Integrin α6 signaling induces STAT3-TET3-mediated hydroxymethylation of genes critical for maintenance of glioma stem cells

Abstract

Both the extracellular matrix (ECM) and DNA epigenetic regulation are critical for maintaining stem cell phenotype and cancer progression. Whether and how ECM regulates epigenetic alterations to influence cancer stem cells (CSCs) remain to be explored. Here we report that ECM through laminin-integrin α6 upregulates ten-eleven translocation enzyme 3 (TET3) dioxygenase. TET3 in turn mediates DNA cytosine 5’-hydroxymethylation (5hmC) and upregulates genes critical for maintenance of glioma stem cells (GSCs). Activating integrin α6-FAK pathway increases STAT3 activity, TET3 expression and 5hmC levels in GSCs. Moreover, targeting STAT3 disrupts integrin α6-FAK signaling and inhibits TET3⁺ GSC maturation in vivo. STAT3 directly regulates TET3 expression and the two proteins are co-localized with 5hmC in GSC clusters. 5hmC is upregulated by STAT3 at the promoters of several tumorigenic genes, including c-Myc, known to be critical for GSCs. In vivo silencing of TET3 in GSC-enriched tumors reduces 5hmC accumulation and expression of the GSC critical genes, leading to tumor growth inhibition. TET3 expression and 5hmC accumulation also co-segregate with integrin α6 in patient malignant glioma. Thus, ECM- integrin α6-STAT3-TET3 axis regulates hydroxymethylation of genes important for GSCs, thereby increasing GSC tumorigenicity and resistance to therapies.

Introduction

Glioblastoma (GBM) is the most lethal brain tumor with few if any effective therapeutics. Although many lines of experimental evidence suggest that CSCs are, too a large degree, culprits for the failure of therapies for GBM patients, further identification of pathways/targets is critical for improvement of therapeutic outcomes.

The ECM consisting of a complex network of macromolecules is a major component of the niche critical for regulating stem cell behavior. The ECM is usually dysregulated in cancer and is critical for promoting cancer metastasis [1–6]. In addition to tumor metastasis, ECM also regulates stem cell differentiation and functions [7]. Integrins are among the best characterized cell receptors on stem cells that interact with ECM [1, 3], thereby connecting the intracellular cytoskeleton with the ECM. Insoluble adhesive cues, such as ECM protein laminin sensed by integrins, can transduce into signals that regulate stem cell differentiation and fates. One of the best examples illustrating the importance of integrin signaling in promoting CSCs is the finding that integrin α6 is critical for GSC proliferation, self-renewal and tumor formation capacity of GSCs [8]. Nevertheless, the detailed molecular mechanism underlying integrin α6-induced GSC tumorigenicity remains elusive.

A critical role of STAT3 in maintaining the cancer stem phenotype has been shown [9–11]. However, STAT3 is also upregulated in non-stem tumor cells. What might enable STAT3 to have unique role in CSCs needs to be explored. Relevant to the questions, GSCs have higher expression of c-Myc, which is required for GSC maintenance as well as survival, and also survivin and Bcl_XL, which endow GSC with increased survival potential, to maintain CSC phenotype and survival [12–14]. The question remains what might propel higher expression of these pro-CSC genes. Is ECM/integrin- α6 pathway critical for upregulating the pro-CSC genes? And if so, what is the pathway and mechanism by which integrin α6 drives GSC phenotype/survival.

Ten-eleven translocation enzymes, TET dioxygenases, are critical for gene promoter conversion from 5mC to 5hmC, thereby favoring gene demethylation [15–17]. Although TET proteins are shown to play a tumor suppressor functions, the findings remain highly contextual [18–24]. In hematopoietic malignancies, TET1 has been found to be a tumor suppressor as well as tumor-promoter [21, 22]. TET3 was recently shown to inhibit GSCs, mainly in the context of nuclear receptor TLX [20]. High levels of 5hmC have been associated with survival for glioma patients [23]. In stark contrast, 5hmC is critical for glioblastomagenesis in proneural glioblastoma [24]. Additionally, a critical role of TET dioxygenases in regulating embryonic stem cells has been described [25]. Nevertheless, whether TET dioxygenase activity may contribute to epigenetic regulation to control CSCs phenotype and/or increase their tumorigenicity requires in depth investigation. In the current study, using highly aggressive human GSC cells, both in vitro and in vivo, we have identified that laminin-integrin α6-STAT3 pathway drives GSCs through TET3-mediated hydroxymethylation of genes critical for GSC maintenance and survival. Moreover, we show that TET3 expression and 5hmC accumulation co-segregate with integrin α6 in GBM.

Results

TET3 promotes 5hmC accumulation and GSCs

Although an important role of TET dioxygenases in embryonic stem cells is known [25–27], whether they are critical for CSC maintenance in tumors remains to be further investigated. The human glioma cells isolated as stem-like cells (GSC008, GSC009, GSC030, and GSC106) are highly tumorigenic, display high STAT3 activity and ready to form spheres [28, 29]. We therefore compared the expression levels of TET enzymes in these human glioma stem-like cells cultured under conditions to enable either sphere formation (GSC) or differentiation (non-stem cells). We show that compared to non-stem human glioma cells, the patient derived primary GSCs expressed considerably higher levels of TET3 protein (Fig. 1A). We assessed expression of TET enzymes in the primary GSCs compared to differentiated non-stem counterparts, only TET3 is consistently upregulated in all 4 GSCs (Supplementary Fig. S1A), while TET1 and TET2 protein expression was barely detectable in GSCs (Supplementary Fig. S1B).

Fig. 1.

TET3 critical for characteristic tumor sphere formation is overexpressed in glioma stem cells favoring 5hmC accumulation. a TET3 protein expression was assessed by Western blotting compared in differentiated non-stem counterparts and primary human GSCs. Actin was re-probed and shown as a loading control. b Reduced tumor sphere formation by TET3 silencing was shown in primary human GSCs, GSC008, GSC009, GSC030 and GSC106. GSC008, GSC009, and GSC030 cells were treated with CpG-luciferase-siRNA or CpG-TET3-siRNAs every other day (n = 6) (upper panels). GSC030 and GSC106 were stably transduced with non-targeting shRNA or inducible TET3 shRNAs, and were treated with 5 μg/ml of doxycycline for shRNA induction (n = 6) (lower panels). SD shown, T-test: *) P < 0.05, **) P < 0.01, ***) P < 0.001. c Increased tumor sphere formation by TET3 overexpression was shown in primary human GSCs, GSC030 and GSC106. GSC030 and GSC106 were stably transduced with full length human TET3 cDNA (n = 6). SD shown, T-test: *) P < 0.05, **) P < 0.01. d Decreased tumorigenicity of primary human GSCs upon TET3 silencing was assessed by LDA. GSCs were stably transduced with non-targeting shRNA or inducible shTET3s, and treated with 5 μg/ml of doxycycline for shRNA induction. e 5hmC accumulation in primary human GSCs, GSC008, GSC030, and GSC106 were confirmed by flow cytometry upon digestion of RNA species. 5hmU accumulation as a TET3 independent DNA deamination alternative to TET3 dependent 5hmC-5fC conversion was included. f 5hmC in GSCs were compared to their non-stem counterparts by flow cytometry. g 5hmC (green) accumulation restricted to pSTAT3⁺MSI-1⁺ and pSTAT3⁺SOX2⁺ GBM cells but not to single positive pSTAT3⁺ GBM cells shown by confocal microscopy upon staining glioma patient tissue sections (left). Scale, 20 μm. Tissue profiles showing 5hmC accumulation confined to GSC-marker⁺ cells (right). a-g Representative data from 3 independent experiments are shown.

To assess whether TET3 contributes to GSC maintenance and sphere formation characteristic for GSCs, we silenced TET3 in the primary patient-derived GSC lines, which significantly reduced their ability of GSC sphere formation (Fig. 1B). By contrast, overexpressing TET3 in those GSC lines increased their ability of GSC sphere formation (Fig. 1C). Moreover, silencing TET3 expression diminished self-renewal capacity of patient-derived primary GSCs as shown by limited dilution assay (LDA), which assesses the frequency of repopulation (Fig. 1D) [30].

Because DNA cytosine 5’-hydroxymethylation (5hmC) is a product of TET3 enzyme activity, we hypothesize that 5hmC becomes a substrate for deamination in GSCs [31]. Direct comparison of DNA cytosine modifications using flow cytometric analysis showed increased levels of DNA hydroxymethylation (5hmC) and considerably elevated levels of its deamination product 5’-hydroxymethyluracil (5hmU) in GSCs (Fig. 1E). Moreover, levels of 5hmC in GSCs were significantly higher than those in their non-stem counterparts (Fig. 1F). Confocal imaging analysis following immunohistochemistry of human glioma tumor tissues indicated that 5hmC was restricted to cells positive for CSC markers, such as MSI-1, SOX2 and activated STAT3 (Fig. 1G).

Our results so far suggest that TET3 is upregulated in GSCs over their non-stem tumor cell counterparts (Fig. 1).

Integrin α6/FAK critical for 5hmC accumulation

The question remains what is unique about GSCs that allows elevated expression of TET3. Integrin α6, the cognate receptor for the integral ECM protein laminin, is upregulated selectively in GSCs, and critical for GSC maintenance [8]. However, the mechanisms by which ECM/integrin α6 drives GSCs remains to be further elucidated. To assess whether integrin α6 driven GSCs requires TET3/5hmC, we compared differentiated non-stem glioma cells with their GSC counterparts. Our results showed that integrin α6 protein expression was considerably elevated in patient-derived human GSCs (Fig. 2A and Supplementary Fig. S2), confirming findings from a previous report [8]. We next assessed induction of TET3 protein expression upon ectopic exposure of GSCs to components of the fibrous ECM, including laminin. Laminin, together with collagen and fibronectin interlaced with embedding non-fibrous proteoglycans, assembles the fibrillar networks of the ECM [2, 7]. We incubated single-cell GSC suspension in collagen, fibronectin, heparan sulfate proteoglycans (HSPG) or laminin. TET3 protein expression was robustly induced by laminin as assessed by flow cytometry and Western blotting (Fig. 2B, upper panels). Moreover, culturing a single-cell suspension of GSCs with laminin resulted in sphere formation characteristic of CSCs in 3D culture (Fig. 2B, lower panels). The GSC spheres were not detectable in collagen, fibronectin or HSPG treated cultures (Fig. 2B, lower panels). Furthermore, treating GSCs with laminin induced FAK and STAT3 activation as well as TET3 expression, which was accompanied by an increase in 5hmC modification (Fig. 2C). Conversely, inhibiting FAK activity by FAK inhibitor blocked STAT3 activation and TET3 expression, as well as 5hmC production (Fig. 2D), which was associated with the loss of GSC sphere integrity (Supplementary Fig. S3A). The cytohistochemistry and Western blotting results confirmed activation of FAK and STAT3 signaling was associated with increased protein expression of TET3 upon laminin treatment (Supplementary Fig. S3B). Furthermore, activation of FAK and STAT3 signaling was inhibited by silencing intergrin α6 or TET3 expression (Fig. 2E and 2F). Moreover, STAT3 knockdown resulted in considerably reduced TET3 expression in GSC-like cells in vivo (Fig. 2G).

Fig. 2.

Integrin α6/FAK signaling feeds STAT3 dependent TET3 expression and 5hmC accumulation. a Integrin α6 protein expressions in GSCs were compared to those of differentiated non-stem counterparts by Western blotting. Actin was re-probed and shown as a loading control. b Integrin α6 signaling mediated TET3 expression was assessed by flow cytometry and Western blotting in primary human stem-like GSC030 upon engagement with laminin compared to other components of the ECM (upper panels). Characteristic sphere formation upon exposure of a GSC single cell suspension to laminin and other ECM components is shown by bright field microscopy (lower panels). Scale, 100 μm. c Ectopic stimulation of stem-like GSC030 with laminin triggers activation of FAK/STAT3, expression of TET3 and 5hmC accumulation as shown by flow cytometry (top to bottom). d Blocking FAK activity with FAK inhibitor during stimulation of GSCs with laminin reduced activation of FAK/STAT3, expression of TET3 and 5hmC accumulation as shown by flow cytometry (top to bottom). e Silencing integrin α6 with inducible shRNA during stimulation of GSCs with laminin reduced activation of FAK/STAT3, expression of TET3 and 5hmC accumulation as shown by flow cytometry (top to bottom).f Silencing TET3 with inducible shRNA during stimulation of GSCs with laminin reduces activation of FAK/STAT3, expression of TET3 and 5hmC accumulation as shown by flow cytometry (top to bottom). g STAT3 silencing reducing TET3 expression was confirmed in vivo using human glioma U251 cells. GFP expression was controlled by the SOX2 promoter. GFP⁺ glioma cells considered stem-like phenocopies were analyzed for TET3 expression by flow cytometry once tumors were dissected and a single cell suspension was prepared. h STAT3 contributing to TET3 expression was validated by ChIP assay in two independent primary human GSCs, GSC009 and GSC030, and quantified. SD shown, T-test: *) P < 0.01, **) P < 0.01, **) P < 0.001. a-h Representative data from 3 independent experiments are shown.

We further assessed STAT3 transcriptional control of TET3 expression. Chromatin immunoprecipitation showed that STAT3 binds to the TET3 promoter in the human primary GSCs (Fig. 2H).

STAT3/TET3 forms complex with DNA 5mC

We next assessed how STAT3/TET3 might epigenetically regulate GSCs. We found pSTAT3/TET3 overexpression and co-localization in various human primary GSC lines derived from patient tumor samples (Fig. 3A). Immunoprecipitation pull-down experiments with either anti-STAT3 or anti-TET3 antibodies showed a robust complex formation between STAT3 and TET3 (Fig. 3B). The interaction between STAT3 and TET3 proteins was robust in laminin treated GSCs, while silencing integrin α6 inhibited STAT3-TET3 complex formation (Fig. 3C). Moreover, confocal STED microscopy revealed that pSTAT3 is co-localized in confined subnuclear domains with TET3 and its product 5hmC in patient glioma tumor tissues (Fig. 3D and Supplementary Fig. S4A–C). Furthermore, STAT3 exhibited a higher binding affinity to a methylated, compared to an unmethylated, STAT3-binding DNA oligo as analyzed by EMSA (Supplementary Fig. S4D). We were able to further demonstrate that TET3 was found in the same DNA-binding complex recognizing a methylated STAT3-consensus binding site, as assessed by an oligo-pull-down assay (Fig. 3E). These results suggest the formation of a complex of STAT3/TET3 with DNA 5mC, which could enable conversion of 5mC to 5hmC.

Fig. 3.

STAT3 and TET3 physically interact and cooperate in binding to methylated STAT3 target DNA sequences. a STAT3 and TET3 subcellular co-localization was assessed in four independent primary human GSC lines GSC008, GSC009, GSC030, and GSC106 by confocal microscopy. Scale, 50 μm. b Physical interaction of STAT3 and TET3 was shown by co-IP pulling down TET3 and detecting STAT3 and vice versa using whole cell lysates prepared from primary human GSCs, GSC030 and GSC106. c Physical interaction of STAT3 and TET3 was reduced by silencing integrin α6 with inducible shITGA6 as shown by co-IP pulling down TET3 and detecting STAT3 and vice versa using whole cell lysates prepared from primary human GSC106. d Nuclear co-localization of activated pY⁷⁰⁵STAT3 with TET3 and 5hmC was assessed by confocal STED microscopy in glioma patient tissue. Scale, 5 μm. e Enhanced cooperative binding of STAT3 and TET3 to methylated STAT3 target DNA sequence (SIE^5mC) was shown by oligo-pull-down assay using unmethylated or methylated SIE oligo. Protein complexes were detected in Western blot procedure after electrophoretic protein separation on SDS-PAGE. a-e Representative data from 3 independent experiments are shown.

Targeting STAT3 and TET3 decreases promoter hydroxymethylation of GSC associated genes and GSCs

We next tested how RNAi-based STAT3 silencing would affect DNA hydroxymethylation of promoters of genes critical for GSC maintenance and key characteristics such as survival and proliferation that enable them to resist therapies. 5-hydroxymethylation of c-Myc, Survivin, and Bcl_XL promoters was readily detectable in GSC-like cells. Upon STAT3 knockdown, 5-hydroxymethylation of c-Myc, Survivin, and Bcl_XL promoters was markedly decreased as demonstrated by PCR using primers flanking the STAT3 DNA-binding site (Fig. 4A).

Fig. 4.

STAT3 contributes to oncogene promoter hydroxymethylation in GSCs in vitro and STAT3 or TET3 knockdown prolongs mouse survival in vivo. a Hydroxymetylation of oncogene promoters upon STAT3 knockdown (shSTAT3) or TET3 knockdown (shTET3) was assessed by PCR using mRNA harvested from human primary GSCs, GSC030 and GSC106. Amplified cDNA was treated to sensitize non-hydroxymethylated cDNA for restriction enzyme digestion and products were separated on DNA gel (upper panels). Decreased oncogene promoter hydroxymetylation upon STAT3 knockdown was quantified by qPCR (lower panels). SD shown; T-test: *) P < 0.05, **) P < 0.01, ***) P < 0.001. Representative data from 3 independent experiments are shown. b Tumors grown from primary human GSCs stably expressing non-targeting shRNA or inducible shTET3 were treated with 2 mg/ml doxycycline water. Mouse survival was plotted out by Kaplan-Meier survival curves. c Tumors grown from primary human GSC, GSC030 were treated with CpG-siRNA conjugates as indicated including control CpG-luciferase-siRNA. Tumor growth kinetics were assessed. SD shown; T-test: ***) P < 0.001. d STAT3 and TET3 sensitive c-Myc oncogene promoter hydroxymetylation of GSC tumors treated as indicated was determined upon isolation of gDNA and treatment of gDNA to sensitize non-hydroxymethylated cDNA for restriction enzyme digestion. SD shown; T-test: *) P < 0.05, ***) P < 0.001.

This prompted us to silence TET3 in human primary GSCs in vivo. We implanted human primary GSCs, GSC030 and GSC106 stably expressing inducible TET3 shRNAs into NOD/scid/IL2R^−/− (NSG) mice, and TET3 knock-down significantly prolonged mouse survival (Fig. 4B). In addition to anti-GSC effects by CpG-STAT3siRNA in vivo as we have reported [9], treatment with CpG-TET3siRNAs as well as a combination of both resulted in significantly tumor growth delay of human primary GSC tumors when compared to a control CpG-siRNA conjugate (Fig. 4C). Furthermore, analyzing gDNA prepared from the tumor tissues indicated that the c-Myc promoter underwent significant 5-hydroxymethylation reduction in vivo upon silencing either TET3 or STAT3, or both via CpG-siRNA approach (Fig. 4D). Moreover, silencing TET3, STAT3, or both combined via CpG-siRNA significantly reduced global 5-hydroxymethylation in vivo (Fig. 5A, 5B and Supplementary Fig. S5). Moreover, silencing TET3, STAT3, or both combined, significantly downregulated integrin α6/FAK/STAT3/TET3, decreased 5hmC accumulation and significantly reduced the expression of c-Myc, Bcl_XL, and Survivin at protein level (Fig. 5C and 5D).

Fig. 5.

STAT3 critically contributes to oncogene promoter hydroxymethylation in GSCs in vivo. a STAT3- and TET3-sensitive protein expressions of integrin α6 as well as 5hmC accumulation in GSC tumors treated as indicated were assessed by confocal microscopy and b quantified. Scale, 50 μm. SD shown; T-test: *) P < 0.05, **) P < 0.01, ***) P < 0.001. c FAK protein expression in GSC tumors treated as indicated were assessed by confocal microscopy. Scale, 50 μm. d Reduced oncogene expression upon treatment of GSC tumors as indicated as determined by confocal microscopy and quantified. SD shown; T-test: **) P < 0.01, ***) P < 0.001.

TET3 expression and 5hmC accumulation co-segregate with integrin α6 in GBM

Several studies indicated a correlation between 5hmC overall levels and glioma disease stages and patient survival [23, 32], the status of 5mhC in GSCs within patient tumors has not been reported. To assess the relationship between TET3–5hmC and GSCs in patient glioma, 19 patient glioma tumor sections, spanning grades 2/3 and GBM, were analyzed for the association of integrin α6 expression with either TET3 protein expression or 5hmC accumulation (Fig. 6A). The majority of cases analyzed show that the expression level of integrin α6 was associated with 5hmC accumulation, although 5hmC was also detected in the absence of integrin α6 expression (Fig. 6A and 6B, upper panels). Correspondingly, integrin α6 expression was shown to be accompanied by TET3 protein expression in the majority of patient samples. Notably, TET3 expression in the absence of integrin α6 expression was not observed (Fig. 6A and 6B, lower panels). Significantly, the association of integrin α6 expression with TET3 and 5hmC was found to be confined to high grade GBM specimens (Fig. 6C).

Fig. 6.

Integrin α6 and TET3 expression resulting in 5hmC accumulation is confined to high grade glioma. a Integrin α6 and 5hmC (upper panels) or TET3 (lower panels) expression was assessed by immunohistochemistry followed by confocal microscopy of human brain tumor patient biopsies. Scale 20 μm (upper panels) and 10 μm (lower panels). b Quantitative case distribution of integrin α6/5hmC (upper panels) or integrin α6/TET3 (lower panels) expression. A total of 19 brain tumor cases were analyzed. c High frequency occurrence of integrin α6+/5hmC+ (left panel) or integrin α6+/TET3+ (right panel) expression is restricted to high grade glioblastoma as assessed from 19 brain tumor cases analyzed. a-c Total 19 patient cases were analyzed, and 5 to 10 images were acquired per slide.

Discussion

Previous studies have emphasized the importance of specialized niches, including perivascular and hypoxic regions, in anchoring GSCs in the tumor microenvironment, which is supported by many growth factors and cytokines, such as HIF1α and IL-6 [10, 33]. More recent studies have also suggested the critical role of ECM, via laminin-integrin α6 axis, in driving GSCs [8]. Nevertheless, the critical events downstream of integrin α6 in promoting GSC phenotype and ability to survive/proliferate thereby resisting both radiation and chemo-therapies remain to be explored. This study reveals that integrin α6, whose expression is characteristic of GSCs, signals through STAT3-TET3 to promote GSCs.

Although much is appreciated about the role of epigenetic regulation in guarding stem cell from differentiation [34–38], relatively little is known whether and how epigenetic reprogramming might promote CSC maintenance and phenotype. Nevertheless, a recent report suggested an important role of TET3 in inhibiting GSCs, mainly in the context of nuclear receptor TLX [20]. While knocking down TET3 was also shown to inhibit GSCs, the cell lines used in the two studies may differ. Our findings that TET3 and 5hmC co-segregate with integrin α6 are also confirmed in patient malignant gliomas. That different lineages of glioblastoma cells display opposing levels of 5hmC in terms of tumorigenicity has been reported [24, 32, 39, 40]. Granted biology is usually contextual, it is of high scientific interest to identify the causes of the discrepancies between our current study and the one by Cui et al [20] in terms of TET3’s role in GSCs.

Nevertheless, more recent studies have suggested oncogenic roles of TET enzymes and their potential to epigenetically reprogram pluripotent cell populations. Our findings are support the recent study [39] showing that TET enzymes play important roles in hypoxia mediated induction of pluripotency in gliomas. The study showed that expression of OCT4 and NANOG, which have known to be important for stem cell properties, increased after induction of TET1 and TET3 by hypoxia in glioma cell line. More recent studies have also suggested STAT/TET1 axis as a therapeutic target for acute myeloid leukemia (AML) [22, 40]. TET1 blockade using a chemical compound suppressed AML progression in vivo. Our data show that GSCs undergo epigenetic editing triggered by ECM/laminin, allowing elevated expression of genes important for maintenance of GSCs by hydroxymethylation.

The transcription factor STAT3 represents a central node in connecting ECM with hydroxymethylation of genes important for GSC maintenance and survival. We have delineated an ECM-intracellular signaling axis—laminin/integrin α6-FAK-STAT3-TET3–5hmC—in upregulating expression of genes critical for GSC maintenance, survival and proliferation, thus promoting GSC phenotype and resistance to therapies. STAT3 also facilitates 5hmC conversion by interaction with TET3. Our in vivo studies also suggest that TET3 can be a target for reducing GSCs thereby reversing drug resistance often associated with an increase in CSCs. The importance of TET3 and 5hmC in GSCs in GBM has also been validated in patient tumors. Taken together, our data indicate a novel and distinct ECM-regulated intracellular signaling pathway that promotes GSC phenotype and underlies GSC strong survival via DNA epigenetic regulation.

Materials and Methods

Cell lines and tissue

Previously characterized and published human primary, low-passage GSC cell lines, GSC008, GSC009, GSC030, and GSC106 [28, 29] (kindly provided by Dr. Christine Brown, City of Hope, in 2018) were cultured in GSC medium [DMEM/F12 (1:1) supplemented with HEPES (Gibco), B-27 (Invitrogen), Heparin-sodium (Sagent), antibiotics-antimycotics (Gibco), 20 ng/ml FGFb, 20 ng/ml EGF (Peprotech), 20 ng/ml leukemia inhibitory factor and GlutaMAX (Gibco)]. Cell culture was performed in ultra-low adhesion plates (Corning). Primary human GSC-like lines were subcultured using Accutase (Innovative Cell Technologies, Inc) to break up sphere clusters. Only the low passages cell lines were used in the study. For differentiation of GSC cell lines, single cell suspensions were washed with HBSS, than cultured in differentiation medium [DMEM/F12 (1:1) supplemented with 7% fetal bovine serum, HEPES (Gibco), GlutaMAX (Gibco), and antibiotics-antimycotics (Gibco)] for 7 – 14 days. For generation of lentivirus expressing shRNAs, non-targeting RNA (ntRNA) (GCGCGATAGCGCTAATAATTT), shSTAT3 (CCTGAGTTGAATTATCAGCTT), shTET3 #1 (GAACCTTCTCTTGCGCTATTT), shTET3 #2 (ACTCCTACCACTCCTACTATG), or shITGA6 (CGGATCGAGTTTGATAACGAT) were cloned into Tet-pLKO-mCherry vector. For TET3 overexpression, lentiviral vector expressing full length human TET3 cDNA and puromycin resistance gene was purchased from Applied Biological Materials Inc. Lentiviruses were produced in 293T cells using cloned or purchased vectors, and GSC030 or GSC106 cells were infected with shRNA or TET3 expressing lentiviruses. The mCherry/shRNA expressing cells were sorted by flow cytometry, and TET3 overexpressing cells were selected by puromycin treatment. TET3 or STAT3 shRNA knock-down was induced by adding 5 μg/ml of doxycycline to complete growth media. All cell lines were tested for Mycoplasma every 6 months using MycoAlert™ PLUS mycoplasma detection kit (Lonza, last tested in December 2018). Human brain tumor tissue sections embedded in paraffin were obtained from pathology core at City of Hope and Biomax (HuCAT018, HuCAT019).

Tumor growth kinetics in vivo and orthotopic brain tumor model

Mice were housed in a pathogen free animal facility at the Beckman Research Institute at the City of Hope National Medical Center, and animal study was carried out in accordance with the established institutional guidance and approved protocols from Institutional Animal Care and Use Committee at the Beckman Research Institute of City of Hope National Medical Center in compliance with NIH guidelines. Both male and female mice were used for the study, and the studies were not blinded. For tumor growth kinetics in vivo, athymic nude mice were purchased from the Jackson Laboratories, and 2.5 X 10⁶ GSC030 cells were engrafted subcutaneously to the mice supported by phenol-red free Matrigel (Corning) with phenol-red free RPMI1680 media at a 1:1 ratio. The mice were randomly assigned to be in the groups, and each group had four mice. Tumor bearing male mice were treated locally with 782.5 pmol/dose of CpG-luciferase-siRNA, CpG-STAT3siRNA [9], or CpG-TET3siRNA (equimolar mix of 3 CpG-siRNA conjugates comprising: sense-strand 1: 5’-GCAGACCAGUGUAACAACATT-3’, antisense-strand 1: 5’-GGTGCATCGATGCAGGGGGG-linker-UGUUGUUACACUGGUCUGCTT-3’; sense-strand 2: 5’-GAAGCAGAGUCUUUAAUGATT-3’, antisense-strand 2: 5’-GGTGCATCGATGCAGGGGGG-linker-UCAUUAAAGACUCUGCUUCTT-3’; sense-strand 3: 5’-GGAAAGCCAUUAGAUGAAATT-3’, antisense-strand 3: 5’-GGTGCATCGATGCAGGGGGG-linker-UUUCAUCUAAUGGCUUUCCTT-3’, TET3siRNA sequences were adapted from Santa Cruz, sc-94636) every other day, and volumes of tumors were measured every other day. For orthotopic mouse model, female NOD/scid/IL2R^−/− (NSG) mice were anesthetized with ketamine (100 mg/kg) and xylazine (20 mg/kg) mixture and 1 X 10⁵ GSC030 or GSC106 cells stably expressing non-targeting shRNA or TET3 shRNAs were engrafted into the striatum. One week after tumor cell injection, the mice were started to treat with doxycycline water (2 mg/ml) to induce shRNA expression. The mice were euthanized when they showed a neurological symptom of brain tumor and mouse survival was analyzed by plotting Kaplan-Meier survival curves using GraphPad Prism 7.0 software.

Immunoprecipitation, oligo-pull-down, and western blotting

Whole cell lysates were prepared using RIPA lysis buffer containing 50 mM Tris-HCl, pH7.4, 150 mM NaCl, 1 mM EDTA pH8.0, 0.5% NP-40, 1 mM NaF, 15% glycerol, and 20 mM β-glycerophosphate. Protease inhibitor cocktail was added freshly to the lysis buffer (miniComplete, Roche). For immunoprecipitation, antibodies were coupled to ProteinG Agarose beads (Invitrogen) prior to adding them to the cell lysates. Precipitation was performed overnight with shaking at 4°C. Then complexes were carefully washed and proteins were electrophoretically separated by SDS-PAGE followed by Western blot detection using antibodies raised against TET1 (ThermoFisher, MA5–16312), TET2 (ThermoFisher, PA5–35847), TET3 (abcam, ab135033 and ThermoFisher, PA5–31860), integrin α6 (Santa Cruz, sc-13542), STAT3 (Santa Cruz, sc-482 and Cell Signaling, 9139), pSTAT3 (Cell Signaling, 9145), pFAK (Cell Signaling, 8556) and β-actin (Sigma, A1978). STAT3-sensitive oligo-pull-down was performed using 2 μg biotinylated SIE ds-oligo (SIE-wt: 5’-AGCTTCATTTCCCGTAAATCCCTAAGCT-3’; SIE-5mC: 5’- AGCTTCATTTCCC^mGTAAATCCCTAAGCT-3’) which was incubated with 400 μg [protein] nuclear extract of cells in 500 μl binding buffer comprising 12 mM HEPES (pH7.9), 12% glycerol, 4 mM Tris/HCl (pH7.9), 150 mM KCl, 1 mM EDTA and freshly added 1 mM DTT, 0.1 μg/μl poly (dI:dC), 0.5 μg/μl BSA for up to 2 h at RT. Add sample to streptavidin magnetic beads (SMB) suspension of 50 μl SMB/sample prewashed thrice with binding buffer and cleared in a magnetic stand, and incubate for oligo-pull-down by SMB for 2 h at RT with rocking. Precipitate was resuspended in 4X Laemmli buffer subsequent to supernatant clearance in a magnetic stand and resolved on SDS-PAGE prior to membrane transfer and immunodetection.

Sphere formation of cell lines and treatments

For sphere formation, single cell suspensions were plated in ultra-low adhesion plates (Corning) with GSC medium. Spherical cell clusters were treated every day with 500 pmol/ml of CpG-siRNA conjugates or were cultured with 5 μg/ml of doxycycline for shRNA induction. Number of spheres was counted every other day.

Limiting-dilution assay (LDA)

Cells were cultured for at least 10 days in sphere forming medium. Prior to sphere assays, cells were placed into single cell suspension. Cultures free of clumping and being of ≥ 90% viability were used. Cells were serially diluted from 500 to 2 cells per well into 96-well and cultured in GSC medium, and fresh EGF and bFGF (PeproTech) were added every other days. The cells were cultured with 5 μg/ml of doxycycline for shRNA induction. Subsequently, cultures were analyzed for spheres in each well. Data were analyzed using the Extreme Limiting Dilution Algorithm (ELDA) [41].

Flow cytometry

Antibodies for flow cytometry were purchased from Cell Signaling Technologies (pY397FAK #3283, pPaxilin #2541), abcam (TET3 #ab135033, 5hmU #ab19735), Novus Biologicals (laminin #NB300–144), Active Motif (5mC #39649) or BD Biosciences (pSTAT3 #557815). Secondary antibodies used were coupled to either Alexa Fluor 647 or Alexa Fluor 488 (Invitrogen) diluted up to 1:4,000 in PBS/1% BSA staining buffer. For blocking FAK activity, GSC cell lines were treated with 10 μM FAK inhibitor (Santa Cruz #sc-203950). Intracellular staining was performed after fixation of single cell suspensions with 2% paraformaldehyde and permeabilization with ice-cold 100% methanol, blocked with PBS/1% BSA for 1h at 4°C, and stained with an antibody diluted at 1:50 in PBS/1% BSA for 30 – 45 min at room temperature. Detection of DNA modification was performed upon RNA digestion using RNase A at 40 μg/ml for 10 min at 37°C prior to blocking procedure. Stained single cell suspensions were analyzed using an AccuriC6 cytometer (BD Biosciences), followed by FlowJo 7.6.1 analysis.

Imaging

Indirect immunofluorescence on frozen tissue sections or paraffin embedded tissue specimen was carried out as previously described [42]. Briefly, after deparaffinization and antigen retrieval at low pH citrate buffer (Vector), tissue microsections were fixed with 2% paraformaldehyde, followed by permeabilization with ice-cold 100% methanol, treated with signal enhancer (Image-iT, Invitrogen) for 30 min, blocked with PBS containing 10% goat serum and 2.5% mouse serum (Sigma) for 1h at room temperature, and stained with a 1:50 dilution of the primary antibody (TET3 #ab135033, MSI-1 #ab21628, SOX2 #ab59776, antibodies were obtained from abcam; integrin α6 #sc-13542 and pSTAT3 #sc-7993 antibodies were purchased from Santa Cruz; pY397FAK #3283 was obtained from Cell Signaling Technologies; anti-5mC #39649, 5hmC #39769, 5caC #61225, 5fc #61223 were purchased from Active Motif and 5hmU #ab19735 was from abcam) in PBS/sera overnight at 4°C. After washing with PBS, secondary antibodies were diluted 1:100 in PBS/sera containing 100 ng/ml Hoechst33342 (Invitrogen). After final washing with PBS, specimens were mounted with Mowiol (Calbiochem).

Indirect immunofluorescence on cultured cells was carried out as described previously [42]. Confocal analysis was performed using LSM510 Meta or LSM700 (Zeiss). Ultra-resolution imaging for detection of distinct distribution of nuclear pSTAT3 was performed using confocal STED microscopy (kindly supported by Laurent Bentolila, at the Advanced Light Microscopy/Spectroscopy Laboratory Light Microscopy Core facility of the UCLA, USA, using a Leica SP5 MP-STED microscope).

Intravital multiphoton imaging was performed as previously described [42]. Blood vasculature of the brain was stained by systemic injection of 100 μg dextran-rhodamine (Invitrogen) diluted in sterile HBSS (Gibco). Fibrous ECM emission is given by second harmonic generation autofluorescence. We used an Ultima 2photon microscope for IVMPM analysis (Prairie Technologies). Acquired Z-stacks were reconstructed to full 3D view using Amira software (Visualization Sciences Group).

Hydroxymethylation (5hmC) detection

Hydroxymethylation of oncogene promoters was assessed using the EpiMark® 5hmC and 5mC Analysis Kit aobtained from New England Biolabs according to the manufacturer’s instructions. Briefly, gDNA isolated by phenol/chloroform extraction was modified using T4 b-glucosyltransferase and UDP-glucose resulting in glucosylation of 5hmC (5ghmC) where 5hmC glucosylation is thought to protect cleavage of hydroxymethylated gDNA by distinct restriction enzyme MspI. Once enzymatic digestion is completed, levels of hydroxymethylation of oncogene promoters was assessed by PCR based amplification of digestion products focusing on oncogene promoter regions bearing STAT3 binding sites. STAT3 binding site flanking PCR primer pairs were designed for c-Myc, Bcl_XL and survivin promoter: Birc5-1fwd: 5’-TCAAATCTGGCGGTTAATGG-3’, Birc5-1rev: 5’-TCAAGTGATGCTCCTGCCTA-3’; Birc5-2fwd: 5’-CTCGATGGGGACAAAGCA-3’, Birc5-2rev: 5’-AGAACTGGCCCTTCTTGGAG; CMYC-1fwd: 5’-AGAAGCCCTGCCCTTCTC-3’, CMYC-1rev: 5’-TGAGTCTCCTCCCCACCTT-3’; CMYC-2fwd: 5’-GCTGCTATGGGCAAAGTTTC-3’, CMYC-2rev: 5’-TCGGGGCTTTATCTAACTCG-3’; BCL2L1-fwd: 5’-CCCTCCTCTCAGGAAGGTCT-3’, BCL2L1-rev: 5’-TTTGCTCTGAATTCCCCAAA-3’; BCL2L2-fwd: 5’-TGCAAGTTCCCCTGTCTCTT-3’, BCL2L2-rev: 5’-CCCCCTAGACCTTCCTGAGA-3’; BCL2L3-fwd: 5’-GGACGGATGAAATAGGCTGA-3’, BCL2L3-rev: 5’-GGTAGCTTCCAGACGCAAGA-3’; BCL2L4-fwd: 5’-AGAGTACTCCTGGCTCCCAGT-3’, BCL2L4-rev: 5’- GTATCACAGGTCGGGAGAGG-3’. Amplified promoter sequences were analyzed by either analytical DNA-agarose gel electrophoresis or q-PCR.

Chromatin immunoprecipitation assay

Chromatin immunoprecipitation was performed using a ChIP assay kit (Upstate Biotechnology) according to the manufacturer’s instructions. We performed chromatin immunoprecipitation in cells based on the protocol provided by Upstate Biotechnology. The chromatin was prepared from 5 × 10⁶ cells and subjected to ChIP assay as previously described [42]. 4 μg of anti-STAT3 rabbit polyclonal antibody or control rabbit IgG were used for immunoprecipitation. The following primers, 5’- ATATTATCTTATCCACATCC −3’, and 5’- AACATCAGACAAACCCAAAT −3’, were used to amplify the human TET3 promoter region (−1440~−1230 bp from the transcription start site), spanning the putative STATx-binding sites that contained a putative STATx-binding sites as suggested by TRANSFAC software.

Statistical analysis

At least three replicates were compiled for all experiments, and n-values listed in the figures and/or the legends. Statistical analysis was performed using GraphPad Prism 7.0 or Microsoft Excel. Data were represented as the mean ± the standard deviation. All data are normally distributed, and the variances are similar. The Student’s t test was used to analyze whether two experimental groups have significant difference. A two-tailed value of P < 0.05 was considered statistically significant, while n.s. indicates not significantly different.