center (MCC) (Tampa, FL, USA) and approved by Moffitt Cancer Center Institutional Review Board, and all patients signed consent forms that were approved by the Institutional Review Board. Fresh resected human brain tissue samples were from the operating room and taken within 2 h to the laboratory to start tissue dissection. Tissue specimens were washed in 1× PBS and dissected under sterile conditions. It was further processed with tumor dissociation process using the Liberase (MilliporeSigma, St. Louis, MO, USA). Afterward, the homogenate was filtered through a 40 µm cell strainer and centrifuged at 800× g rpm for 5 min to pellet the cells. Cells were then used for culture and downstream gDNA extraction and assays.

The patient-derived GSCs were well characterized (Table 1) and cultured in NS-A medium (90% NeuroCult NS-A Basal Medium Human plus 10% Human NeuroCult NS-A proliferation Supplements, StemCell Technologies, Vancouver, British Columbia, Canada in 3D sphere as we described before [45]. Complete medium was supplied with recombinant human epidermal growth factor (R&D System, Minneapolis, MN, USA) and 100 units/mL penicillin-100 μg/mL streptomycin (Life Technologies, Carlsbad, CA, USA). Anti-TUSC3 and anti-MGMT antibodies were obtained from Abcam (Cambridge, United Kingdom); anti-β-actin IgG-HRP was obtained from Santa Cruz Biotech (Dallas, TX, USA) and anti-DNMT1, Goat anti-Rabbit IgG-HRP and Goat anti-mouse IgG-HRP were obtained from CellSignaling Technology (Danvers, MA, USA). Temozolomide (TMZ) was obtained from Sigma (St. Louis, MO, USA). 5-Azacitidine (5-Aza) and Lomeguatrib were obtained from Selleckchem (Houston, TX, USA).

Human TUSC3 expression and empty vector constructs (bacterial stock) were obtained from Genscript (Piscataway, NJ, USA). Plasmid was purified using QuickLyse Miniprep Kit (Qiagen, Germantown, MD, USA). GSCs (2 × 10⁵) were seeded in 6-well plate 24 h before transfection. Lipofectmine 3000 (ThermoFisher Scientific, Waltham, MA, USA) was used for transfection following manufacture’s protocol. Puromycin (2 μg/mL, InvivoGen, San Diego, CA, USA) was used for selection after transfection and the puromycin concentration was then reduced to 1 μg/mL 5 days later for maintenance.

CRISPR/Cas9 vectors lentiCRISPR-v2-puro was obtained from Addgene. sgRNA targeting human TUSC3 and sgRNA control were cloned into lentiCRISPR-v2-puro. TUSC3 targeting forward primer: CACCGCATTCGGAAGATTGAGCGTC, reverse primer: AAACGACGCTCAATCTTCCGAATGC; control sgRNA cloning forward primer: CACCGCACTCACATCGCTACATCA, reverse primer: AAACTGATGTAGCGATGTGAGTGC. Lentivirus was packed by 293T cells and cells were infected by lentivirus, followed by extensive selection of 2 μg/mL puromycin (InvivoGen, USA).

Heat-denatured proteins (50–100 µg) were loaded on 4–15% precast polyacrylamide gel (Bio-Rad, Hercules, CA, USA) along with molecular weight marker. The gel was run for 1–2 h at 120 V and then transferred to PVDF membranes (Bio-Rad) by semi-dry system. Membranes were blocked with 5% non-fat milk, 0.1% TBS-T (Tween) solution for 1 h at room temperature. The target proteins were then detected by the primary antibody at 4 °C overnight, washed with TBS-T (5 min each) and incubated with appropriate secondary antibody at room temperature for 1 h. The membrane was then washed three times with TBS-T, 5 min each. The target proteins were detected with luminol reagent and X-ray film (Santa Cruz Biotech, Dallas, TX, USA).

Total RNA was extracted for cells or tissues using RNeasy mini-prep kit (Qiagen) and quantified with Nanodrop 2000 (ThermoFisher Scientific, Waltham, MA, USA). Real-time PCR was performed by the Bio-Rad CFX96 Touch Real-Time PCR Detection system using cDNA synthesized with the iScript cDNA Synthesis kit (Bio-Rad). Human MGMT and GAPDH primers were used as we described before [46]. Human DNMT1 forward primer: AAAGTCAGCATTTGCGGGTT, reverse primer: TACTTTACGTGGCCAAAGCG; human TUSC3 forward primer: GAGAGCTGATACTTTTGACCTCC, reverse primer: CCCGAATATGAACATCCGTTCTG.

Cell viabilities were measured by an XTT Cell Viability Assay Kit (Cell Signaling Technology, Danvers, MA, USA) following treatment/transfection of cells. The absorbance was measured at 450 nm using a microplate reader (Molecular Device, San Jose, CA, USA). For proliferation assay, 2000 cells were seeded in 96-well plate. Parallel cultured cells were counted and recorded every 2 days. Proliferation curve was carried out using GraphPad Prism. When tested IC-50 in GSCs and drug response in GSCs and NSCs, 5000 cells were seeded in 96-well plates. Drugs were added the next day in triplicates and treated for 72 h (5-Aza: 10 µM; Lomeguatrib: 10 µM). Following drug treatment, cell viability was assessed using XTT Cell Viability Assay Kit (Cell Signaling). Data analysis and calculation of IC-50 were carried out using GraphPad Prism 9 (Dotmatics, Boston, MA, USA).

Nuclear extract was carried out according to the manufacturer’s instruction (Epigentek, New York, NY, USA). The activity of DNMT1 was detected by human DNMT1 ELISA kit according to the manufacturer’s instructions (Epigentek). The concentration of DNMT1 in samples was calculated by the working standard curve.

The trans-well migration and invasion assays were performed using 6-well Transwell^® Cell Migration and Invasion inserts according to the manufacturer’s instructions (Corning, New York, NY, USA). 1 × 10⁵ GSCs were seeded in trans-well and incubated 48 h (with DMSO or 5-Aza treatment) for migration and invasion. Migrating or invading cells were fixed with ethanol and stained with crystal violet. The numbers of migrated and invaded cells were quantified by counting under microscope.

To quantify drug interactions between 5-Aza and TMZ and also to classify the interactions, we calculated combination index and isobologram. A cross design was made to test the synergy and sensitivity of a drug pair in MGMT-M GSCs. We simulated isobologram for the pair of 5-Aza and TMZ with five effective doses cross combinations 1/10 titrated (5-Aza from 10 µM, TMZ from 1000 µM) for 72 h. Loewe additivity model was used for two drugs combination. To further quantify drug interactions among 5-Aza, Lomeguatrib and TMZ, we calculated drug combination index utilizing three-way (5-Aza vs. Lomeguatrib vs. TMZ in MGMT-M GSCs) drug interactions model. We chose five doses for each drug treatment with indicated titration for 72 h. For TMZ, the dose was titrated from 330 µM, while 5-Aza, Lomeguatrib were titrated from 33 µM and 20 µM. Drugs were combined in a 3D grid, where each drug was linearly increased in one axis. ZIP model was used for three drugs combination. All drug responses were determined by viability test. For synergy score measurement, data matrix was input and analyzed by SynergyFinder v3.0 (https://synergyfinder.fimm.fi↗).

Genomic DNA (gDNA) was extracted from GSCs using Quick-DNA Miniprep kit (Zymo, Irvine, CA, USA) and bisulfite converted using EpiTect Fast DNA Bisulfite Kit (Qiagen, Germantown, MD, USA) according to the manufacturer’s protocol. The bisulfite-converted DNA was subjected to PCR amplification using specific primers of MGMT promoter to amplify a 310 bps region of the promoter CpG island [46], forward: TTGAGTTAGGTTTTGGTAGTGTTTAG, reverse: CCTTTTCCTATCACAAAAATAATCC. PCR products were purified and cloned using pGEM-T Easy Vector System (Promega, Madison, WI, USA). DNA from positive clones were extracted by Qiaprep Spin Miniprep Kit (Qiagen) for sequencing.

Total RNA was extracted using RNeasy mini-prep kit (Qiagen) and followed by DNase digestion using RNase-Free DNase Set (Qiagen). Extracted RNA was labeled and hybridized onto the GeneChip PrimeView Human gene expression array cartriade (Affymetrix, Santa Clara, CA, USA).

GSCs were transfected with small interfering RNA control (MilliporeSigma) or indicated gene siRNA (MilliporeSigma) using Lipofectamine RNAiMAX Transfection Reagent (ThermoFisher Scientific). One day prior to transfection, the cells were seeded in 24-well plate (5 × 10⁴) and siRNAs (10 pmol per well) were prepared according to the manufacturer’s instructions and added to the cells. Cells were cultured 48 h and then for further assays.

Rescue experiments were performed in wild-type and TUSC3 knockout GSCs. In rescue groups, these GSCs were transiently transfected with either empty vector or full-length TUSC3 cDNA. Transfection was followed manufacturer’s instructions (Lipofectmine 3000, ThermoFisher Scientific).

Further, 6–8 weeks female NCRNU athymic mice were purchased from Taconic Biosciences. All animals were housed in the American Association for Laboratory Animal Care–accredited Animal Resource Center at Moffitt Cancer Center. All animal procedures and experiments were carried out under protocols approved by the Institutional Animal Care and Use Committee of the University of South Florida and Moffitt Cancer Center. All animal studies were performed in accordance with relevant guidelines and regulations of University of South Florida and Moffitt Cancer Center. For orthotopic model, xenograft tumors were established by intracranially injecting 1 × 105 indicated GSCs in a 2–3 μL volume of PBS in the right striatum of mice (n = 5/group) on a Stoelting Digital Stereotaxic Instrument (Stoelting, Wood Dale, IL, USA). Seven days after implantation, the mice were randomized into distinct groups for treatment (as indicated in figures). TMZ, 50 mg/kg/day; 5-Aza, 5 mg/kg/day; Lomeguatrib, 20 mg/kg/day. For survival studies, animals were followed every day until they lost 20% of body weight or had trouble ambulating, feeding or grooming.

Total genomic DNA was extracted using the Quick-DNA Miniprep kit (Zymo, Irvine, CA, USA) and followed by RNase A digestion. A DNA methylation assay using an Illumina DNA methylation MethylationEPIC array (Illumina, San Diego, USA) was performed on genomic DNA extracted from 16 GBM patient tissues. Minfi was used for normalization using Noob and FunNorm functions [47,48,49].

In-house gene expression data. The Affymetrix U133 plus 2.0 CEL files were normalized using IRON and log2-transformed before analysis [50]. The following criteria were used for selecting the most important genes: the fold change (FC) between the naïve (DMSO-treated) and the 5-Aza-treated cells was calculated. Probe sets with FC > 1 for both the MGMT-M and MGMT-UM, max (expression value naïve, 5-Aza treated) > 4 and matching signs were selected. Probe sets mapping to multiple genes or un-annotated were discarded. The final gene-level list was derived by selecting a single probe set for each gene.

TCGA GBM TUSC3 expression vs. Methylation. The GBM methylation data were downloaded as raw IDAT files and normalized via internal control probes followed by background subtraction. 4 TUSC3 CpG-probes located in the TSS200-5′UTR with r < −0.5 were averaged and correlated to TUSC3 expression level.

TCGA GBM survival analysis. Bioinformatics analysis and figures were completed using MATLAB 9.8 and Statistics and Machine Learning Toolbox 11.7, The MathWorks, Inc., Natick, Massachusetts, United States. Survival analysis was completed in MatSurv v1.1.03 [51].

Student’s t-test (for two-condition experiments) and ANOVA (for multiple-condition experiments) were employed. Survival was assessed using Kaplan–Meier analysis with statistical comparisons made by log-rank (Mantel–Cox) test. All statistical tests were considered significant at p < 0.05. * Means p < 0.05. For survival data, a sample size of 5 mice per group to obtain 80% power to detect a hazard ratio of 0.22 (control versus treatment group) at 5% type I error by a two-sided log-rank test.

All the datasets analyzed during the current study are provided by links in Results and Methods. The data that support the findings of this study are stored at Moffitt Cancer Center and Research Institute and are available from the corresponding authors upon reasonable request if required.