INTRODUCTION
With a median survival of about 15 months (Schwartzbaum et al,
2006; Stupp et al,
2005), glioblastoma is the most frequent and aggressive of all gliomas, with a propensity to invade the surrounding parenchyma (Wen & Kesari,
2008). Individual tumour cells can be found far from the primary tumour site, often crossing great distances into the contralateral hemisphere (Wilson,
1992). These cells cannot be isolated for surgical resection, or easily targeted by irradiation, and thus represent sources for tumour recurrences (Glas et al,
2010). Adjuvant chemotherapy (
e.g. Temozolomide, TMZ) is therefore included as a critical component of the current standard of care, in attempt to address these residual invasive cells. Given the exceedingly poor prognosis, it is critical to understand the biology of treatment‐resistant glioblastoma cells.
The cancer stem cell hypothesis (Reya et al,
2001) postulates an intra‐tumoural hierarchy, where a small population of tumour cells has greater abilities to initiate and propagate tumours (Ignatova et al,
2002), rendering cancer stem cells an important therapeutic target (Vescovi et al,
2006). Cancer stem cells have been shown to be more invasive and therapy resistant than other cells of the same tumours (Bao et al,
2006; Cheng et al,
2011; Lathia et al,
2011). Tumour heterogeneity is a direct implication of the cancer stem cell hypothesis, and indicates that cell populations with different properties (such as drug resistance or higher capacity for tumour/recurrence formation) exist within the same tumour (Siebzehnrubl et al,
2011). The triad of tumourigenesis (cancer stemness), invasion and therapy resistance is a fatal combination if merged in a single cell population, and renders such a population an important contributor to poor outcome.
This triad is induced by Epithelial–Mesenchymal Transition (EMT) in cancers outside the CNS, where EMT is the major cause of invasion and metastasis (Chaffer & Weinberg,
2011), and cancer cells undergoing EMT have been shown to acquire stem cell traits and are frequently more therapy resistant (Mani et al,
2008; Polyak & Weinberg,
2009; Singh & Settleman,
2010). Therefore, EMT can generate cell populations that combine these three above‐mentioned hallmarks. However, the role of EMT and related processes in brain cancer has received little attention thus far (Kahlert et al,
2012; Lu et al,
2012; Mikheeva et al,
2010), likely because the brain is lacking critical tissue components (
i.e. epithelium and mesenchyme). Yet, it is conceivable that key invasion pathways overlap between CNS and other cancers, and that factors inducing EMT outside the brain also activate the triad of invasion, stemness and chemoresistance in malignant gliomas.
Many factors, including reduced cell adhesion (Asano et al,
2004), reduced matrix adhesion (Nakada et al,
2007), matrix protease secretion (Rao,
2003) and cytoskeletal remodeling (Giese et al,
1996) have been advanced as determinants of glioma invasion; several of these pathways are induced by EMT outside the CNS (Chaffer & Weinberg,
2011). Therapy resistance in glioma is mediated by expression of DNA repair enzymes (Bao et al,
2006), and/or expression of drug efflux transporters (Bleau et al,
2009). Of particular note,
O‐6‐Methylguanine DNA Methyltransferase (MGMT) confers resistance to the standard of care drug TMZ (Bocangel et al,
2002). While MGMT is also expressed in several non‐CNS cancers (Gerson,
2004), it is currently unknown whether EMT can induce MGMT expression. Glioma stemness has been linked to a number of transcription factors, such as SOX2 (Gangemi et al,
2009), OLIG2 (Ligon et al,
2007), and BMI1 (Facchino et al,
2010). SOX2 and BMI1 are targets of EMT activators, in particular of ZEB1 (Wellner et al,
2009).
ZEB1 is an inducer of EMT, transcriptional repressor of cell‐adhesion molecules, miRNAs—particularly the miR‐200 family—and cell polarity‐associated genes (Brabletz & Brabletz,
2010; Wellner et al,
2009). It has emerged as one of the master regulators for metastasis (Brabletz & Brabletz,
2010) and plays a critical role in tumour initiation at distant sites (Wellner et al,
2009). Edwards et al (
2011) showed induction of ZEB1 through the tumour microenvironment in glioma, and related ZEB1 expression to repression of E‐cadherin and thus invasion. Given the low prevalence of E‐cadherin in glioma (Utsuki et al,
2002), we asked whether ZEB1 could affect glioma invasion through other mechanisms. We further aimed to elucidate if ZEB1 regulates other, typically EMT‐related processes in brain cancer, such as potential for recurrence and therapy resistance. Expression of ZEB1 has been observed in chemoresistant cells in cancers outside the CNS (Li et al,
2012; Wang et al,
2009), but whether this relation is correlative or causative is thus far unknown.
Our study provides a systematic analysis of the functions of ZEB1 in glioblastoma pathobiology. Specifically, we address how ZEB1 exerts its effects on key malignant processes in glioma, i.e. invasion, tumourigenesis and therapy resistance.
DISCUSSION
The findings presented here support that inducers of EMT in non‐CNS cancers can promote single‐cell invasion, chemoresistance and tumourigenesis in glioblastoma. While we identify a number of EMT‐related factors in primary cell lines and tumour specimens, ZEB1 seems to dominate these processes, in up to 50% of patients with glioblastoma. However, the apparent heterogeneity of EMT‐associated proteins in our samples indicates that these molecules may have redundant functions, fine‐tune downstream pathways and/or serve specific purposes in a temporal and spatially regulated fashion.
Invasion is a common feature of brain malignancies from WHO grade II onwards, and we find ZEB1 expressed in invasive tumours (grade II–IV), but not in non‐invasive neoplasms (grade I, data not shown). Thus, expression of ZEB1 seems to be associated with infiltrating tumour cells across a spectrum of glioma. A number of invasion pathways have been identified (Dirks,
2001; Hoelzinger et al,
2007; Lefranc et al,
2005), and it is likely that several contribute to glioma invasion. Invasive ZEB1‐positive cells possess a greater degree of freedom for cellular locomotion due to the actions of ROBO1, uncoupling N‐cadherin from the cytoskeleton (Rhee et al,
2002). This uncoupling allows N‐cadherin to diffuse freely across the cell membrane, which contributes to the ability of invasive cells to break free from neighbouring tumour cells and infiltrate into the surrounding tissue. Tumour infiltration may further be facilitated by a variety of cellular migratory mechanisms (
e.g. chemotaxis, autocrine/paracrine signalling, etc.).
In solid‐tissue cancers outside the CNS, tumour invasion and EMT are driven by an E‐cadherin to N‐cadherin switch (Thiery et al,
2009), characterised by the loss of E‐cadherin and a concomitant increase in N‐cadherin. Here, we observe no significant levels of E‐cadherin in our sample cohort, and also find no change in N‐cadherin expression. In contrast, our data suggests that increased tumour invasion is mediated by a change in N‐cadherin dynamics, mediated through ROBO1 by ZEB1. Of note, there are some cases of glioma where E‐cadherin is present (Lewis‐Tuffin et al,
2010). It is tempting to speculate that the subset of glioma in which E‐cadherin is abundant has a more pronounced epithelial character, and might present as gliosarcoma. In these tumours, it is conceivable that the ‘classic’ E‐to‐N switch is the driving force of tissue infiltration.
Our data strongly support that invasive cells are also more resistant to the current standard of care drug TMZ, rendering these cells prime candidates for tumour recurrence. An intricate regulatory pathway, including miR‐200c, c‐MYB and MGMT, maintains this resistance. To date,
MGMT promoter methylation is the most reliable prognostic marker for therapy resistance (Hegi et al,
2005). As ZEB1 does not influence methylation in our primary cell lines, it is possible that ZEB1 protein analysis may yield prognostic information that complement
MGMT methylation data. ZEB1, miR‐200c and c‐MYB may constitute a novel pathway for the chemoresistance enzyme MGMT. We observed differences in the subcellular localisation of MGMT between xenografts (cytoplasmic) and clinical specimens (nuclear). Of note, Ishibashi et al. observed that MGMT is present in both fractions, and postulated that cytoplasmic MGMT is translocated to the nucleus after nuclear MGMT is depleted (Ishibashi et al,
1994). It is therefore possible that the nuclear staining pattern in patient specimens is due to previous treatment with alkylating agents, resulting in nuclear accumulation of MGMT. ZEB1 was a better predictor of outcome and therapy response at the protein level than MGMT. This may indicate that additional pathways regulate MGMT, as we found more specimens MGMT positive than ZEB1 positive. However, the strong correlation between ZEB1 and MGMT shows that ZEB1‐positive cases are highly likely to express MGMT, explaining their poor response to TMZ.
In accordance with the cancer stem cell hypothesis, ZEB1 is linked to the expression of stemness‐associated factors and tumourigenesis. Our data indicate that critical stem cell regulators, such as SOX2 and OLIG2, are induced by the ZEB1‐miR‐200 feedback loop in glioblastoma. The presence of ZEB1‐positive cells at the tumour invasion front strongly supports an invasive niche that contains cancer stem cells (Cheng et al,
2011; Lathia et al,
2011). As the ZEB1‐positive population is characterised by high motility and increased chemoresistance, these cancer stem cells are a candidate population for tumour recurrence. Of note, we have previously identified a more quiescent population of glioblastoma stem cells (Deleyrolle et al,
2011; Piccirillo et al,
2006), and we observed here that increased levels of ZEB1 are associated with slower proliferation. Whether the ZEB1 pathway directly reduces proliferation rates of glioblastoma stem cells remains to be tested.
Protein‐level analysis shows that about 45% of human glioblastomas express ZEB1, and patients with ZEB1‐negative glioblastomas have survival benefits that are likely related to an improved response to TMZ therapy. We cannot exclude that ZEB1 protein analysis is affected by sample collection site (i.e. core vs. edge), but fluorescence immunostaining detected ZEB1 only in samples that were found ZEB1 positive in immunoblots. Sample bias may also affect collections in databases (e.g. TCGA), which might explain the comparatively low prevalence of ZEB1 (and other EMT‐related factors) in these datasets. Another possibility is post‐transcriptional regulation of ZEB1 expression, e.g. through microRNAs, which may result in divergent levels of mRNA and protein. Further studies are required to address these issues.
ZEB1 expression in patient tumour samples is significantly enriched in the proliferative subclass (Brennan et al,
2009; Phillips et al,
2006), which appears in conflict with the lower proliferation of ZEB1‐expressing cells in patient samples. A potential explanation is that ZEB1‐positive cancer stem cells generate large numbers of rapidly dividing progenies, which in turn drive classification. Indeed, it has recently been observed that CD133‐positive glioblastoma stem cells generate rapidly proliferating, but less invasive and less tumourigenic progenies (Chen et al,
2010). This study also found significantly lower percentages of CD133‐positive cancer cells in mesenchymal
versus proliferative glioblastoma.
An enrichment of EGFR amplification has been described in the proliferative subclass (Huse et al,
2011), which we confirmed in our cohort. Given the strong correlation with EGFR expression, it is conceivable that EGF signalling induces ZEB1 in these tumours. Since others have found induction of ZEB1 through beta‐catenin or NF‐KB signalling (Edwards et al,
2011; Kahlert et al,
2012), it is possible that different pathways activate ZEB1 in different subclasses. The enrichment of EGFR expression in ZEB1 positive tumours suggests that tumour treatment may be more efficacious via EGFR inhibition in combination with TMZ treatment. However, clinical trials employing this strategy have shown only limited effects to date (Wick et al,
2011). Further research is needed to determine whether better patient selection or new pharmacological approaches may be more successful and whether ZEB1 status may be of prognostic value in these cases.
Our findings establish ZEB1 as a regulator of invasion and chemoresistance in glioblastoma, and a candidate agent for tumour recurrence. The multiple ZEB1‐associated regulators of brain tumour growth and invasion outlined in this study provide potential targets for future therapeutic approaches intervening at the level of invasion and/or chemoresistance.
MATERIALS AND METHODS
Cell culture
Tumour cell lines were generated (Piccirillo et al,
2006) and maintained (Siebzehnrubl et al,
2009) as described. Briefly, 50,000 cells were seeded per ml of culture medium (N2, Invitrogen, Carlsbad, CA) in the presence of mitogens (20 ng/ml each of EGF and FGF2, Sigma, St. Louis, MO). Cells were propagated as spheres and passaged using Accutase (PAA, Cölbe, Germany) every 7 days. For experiments with adherent cells, spheres were dissociated and plated in N2 medium supplemented with 1% foetal bovine serum (FBS). For scratch assays, 2 × 10
6 cells were plated per well of a six‐well culture plate coated with poly‐
l‐ornithine and laminin (15 µg/ml) in N2 containing 1% FBS, and grown to confluence overnight. Confluent monolayers were scratched with a pipette tip, and imaged at the time of the lesion and 24 h later. For sphere formation assays, 1,000 cells were plated per well into 96‐well culture plates in N2 containing EGF and bFGF. Spheres larger than 50 µm were counted 7 days after plating.
Flow cytometry
Immunostaining of cancer cells and flow cytometry was performed as described (Deleyrolle et al,
2011; Piccirillo et al,
2006). Briefly, live cells were dissociated using PBS and 0.5 mM EDTA and subsequently incubated with primary antibody or IgG controls in PBS containing 0.1% BSA for 1 h on ice. Following 2 washes in PBS, cells were analysed on a BD LSR II (BD Biosciences, San Jose, CA). Data was analysed and dot plots were generated using FlowJo Ver. 8.8.7 (Tree Star, Ashland, OR).
Cell viability assay
The Methyltetrazolium bromide (MTT) assay was used as indicator of cell viability and performed as described (Holsken et al,
2006). Briefly, 10,000 cells were plated per well into 96‐well cell culture plates and treated 1 h after plating with varying concentrations of TMZ (ranging 5 µM–5 mM, Tocris, Ellisville, MO). Concentration‐effect curves for TMZ treatment were generated by nonlinear regression analysis as described (Holsken et al,
2006). Bar graphs are derived from individual concentration measurements, compared to the appropriate controls.
Knock‐down experiments
Plasmids for knockdown of ZEB1 and expression of hsa‐miR‐200c, as well as antago‐miR‐200c and control sequences are as described previously (Wellner et al,
2009). Plasmids for knockdown of c‐MYB, MGMT and ROBO1 were obtained from OpenBiosystems (Lafayette, CO). The expression plasmid for c‐MYB (Clarke et al,
1988) was a kind gift of Dr. J.S. Lipsick (Stanford University). Expression plasmids for MGMT and ZEB1 were obtained from Origene (Rockville, MD). Cancer cells were transfected using Lipofectamine LTX (Invitrogen) according to the manufacturer's instructions. Transfected cells were selected using puromycin or geneticin (Sigma) before being used for subsequent experiments.
Animal experiments
Adult female Fox‐Chase SCID mice (Charles River, Wilmington, MA) were used for in vivo tumour transplants. All procedures were performed according to NIH and institutional guidelines for animal care and handling. After animals were deeply anaesthetised using USP grade Isoflurane (Halocarbon, North Augusta, SC), an incision was made in the scalp, the skull demonstrated and a hole drilled at the coordinates Bregma −0.5 mm anterior and −1.5 mm lateral. A Hamilton syringe was lowered 2.5 mm into the burr hole, and 1 µl of a cell suspension was injected over 5 min before the needle was retracted. After the incision was closed with surgical staples the animal was allowed to recover before being returned to the cage. Animals were transplanted with doses ranging from 1,000 to 100,000 cells, and tumour‐bearing animals were scored regularly for tumour‐related symptoms. Moribund animals were anaesthetised and transcardially perfused with 4% paraformaldehyde in saline, the brains removed, postfixed and prepared for histology. For in vivo TMZ treatment, animals received orthotopic grafts of 150,000 cells. One week after transplantation, tumour‐bearing animals were intraperitoneally injected with 20 mg/kg TMZ in saline (final DMSO concentration 25%). Animals received five injections over 5 days, corresponding to one cycle of TMZ treatment. Animals at endpoint (defined as bodyweight loss ≥20% or observation of severe neurological symptoms) were perfused and tumour presence was confirmed histologically. For in vivo analysis of inducible ROBO1 knockdown, animals received orthotopic grafts of 150,000 cells. One group received doxycycline injections i.p. (10 mg/kg in saline) at day 4 post surgery and then every other day.
Patient sample collection
Tissue specimens from glioblastoma patients were obtained from the Florida Center for Brain Tumour Research (FCBTR) and the UF and Shands Department of Pathology with approval from the UF Institutional Review Board (IRB). Informed consent was obtained from all subjects and all experiments conformed to the principles set out in the WMA Declaration of Helsinki and the NIH Belmont Report.
Immunohistochemistry and Immunocytochemistry
Immunostainings were performed as described (Siebzehnrubl et al,
2009; Zheng et al,
2006). Paraffin‐embedded patient material was deparaffinised, followed by heat‐mediated antigen retrieval in a 10 mM citric acid buffer (pH 6.0) or Trilogy (Cell Marque, Rocklin, CA), and tissue was subsequently immunostained with 3,3′‐diaminobenzidine (Vector Elite Kit, Vector labs, Burlingame, CA) or fluorescence‐conjugated antibodies using standard protocols. A table of employed antibodies, suppliers and dilutions can be found in the Supporting Information.
Image acquisition and data analysis
Low‐power fluorescent images were taken on a Leica DMLB epifluorescence microscope (Bannockburn, IL) equipped with a CCD camera (Spot Imaging Solutions, Sterling Heights, MI). To obtain full images of brain sections, multiple grey‐scale images were acquired per section using Spot Advanced software (Spot Imaging Solutions) and merged into a full image and inverted into black‐on‐white images using Photoshop CS4 (Adobe Systems, San Jose, CA). Photomerged images were imported into ImageJ and threshold levels were adjusted to distinguish tumour from background. Using the wand tool, all outlines of positively stained (black) tumour areas were selected in each section and the perimeter (line surrounding the tumour) and area of the tumour were measured. The wand tool allows an exact distinction between black (tumour) and white (parenchyma) regions; hence, the measurement of tumour outline and area is unbiased. The ratio of the squared perimeter distance over the area (P2/A) was calculated and used to compare invasive properties of different tumours. Since P2/A is a dimensionless number, the resulting figure is termed ‘invasion index’. A higher invasion index is indicative of a more dissociated tumour, whereas a lower invasion index represents a more spherical tumour. High‐power images were taken on an Olympus BX‐81 DSU spinning disc confocal microscope (Olympus, Center Valley, PA) and projection images of z‐stacks were generated using Slidebook (Olympus) software. For mean fluorescence intensity analysis, two visual fields within the tumour core or edge were selected at random per animal, and a confocal z‐stack through the entire section was obtained. For each stack, one plane of section was selected and mean gray values for each channel obtained using ImageJ, and the ratio of average channel intensities for ZEB1 and Hoechst was calculated.
RNA isolation and quantitative real‐time PCR
Total RNA was isolated from tumour sphere or adherent cultures using the RNeasy Mini Kit (Qiagen, Valencia, CA) according to the manufacturer's instructions. RNA was quantified on a Nanodrop Spectrophotometer (Thermo, Wilmington, DE), and 1 µg of total RNA was used for cDNA synthesis as described (Siebzehnrubl et al,
2009). Twenty‐five nanograms of cDNA were used for quantitative PCR using the SYBR green PCR master mix (Applied Biosystems, Carlsbad, CA) on an ABI 7900HT (Applied Biosystems) as previously described (Siebzehnrubl et al,
2009). Expression levels of ZEB1 were quantified in triplicate relative to beta‐actin using the ΔΔ
Ct method. Primer sequences and amplification times are described elsewhere (Wellner et al,
2009).
Protein isolation and Western blotting
Proteins were isolated from cancer cell cultures and primary tumour specimens as described (Siebzehnrubl et al,
2009). For Western blotting, 5–40 µg of denatured protein were loaded on 4–12% Bis–Tris reducing gel (Invitrogen), separated and blotted onto a PVDF membrane (iBlot, Invitrogen). Blots were blocked and probed with respective primary and secondary antibodies (see Supporting Information) as described (Siebzehnrubl et al,
2009), and developed using the ECL Plus kit (Amersham, Piscataway, NJ) on a FluorChemQ Multi Image III (Cell Biosciences, Santa Clara, CA) and AlphaInnotech software version 1.0.1.1. Band densitometry was performed using ImageJ.
Chromatin immunoprecipitation
Chromatin was cross‐linked and isolated from 2.0 × 107 cells in sphere culture using the Simple ChIP kit (Cell Signaling, Danvers, MA) according to the manufacturer's instructions. Following immunoprecipitation with anti‐MYB and IgG control antibodies, the MGMT promoter region was amplified with specific primers (primer sequences and PCR conditions available upon request).
Bisulfite genomic sequencing
DNA was prepared and bisulfite converted as previously described (Pardo et al,
2011). Bisulfite converted DNA was amplified using
MGMT‐specific primers (sequences available upon request). PCR products were gel‐extracted and cloned into pGEM T‐easy vector (Promega, Madison, WI) followed by transformation of TOP10 cells (Invitrogen). Individual clones were sequenced and data was analysed using Sequencher for sequence alignment and MethylViewer to assign site‐specific methylation status (Pardo et al,
2011). Molecules with <97% conversion efficiency were excluded.
Statistical testing
Statistical analyses were performed in GraphPad Prizm 5.0 (GraphPad Software, La Jolla, CA). Statistical tests are indicated in the text. In all analyses, a p‐value < 0.05 was deemed significant. We used the D'Agostino–Pearson test for normal distribution of values. A Bonferroni Multiple Comparison test was applied to all ANOVA analyses post‐test. Observers were blinded to the patient data (including survival time) when performing tumour sample protein analyses. We used the R statistical software package (V. 2.15.0) to calculate descriptive statistics, create graphics and perform all analyses of patient specimens. To evaluate the possible associations between molecular subclass (mesenchymal, proneural or proliferative) and the other variables in the study (age, gender, mortality, KPS, ZEB1 and MGMT), we used chi‐square, Fisher's exact or Kruskal–Wallis tests, as appropriate. We used the Kaplan–Meier method to perform survival analyses comparing groups classified by ZEB1 or MGMT level, and to create survival plots. To estimate the effects of ZEB1 and MGMT when controlling for age and gender, we used Cox Proportional Hazards models. In all survival analyses, the outcome variable was time from start of treatment until death. Subjects still alive at the time or analysis and subjects lost to follow‐up were considered censored.