Glioma cell tissue culture
Ten GBM cell lines were selected for this study: seven established (U87MG, T98G (ATCC), U87ΔEGFR  G120, G112MS, SF763 and SF767 , and three primary cultures (GH3, 4, and 6) that were kindly provided by R. Goldbrunner, Dept. of Neurosurgery, Munich. Germany. All tissue culture was done using 10% FBS MEM and cells were kept at 37°C, 5% CO2 and 100% humidity. Primary cultures were used at or under passage 5, maintained in 20% FBS MEM and switched to 10% FBS MEM twenty-four hours prior to use in migration assay. All cell cultures were tested for, and found to be free of, Mycoplasma sp. infection by DAPI staining regularly.
Radial Migration Assay
The radial migration assay was performed as described previously [11, 12]. To simulate a GBM migratory front (rim) and proliferating core, three thousand glioma cells were seeded as a defined, confluent circular monolayer using cell sedimentation manifolds (CSM, Inc., Phoenix, AZ) on glioma-derived extra-cellular matrix coated 10-well slides. Migration was initiated by removing the manifold 4 hours after seeding and cells then radially dispersed for 24 hours. Migration rates were estimated from photomicrographs of migration assays taken at time point 0 hours and time point 24 hours .
For microarray experiments, stationary (core) and migratory (rim), cells were harvested under an inverse microscope (Axiovert 100, Zeiss, NY) using P2 pipette in three independent biological replicates. Thirty individual dispersion assays (three 10-well slides) were collected per cell line and all materials were stored at -20°C until RNA isolation. Core and rim cells were separately lysed with Trizol (Invitrogen, Carlsbad, CA).
RNA isolation, amplification and labeling
Total RNA was extracted from Trizol solution, according to manufacturer's instructions and followed with purification through RNeasy mini kit (Qiagen, Valencia, Ca). Total RNA quality and quantity were assessed by Bioanalyzer RNA 6000 Nano LabChip Kit (Agilent Technologies, Palo Alto, CA). Samples selected for the study contained intact RNA as determined by visual inspection of electropherograms for distinct 18S and 28S ribosomal peaks. Samples that failed this assessment were not included in the study.
Sample (either rim or core) and human universal reference RNA (Stratagene, La Jolla, CA) underwent one round of linear amplification and fluorescent labeling in Agilent's Low-Input Linear Amplification and Labeling Kits yielding fluorescent cRNA. Sample RNA was labeled with Cyanine 5 (Cy5) while reference RNA was labeled with Cyanine 3(Cy3) - CTP (PerkinElmer, Boston, MA) and cRNA concentration was assessed spectro-photometrically (GeneQuant II, Bottom Ltd., Cambridge, UK). One hundred nanograms of total RNA from each sample were labeled for hybridization to microarray chips and remaining unamplified sample RNA was saved for validation studies. Reference RNA was amplified/labeled in replicate reactions and pooled.
Microarray processing and quality testing
Each Cy5-labeled core and rim sample (three biological replicates) was hybridized against Cy3-labeled universal reference RNA on whole human genome 40 K oligonucleotide microarray chips (Agilent) according to manufacture's protocol . Agilent DNA microarray scanner and Feature Extraction Software with default settings for oligonucleotide expression arrays were used to capture and process array images. Spot intensities were extracted and preprocessed by subtracting background noise and applying a LOWESS (locally weighted linear regression) correction for dye-bias . Preprocessed data was then filtered by removing genes with a background signal higher than feature signal. Log2 (S/R) (S = sample (rim or core) and R = universal reference) for each gene was calculated.
Technical variation (i.e. variation due to instrumentation, assays and reagents) was assessed with two sets of technical triplicates. One collected at the beginning of the study and one at the end. A comparison of variation in technical and biological triplicates from the same cell line (T98G) was done to determine if technical variation was the same or greater than biological variation. Variation was qualitatively assessed by overlaying standard deviation versus average gene expression for each set of technical triplicates with their corresponding biological replicates. Expression values were log2 ratio of ratios, which is calculated by dividing rim ratio (rim/universal reference) by core ratio (core/universal reference).
A minimum fold change threshold was established by calculating the median standard deviation of rim/core expression for each biological replicate (n = 3). Genes that were not statistically (p > 0.1) differentially expressed were identified with ANOVA and removed. Genes that are not differentially expressed are considered non-informative and removed to reduce the complexity of the data.
In order to identify genes whose expression was consistently up- or down-regulated in rim cells as compared to their corresponding core cells across all cell lines, a pattern recognition approach integrating a priori knowledge about transcriptional differences between rim and core populations was employed. Principal component analysis (PCA) was used to determine components of variation in the data whereby patterns can be identified that discriminate samples (i.e. rim from core) [15, 16]. Genes whose profiles correlate (r > 0.6 and are visually similar) with the discriminatory PCA-derived patterns serve as probands and are used in the proband-based rule function within the Genetic Analysis By Rules Incorporating Expert Logic (GABRIEL), a platform of knowledge-based algorithms that incorporate biological knowledge to enable systematic microarray analysis, to find sub-sets of genes with similar expression patterns . One of GABRIEL's functions can readily identify additional genes whose expression profiles correlate with those of pre-selected proband genes. Both false discovery and false negative rates (FDR and FNR, respectively) are used to estimate significance of selected genes compared to random chance . Proband analysis was performed with the correlation coefficient threshold set at 0.8 for each proband. Genes identified were categorized according to their Gene Ontology into classes deemed significant for glioma biology.
Migration rates from the replicates of the 7 glioma cell lines and 3 primary cultures were averaged and binned into fast (>average migration rate) and slow migration (<average migration). Significance analysis of microarrays (SAM) was carried out on the union of migratory and stationary signatures derived from GABRIEL in order to identify significantly different genes between fast and slow migration groups. Support vector machines (SVM) with a linear kernel using leave-1-out crossvalidation were used to predict migration rate (e.g. fast or slow) based on gene signatures. A permutation test (10,000 iterations) was performed to assess the performance of the classification.
Clustering of the migration rate signature in the malignant astrocytoma cases was performed using k-means and visualized using multidimensional scaling (MDS) plots. Kaplan-Meier analysis were used to assess survival differences between clusters.
TM4  was used for SAM. Survival analysis was done with SPSS (SPSS Inc., Chicago, IL). ANOVA, PCA, SVM, k-means, and MDS was performed with MATLAB (MathWorks, Inc. Natick, MA)
Gene expression profiling of human brain tumor specimen
Gene expression data of 111 glial tumors and 24 normal brain specimens was kindly provided by Dr. T. Mikkelsen, Henry Ford Hospital, Detroit MI and Dr. H. Fine Neuro Oncology Branch, National Cancer Institute, National Cancer Institute, Bethesda, MD. Array data was processed according to Affymetrix® MAS5 (Microarray Suite 5) algorithm implemented in Affymetrix® GCOS (GeneChip Operating Software). Text files exported from GCOS were uploaded into GeneSpring®7.2 for data management (Silicon Genetics, Redwood City, CA).
Gene symbols and accession numbers were used to map Agilent probes to Affymetrix probes using Affymetrix's Netaffx tool.
QRT-PCR was performed with the unamplified RNA collected for use in microarray experiments. One hundred nanograms of RNA from each biological replicate were reverse transcribed (Supersrcipt III, Invitrogen) using oligo-dT primers. cDNA was diluted 1:4 in water and aliquoted until use. Primers were designed to hybridize in the 3'prime region of the genes and to produce amplicons ranging from 100 to 300 base pairs. Relative quantification was performed on the LightCycler Instrument (Roche, Mannheim, Germany) using LightCycler FastStart DNA Master SYBR Green I and normalized to Histone 3A as a housekeeping gene. Specificity of PCR amplicons was confirmed by melting curve analysis  and agarose gel electrophoresis.
Crossing points of target gene vs. housekeeping gene Histone 3A were used to calculate relative fold up-/down-regulation in the invasive cells with the following formula: F = 2(IH-IG)-(CH-CG), adapted from , where F = fold difference, C = core cells, I = invasive rim cells, G = gene of interest, H = housekeeping (Histone 3A).
In vitro protein expression of CTGF was assessed by immunofluorescence staining of migrating and stationary T98G cells in migration format. After induction of migratory phenotype for 24 hours cells were fixed with 4% paraformaldehyde, permeabilized with 0.5% triton X-100, blocked with 3% goat serum in tris-buffered saline (TBS) for 30 minutes, and incubated at room temperature for one hour with anti-CTGF antibody (SC-25440, Santa Cruz Biotechnology, Inc., Santa Cruz, CA) at 1:100. Addition of Cy3 conjugated secondary antibody (Jackson Laboratories, West Grove, PA) for one hour at room temperature (1:2000) was completed and cells were viewed with a LSM 5 Pascal laser scanning confocal microscope (Zeiss, Thornwood, NY).
A glioma invasion specific TMA  was used for immunohistochemical evaluation of proteins in stationary and invasive glioma cells. Briefly, the TMA was heated for 2 hours at 65°C, deparaffinized in xylene and hydrated in a graded alcohol series. Endogenous peroxidases were quenched with 3% hydrogen peroxide in PBS for 15 minutes, followed by antigen retrieval using reveal solution in the Decloacking Chamber (Biocare Medical, Concord, CA). Non-specific binding was blocked with 10% normal goat serum in 0.1% Triton X-100 TBS for 1 hour at room temperature. Slides were then incubated with CTGF antibody (1:200) at 4°C overnight, washed and secondary antibody (Vectastain Kit; Vector Laboratories, Burlingame, CA) was applied at room temperature for 1 hour. Slides were exposed to diaminobenzidine (Sigma, St. Louis, MO) for 1 minute and counterstained with hematoxylin 2 (Richard-Allen Scientific, Kalamazoo, MI). TMA was evaluated by a pathologist (S.N.) where percentage of cells exhibiting no(0), weak(+), moderate(++) and strong immunopositivity(+++) was assessed as described previously [7, 22]; Pearson chi-square test was employed to compare staining intensity in rim and core population.
Purified, duplexed siRNAs for CTGF and for luciferase (control) were purchased from Qiagen (Valencia, CA). The siRNA sequences targeting human CTGF (GenBank accession number NM_001901) were GTCCCGGAGACAATGACATCT (C1) and ATCGGAATCCTGTCGATTAGACT (C2). The sequences were designed to be unique when compared with the sequence of other CCN members. The sequence targeting luciferase was AACGTACGCGGAATACTTCGATT. Glioma cells (8 × 105) were transfected with 20 nM of siRNA using lipofectamine 2000 (Invitrogen) and cultured for 48 hours prior to use.
In-vitro migration assays were performed on 10-well slides coated with 10 μg/ml laminin.
Ex Vivo Invasion Assay on Rat Brain Slices
An ex vivo invasion assay on rat brain slices was carried out as described previously [13, 23]. Briefly, 400 μm thick sections were prepared from Wistar rat (Crl:(WI)BR; Charles River Lab, Wilmington, MA) cerebrum. Approximately 1 × 105 glioma cells stably expressing green fluorescence protein (GFP) were gently placed (0.5 μl transfer volume) on the brain slice in 4–6 replicates per condition. After 72 hours, glioma cell invasion into the rat brain slices was quantitated using a LSM 5 Pascal Laser scanning confocal microscope (Zeiss, Thornwood, NY). Serial optical sections were obtained every 5 μm downward from the surface plane to the bottom of the slice. The invasion rate was calculated as described previously . In brief, for each focal plane, the area of fluorescent cells was calculated and plotted as a function of the distance from the surface of the brain slice.
Antibodies and Immunoblotting
Glioma cells or glioma tissue specimens were lysed in sample buffer as described previously  and separated by 10% SDS-polyacrylamide gel electrophoresis, transferred to nitrocellulose membrane (Invitrogen) and probed with specific antibodies. Horseradish peroxidase-conjugated secondary antibodies were detected using a chemiluminescence system (NEN, Boston, MA). Following stripping, membranes were re-probed with α-tubulin antibody. Bound secondary antibodies were detected using a chemiluminescence system (NEN, Boston, MA). CTGF antibody was from Santa Cruz Biotechnology (Santa Cruz, Ca) and α-tubulin antibody was from Upstate Biotechnology (Lake Placid, NY). Horseradish peroxidase coupled secondary antibodies were purchased from Promega (Madison, WI).