Background
Invasion is arguably the feature of human glioblastoma (GBM) most responsible for their dismal outcomes with average survival less than 1 year. Diffuse tumor invasion into adjacent brain restricts curative resection and limits effective delivery of chemotherapy and radiation. In addition, migratory GBM cells can activate mechanisms that increase resistance to these therapies further compounding efforts to eradicate them. Despite the importance of glioma invasion, little is known about how this complex phenotype is regulated in gliomas, a prerequisite to development of effective anti-invasion therapies. By contrast, the process by which human epithelial cancers, or carcinomas, acquire an invasive phenotype has been more extensively characterized at both the cellular and molecular levels.
Carcinoma invasion and metastasis are driven by a process termed epithelial to mesenchymal transition (EMT) (For review see [
1]). Mesenchymal transitions lead to acquired potential for cell migration, changes in cytoskeletal organization, reduced cellular adhesion and changes in expression of transcription factors. Among the transcription factors that play fundamental roles in regulating these changes is the basic helix-loop-helix protein TWIST1. TWIST1 activates EMT in the context of embryonic morphogenesis [
2], tissue fibrosis [
3,
4] and cancer metastasis [
5‐
7]. A central feature of TWIST1-mediated EMT is the repression of the epithelial marker E-cadherin, and activation of the mesenchymal marker N-cadherin [
5‐
7], a hallmark feature of carcinoma EMT termed the "cadherin switch". The recent recognition of mesenchymal change in glioblastoma [
8‐
10] and its association with more aggressive clinical phenotypes [
8,
9] suggests that mechanisms that promote EMT in carcinoma may be of great clinical relevance in GBM.
We previously reported that TWIST1 is up-regulated in malignant gliomas and promotes glioma cell invasion of the SF767 glioma cell line
in vitro[
11]. However, the role of TWIST1 in promoting glioma invasion has not been investigated in the context of the brain microenvironment or as a mediator of mesenchymal change as occurs in carcinomas. In addition, the identification and clinical relevance of putative TWIST1 target genes in GBMs is not known. In this study we report that TWIST1 promoted GBM invasion through activation of mesenchymal molecular and cellular changes. This effect was not dependent on a "cadherin switch" indicating that TWIST1 promotes invasion through mesenchymal changes distinct from those associated with carcinoma EMT. The highly correlated expression of TWIST1 and mesenchymal target genes SNAI2 and FAP in human gliomas supported the clinical relevance of TWIST1 mesenchymal change. Together these results demonstrated an important role of TWIST1 in glioma invasion through activation of mesenchymal change and suggest its potential as a therapeutic target.
Discussion
The fundamental role for mesenchymal change in promoting invasion, malignancy treatment response and even cancer stem cell function in human carcinoma and GBM invasion is increasingly recognized [
8,
9]. TWIST1 is a central regulator of mesenchymal change in carcinoma [
5] but its relevance to invasion and mesenchymal change in GBM models has not been studied. Since tumor invasion is perhaps the major obstacle to improved outcome for patients with carcinomas and gliomas the elucidation of TWIST1 function in GBMs is potentially of great clinical importance. Following on our previous observation in the SF767 GBM cell line [
11] this study validated the pro-invasive function of TWIST1 in multiple cell lines
in vitro and
in vivo and demonstrated that TWIST1 promoted clinically relevant mesenchymal molecular and cellular phenotypes that partially recapitulated those associated with carcinoma EMT. These findings identify TWIST1 as a regulator of mesenchymal change and invasion in GBM that can be leveraged for further investigation of the clinical potential of subverting mesenchymal change as a therapeutic strategy in treating GBM.
Collectively TWIST1 promoted invasion
in vitro of all GBM cells tested to date (including a GBM stem cell line). We further established that TWIST1 enhanced invasion in the more relevant settings of brain slice culture and orthotopic xenotransplant models using SNB19 and T98G GBM cell lines. Of interest, the patterns of enhanced invasion generated by TWIST1 over-expression were cell-line specific with SNB19 TW cells invading as single cells or small aggregates from a central core while T98G TW cells diffusely invaded throughout the brain. These extreme patterns of invasion are similar to those in cases of gliomatosis cerebri [
33]. These findings clearly demonstrate the generic pro-invasive function for TWIST1 in GBM and suggest that cell-intrinsic factors can modify TWIST1- mediated patterns of GBM invasion.
Consistent with this, TWIST1 over-expression generated cell-specific changes in gene expression with shared pro-invasive functional attributes. TWIST1-mediated changes in expression of specific genes in SNB19 and T98G were heterogeneous but overlapped at the functional level within five common categories related to the cellular requirements for glioma invasion and EMT including cell adhesion, extracellular matrix, cell motility and locomotion, cell migration and actin cytoskeleton organization. Importantly, TWIST1 over-expression generated cell phenotypes highly consistent with the over-representation of genes within these functional categories that reflect critical individual cellular features required for carcinoma and GBM invasion [
1]. We also determined that TWIST1 induced re-localization of activated FAK to sites of abundant lamellipodia formation, a significant finding given the association between FAK activation, cytoskeletal organization and its role in EMT and glioma malignancy (reviewed in [
34‐
36]. Loss of apical-basal polarity (relative to a basement membrane) is an additional feature of EMT in carcinomas which was not tested here since assays of polarity for GBM cells
in vitro are not well established. However, the recent description of polarized ciliated neural stem cells within the ventricular zone neuroepithelium [
37‐
39] suggests that such studies could be attempted
in situ or with novel co-culture systems or at earlier stages of glioma development. This approach could reveal polarity changes (analogous to carcinoma EMT) as fundamental steps in the process of gliomagenesis and acquisition of an invasive phenotype. Together gene expression analysis and cellular assays demonstrated that TWIST1 over-expression in glioma cells orchestrated the acquisition of a robust mesenchymal phenotype and cellular changes that closely mirror those of carcinoma cells undergoing mesenchymal transformation [
40] and required for tumor invasion and metastasis [
41].
TWIST1-mediated molecular changes also provided important insight into its role in mesenchymal change in GBM. Many genes related to carcinoma EMT were also up-regulated by TWIST1 in GBM indicating potential mechanistic overlap between the two processes. However, the lack of a TWIST1-mediated "cadherin switch" in GBM cells suggested that alternative mechanisms in nervous tissue and gliomas function to modulate cell adhesion and invasion. Alternatively, a cadherin switch could occur early in gliomagenesis or require specific anatomic or environmental interactions not present in our experimental system. The recent discovery that normal neural stem cells -- putative GBM cells of origin -- express E-cadherin supports this possibility [
42,
43]. Further studies are warranted to examine the impact of TWIST1 and other factors related to mesenchymal change in normal GBM cells of origin (neural stem and progenitor cells) or in cells at early stages of gliomagenesis to better define how alterations in E-cadherin or other cell-cell adhesion molecules impact the acquisition of an invasive malignant phenotype.
The clinical relevance of identified putative TWIST1 targets was established through correlation between TWIST1, SNAI2 and FAP expression levels in 39 human gliomas of different grades. These studies demonstrated that the current
in vitro model of TWIST1 pro-invasive function was capable of identifying clinically relevant pro-invasive targets and candidate downstream mechanisms of TWIST1-mediated glioma invasion. Our data also confirms prior reports that expression of SNAI2 [
31] and FAP [
44] is directly linked to malignant glioma grade and further showed that they are coordinately upregulated in gliosarcoma, the grade IV glioma with the most overt mesenchymal differentiation. As regulators of invasiveness, TWIST1 and SNAI2 are potential targets for therapeutic modulation, a proposition further supported by their known functions to promote cell survival and treatment resistance in other cancer types [
45‐
50]. FAP is expressed in wounds and fibrotic tissues as well as carcinoma-associated fibroblasts in multiple cancer types and is thought to degrade tumor matrix and facilitate carcinoma invasion [
51]. Further studies are needed to determine which cell type(s) express FAP and whether it serves a similar role of altering tumor stroma to promote invasion in GBM.
The significance of TWIST1 function to promote invasion through mesenchymal change in GBMs is underscored by recent reports of clinically relevant mesenchymal phenotypes in GBMs. Gene expression array studies identified a mesenchymal stem cell (MSC) phenotype in human GBMs [
10] and distinct pro-neural, proliferative and mesenchymal gene expression signatures among malignant grade III and IV human gliomas [
9]. The mesenchymal signature is associated with poor prognosis, increased angiogenesis and tumor recurrence [
9]. Therefore, along with other transcription factors such as STAT3 and C/EBP which were recently identified as regulators of mesenchymal transformation in GBM cells [
8] the correlation of TWIST1 with induction of mesenchymal changes, increased glioma grade and invasiveness implicate TWIST1 as an additional central regulator of this process in human GBM. Of note, STAT3 transcriptionally upregulates TWIST1 expression and promotes breast carcinoma cell migration [
52] prompting speculation that STAT3-TWIST1 interactions in GBM may also contribute to invasion and mesenchymal change.
Inhibitors of TWIST1 are not available; therefore, to investigate the therapeutic relevance of inhibiting TWIST1 in GBM we knocked down TWIST1 expression using shRNA and assayed its effects on cell invasion and glioma stem cell properties. Specific inhibition of TWIST expression resulted in marked reductions in glioma cell invasion
in vitro. These findings are consistent with the pro-invasive function of TWIST1 in GBM and support the therapeutic potential of inhibiting TWIST1 or TWIST1-mediated signaling to inhibit GBM invasion. Glioma stem cells are recognized as tumor-initiating cells that determine tumor malignancy and growth. Through activation of EMT, TWIST1 promotes the formation and maintenance of breast cancer stem cells [
53] and TWIST1 over-expression is implicated in mesenchymal stem cell activity [
54]. Given these observations we propose that targeting TWIST1 may have additional therapeutic relevance in gliomas by abrogating glioma stem cell functions. Our data showed that inhibition of TWIST1 expression resulted in a dramatic reduction in GBM stem cell sphere formation and growth. These results suggest that a unique therapeutic potential of inhibiting TWIST1 may result from simultaneous targeting of glioma cell invasiveness and stem cell function -- hallmark GBM properties that both contribute to tumor growth, progression and treatment resistance. To address this potential, ongoing and future studies will address the effects of TWIST1 inhibition in GBM cells on tumor growth, invasion and response to therapy
in vivo.
Methods
Cell lines and tissue
Glioblastoma cell lines T98G, SNB19, SF767, U87MG were maintained in DMEM/F12 with 10% FBS (Hyclone). Human primary GBM cancer-initiated cells (GBM4, GBM6) were cultured as described [
12] in the presence of EGF and bFGF. Human glioma tumor samples were acquired according to a protocol approved by the Institutional Review Board of the Human Subjects Division of the University of Washington. Samples were immediately snap frozen in liquid nitrogen and stored at -80°C before processing. Type and grade of tumors were confirmed by histopathological examination.
Expression and shRNA constructs, and cell transduction
A retroviral human TWIST over-expression construct and methods for infection of SNB19 and T98G cell lines were described previously [
11]. Myc-tagged SNAI2 expression construct (SNAI2myc) was generated by PCR followed by subcloning in LXSN expression vector. Exogenous protein expression was confirmed by Western blot analyses with corresponding antibody. Lentiviral shRNA construct for inhibition of TWIST1 expression [
5] and control shRNA were purchased from Addgene. Lentivirus was generated using a standard method in HEK293T cells. A pool of infected cells was selected with Puromycin (1 μg/mL).
Cell aggregation assay
Single-cell suspensions (105 cells/mL in DMEM-F12 without FBS) were plated into each well of a 6-well plate coated with 0.6% agarose/DMEM-F12. The plate was incubated at 37°C on a rocking platform for 16 hrs. Cells were fixed in 5% formalin to preserve cell-cell interactions and photographed.
Cell adhesion
SNB19 LXSN and TWIST cells (5 × 104/1 mL) were allowed to adhere to 24-well plates coated with BSA or fibronectin (5 μg/mL) for 1 hour. Cells were then washed, fixed, stained and counted. Differences in cell adhesion are shown as percent of SNB19 control cells attached to BSA-coated wells. Three wells from 3 separate experiments were analyzed and the significance of differences was determined by Student t-test. Data shown are mean ± SE.
GBM stem cell sphere assays
GBM6 and GBM8 stem cells with TWIST1 knockdown and control cells (scrambled shRNA lentiviral vector) were dissociated and viable cells were counted using ViCell. Viable GBM6 cells plated at 3200 cells per well in 6-well plates were used to establish the effect of TWIST1 knockdown on sphere size. After 5 days spheres were photographed and sphere sizes were measured using Adobe Photoshop. To determine the effects of TWIST1 on sphere-forming activity GBM8 cells were plated at clonal dilution (20 viable cells per well in 96-well plate). After 7 days, wells with spheres were counted and presented as a percent of wells with spheres. Average numbers of spheres per well with spheres were also calculated. Fisher exact test and t-test were used for statistical analysis as appropriate.
Immunocytochemistry and F-actin staining
Cells were grown in 8-well chambers and fixed in 4% PFA for 10 min. Following treatment with 0.1% Triton X-100/TBS for 5 min, cells were blocked in 1%BSA/TBS, washed and incubated with FAK or phospho-FAK antibody (Upstate) according to manufacturer protocol. Appropriate secondary antibody conjugated with FITC (Pierce) was used for antigen detection. F-actin was stained with TRITC conjugated phalloidin (1 μg/mL) followed by DAPI (1 μg/mL) staining. Antigens were visualized using confocal microscopy (Delta-Vision).
Western blot analysis
Cell harvesting, cell lysis and Western blot procedures were performed as described previously [
11]. Total cell lysates were used for detection of TWIST1 over-expression. Nuclear extracts were used to increase assay sensitivity in detecting endogenous TWIST1 expression. For protein loading control, anti-β-actin antibody (Sigma) for total proteins or Ini1 (H-300) antibody (Santa Cruz) for nuclear proteins were used. Immunoblot for MycTag Antibody (Upstate) was performed according to manufacturer's recommendation.
Invasion and migration assays
The invasion and migration assays were performed using 24-well Matrigel invasion chambers or uncoated Control inserts (BD Biosciences) as previously described [
11]. Briefly, cells were resuspended in a serum-free DMEM and loaded into inserts (5 × 10
4 cells/500 μL). DMEM/F-12 with 10% FBS (SNB19) and without FBS (T98G) was added to the lower chamber (750 μL). Following incubation at 37°C, cells that invaded or migrated to the underside of the membrane were fixed, stained, digitally imaged and counted. Differences in cell invasion were expressed as a percent of invading/migrating relative to control cells. Data shown are mean ± SE.
Invasion in organotypic brain slices
Coronal brain slices (400 μm) from 21-day-old mice were cultured as previously described in the media supplemented with 10%FBS [
55]. SNB19 or T98G cells with LXSN or TWIST1 over-expression (labeled with GFP expressing lentivirus) were placed at the corpus callosum. After 10 days
in vitro, the co-cultures were fixed and analyzed by confocal microscopy. Slices were imaged using a FluoView FV1000 confocal microscope (Olympus). Collected Z-stacks were processed for visualization and cell counting using NIH Image software. The morphometric software Metamorph (Molecular Devices Corporation) was used to measure cell migration. The distances from the border of the cells' aggregate to each of the 20 furthest cells were measured for each of the 9 tissue slices (total n = 178 measurements) and compared using a repeated measures ANOVA model. The modeling was done with the 'proc mixed' procedure available in SAS.
Intracranial injection and whole brain imaging
Animal experiments were performed according to procedures approved by the University of Washington IACUC. Glioma cells harboring empty vector and cells with TWIST1 over-expression were labeled with GFP-expressing lentivirus (pLL3.7) prior to implantation. T98G cells were injected in 9- to 10-week-old SCID-NOD mice. SNB19 cells were injected in 7- to 8-week-old nude mice. Following animal sedation 3 × 105 labeled cells were injected intracranially into the right caudate nucleus using a stereotactic apparatus and a Hamilton syringe. To determine the effect of TWIST1 on tumor growth and invasion, animals (6 mice with SNB19 TW and 4 mice with SNB19 Ctrl) were sacrificed 17 days after injection when the animals first showed signs of morbidity. Animals injected with T98G TW or control cells (3 mice per group) were sacrificed 90 days after injection. Animals were perfused with 4% paraformaldehyde (PFA). The entire brain was dissected from the calvarial vault and fixed for an additional 24 hours in 4% PFA at room temperature with light agitation. Brains were washed with PBS and transferred to 50% glycerol in PBS for 24 hours, 75% glycerol for 24 hours, then 90% glycerol at 4°C until imaged. The whole brain was sliced in the axial plane to obtain 3 slices, each approximately 2 mm thick. Each slice then was imaged using the FV1000 laser scanning confocal microscope in the axial plane to detect GFP-expressing cells and perform automated image splicing to reconstruct the entire tumor in a single axial slice.
Image analysis for invasion
A semi-quantitative scale was used for initial characterization of tumor growth patterns as follows: Type 1 = solid compact core with mainly localized expansile growth; Type 2 = easily detectable non-contiguous individual cells or cell clusters invading into brain parenchyma adjacent to the solid tumor core; Type 3 = poorly defined or absent core with diffusely invasive contiguous or non-contiguous growth. Type 3 represents the most invasive growth pattern. For circumstances where the scale did not provide clear indications for the degree of invasiveness, differences in tumor growth pattern were quantified as follows: Reconstructed wide-field brain images with tumors were analyzed using Huygens software (Scientific Volume Imaging, Hilversum, The Netherlands). Individual signal intensities from each individual optical section (Z-stack images) collected by confocal microscopy were integrated into a brightest point projection image (BPI) to provide a 2-dimensional summary of total tumor cell density and spatial distributions (Image J). Information collected includes tumor core volume, invasive cell volume and a number of invasive aggregates. Numerical values were compared using t-test.
Microarray processing methods for the Affymetrix microarray platform
Gene expression experiments were performed using the GeneChip platform by Affymetrix (Santa Clara, CA) and the manufacturer's protocol. Statistical analysis and data normalization for the Affymetrix arrays were carried out with Bioconductor software [
56], and GeneTraffic
® (Iobion Informatics LLC, La Jolla, CA). Modified t-test was applied for two-group comparison. Bioconductor was used to calculate p-values using a modified t-test in conjunction with an empirical Bayes method to moderate the standard errors of the estimated log-fold changes. P-values were adjusted for multiplicity with the program q-value. Genes with absolute change greater than or equal to 1.5 fold and p < 0.05 were considered differentially regulated by TWIST1. To demonstrate the correlation of changes in expression (up or down-regulation) between genes differentially regulated in both cell lines, we applied a Pearson's correlation analysis using Bioconductor software (see above). Gene Ontology (GO) categories were analyzed using GoMiner [
57] to detect gene category over-representation with cut off false discovery rate (FDR) 0.1.
Quantitative RT-PCR (qRT-PCR)
Total RNA from cells, brain or tumor samples was extracted using Qiagen RNeasy mini kit. RNA (1 μg) was reverse-transcribed with Clontech kit. SYBR Green PCR Master mix (ABI) was used for template amplification. Thermocycling for all targets was carried out in 30 μL reaction for 40 cycles in triplicate. Each cycle consisted of: 94°C for 15 seconds, 58°C for 30 seconds and 72°C for 30 seconds. For all samples, reactions were run in triplicate. PCR reactions where reverse transcriptase was omitted were used as negative controls. SYBR Green incorporation was monitored in real time with an ABI PRISM 7000 sequence detection system (Applied Biosystems) and threshold exponential amplification cycle (CT) was calculated by SDS system software. Differences in the CT values (ΔCT) between the target transcript and GAPDH endogenous control determined the relative gene expression level and the ΔΔCT method was used to calculate fold differences in expression. Relative expression in each tumor sample is normalized by expression of corresponding target in a pool of normal brain samples (n = 4). Correlations between TWIST1 and SNAI2 or FAP expression were calculated using a regression coefficient and ΔCt values for each tumor sample. Statistical analysis of relative expression levels in human tumors was performed using unpaired t-test. Relative expression values were log transformed before testing to ensure normal distribution. Absolute quantification was used to compare levels of E- and N-cadherin mRNA expression. Standard curve was built using plasmid harboring corresponding cDNA targets diluted from 0 to 107 copies per reaction. Final results are shown as a number of copies per μg of total RNA. The specificity of amplifications was confirmed by amplicon melting profile.
Acknowledgements and Financial Support
We acknowledge Rosemary Kimmel for her expert editorial assistance, J. Barber for the help with statistical analysis, Drs Jing Zhang, Eduardo Mendez and Daniel Silbergeld for critical review of the manuscript, Theo Bammler and Frederico Farin of the UW CHDD Genomics Core for assistance with microarray studies, Glen MacDonald of the CHDD Cellular Morphology Core for assistance with development of whole brain tumor imaging techniques, Drs Daniel L Silbergeld, Alexander M Spence, Jason Rockhill, and Maciej Mrugala whose clinical efforts made this work possible. This research was funded in part through an NIH/NINDS T32-NS-0007144 Clinical Neuroscience Training Grant (RO, LK, JM) and a University of Washington Institutional Bridge Funding Grant (RCR, AKM). Research in ISG group is supported partially by FEDER and by MICINN (SAF2009-08803), Junta de Castilla y León (CSI13A08 and proyecto Biomedicina 2009-2010), MEC OncoBIO Consolider-Ingenio 2010 (Ref. CSD2007-0017), Sandra Ibarra Foundation, NIH grant (R01 CA109335-04A1) and by Group of Excellence Grant (GR15) from Junta de Castilla y Leon. There are no conflicts of interest.
Competing interests
The authors declare that they have no competing interests.
Authors' contributions
SAM generated expression constructs and performed in vitro cellular assays, qRT-PCR and microarrays. AMM carried out in vivo studies, confocal imaging, microarray interpretation and statistical data analysis. AP performed ex vivo invasion assays. RB performed microarray analysis and bioinformatics support. RGO and LK participated in tumor collection and assisted with cell culture experiments. JPM participated in generation of recombinant DNAs and analysis of gene over-expression and knockdown. CAG generated anti-TWIST1 antibody. HW generated primary GBM cells. IG-H and IS-G generated reagents for Snai2 analysis. JRS and PJH assisted in experiment design, data interpretation and manuscript writing. RCR conceived of the study, and participated in its design, coordination and manuscript writing. All authors read and approved the final manuscript.