Acquisition of an angiogenic phenotype is a necessary step during human tumor progression.1 Therefore, inhibiting angiogenesis has become a promising strategy to complement standard chemotherapy. Vascular endothelial growth factor (VEGF) and its receptors are key regulators of blood vessel growth and the main target of current anti-angiogenic drugs. Inhibition of VEGF function alone is sufficient to stop or slow-down solid tumor growth as well as metastasis formation in numerous animal models.2, 3 Blocking VEGF can also lead to tumor regression of established tumors in mice4 and to reduction of blood vessel density in human tumors.5 In the long-term, anti-angiogenic therapy leads to “tumor dormancy,” a biological state balanced by cell proliferation and apoptosis of cancer cells.6 However, targeting only one angiogenic pathway (e.g., VEGF) might not be sufficient to cure. There is evidence that tumor hypoxia caused by angiogenesis inhibition actually increases malignant potential of surviving cancer cells. Tumors often persist,7 show increased vessel cooption and invasion8 or switch on alternative pro-angiogenic pathways.9 Consequently, impairing two anti-angiogenic pathways at a time is more efficient than monotherapy.10, 11
Given the heterogeneous responses to anti-angiogenesis in animal models and the increasing use of anti-VEGF therapy in humans (Bevacizumab, Avastin™), it is important to comprehend how tumor cells react to such therapies on a molecular level. This knowledge most certainly will help to identify new “druggable” targets12 for adjuvant anti-angiogenic intervention. Knock down of VEGF by RNA interference (RNAi) strongly slow tumor growth and angiogenesis of in various animal tumor models.7, 13–17
We have previously shown that U87 glioma cells deposited on the chick chorio-allantoic membrane (CAM) form highly vascularized tumors in a VEGF-dependent manner, and recapitulate key features of glioblastoma in patients on a morphological and molecular level.18 To generate a defined anti-angiogenic in vivo environment we transfected U87 glioma cells with short-interfering RNAs (siRNAs) targeting all isoforms of VEGF-A and implanted them on the CAM.
An advantage of the model is enhanced visual control of morphological characteristics of the tumor (e.g. size, prominence, vascularization) prior to selection of samples for gene expression studies. This eliminates contamination with adjacent, nonrelevant tissue. Furthermore, the use of human oligonucleotide microarrays (Affymetrix) maximizes specific hybridization of human transcripts regulated in tumor cells and not of mRNA present in associated supporting cells like endothelial cells, pericytes and fibroblasts, which are of chick origin. This approach maximizes the chance to identify genes, which are specifically regulated in tumor cells in response to anti-angiogenesis.
Cells and embryos
U87 human glioma cells (ATCC/LGCpromochem, Molsheim, France) were maintained in DMEM (Invitrogen, Cergy Pointoise Cedex, France) with 10% fetal bovine serum, antibiotics and L-glutamine. Fertilized chicken eggs (E.A.R.L. Morizeau, Dangers, France) were handled as described.18
Vascular endothelial growth factor RNA interference and transient transfection of U87 cells
Small interfering RNAs (siRNAs) were purchased from Eurogentec (Liege, Belgium). siRNA2 targets all isoforms of VEGF-A (sense: 5′-GGAGUACCCUGAUGAGAUC-TT14). A siRNA with no human sequence match served as negative control. U87 cells were transfected with siRNAs in Opti-MEM using LipofectAMINE PLUS, according to manufacturers indications (Invitrogen; final concentration 100 nM). After 24, 48, 72 or 96 hr, conditioned medium or cells were collected for ELISA or RNA isolation.
Experimental glioma model (chick embryo)
siRNA transfected U87 cells or untreated cells were deposited in plastic rings on the chick CAM at E10 (= tumor day 0; n = 336), as described.18 Digital photos were taken every other day using a stereomicroscope (Nikon SMZ800). Tumors were harvested 2, 4 or 7 days after implantation. Some VEGF siRNA tumors (n = 30) were treated topically with 3 μg of recombinant human VEGF165 protein 1 and 2 days after implantation.
RNA extraction and reverse transcriptase (RT)-PCR
Total RNA from tumor tissues were extracted using the Qiagen RNeasy mini kit (Qiagen, Courtaboeuf Cedex, France), and verified by electrophoresis. First strand cDNA was prepared from 1 μg of RNA by reverse transcriptase (SuperScript II Rnase H, Invitrogen) in the presence of oligo (dT)15 primer (Promega, Charbonnières, France).
Semi-quantitative real-time PCR and VEGF mRNA detection
Reduction of VEGF transcripts was determined by quantitative real-time PCR (Mx3000P thermocycler, Stratagene, La Jolla, CA) using SYBR Green dye (ABgene, Billebon-sur-Yvette, Courtaboeuf, France). Mean ± S.E.M. of 3–4 independent experiments is shown. Human-specific primers were designed using Primer Bank19 (supplementary Table II) and evaluated for amplification efficiency using Universal Human Reference RNA (Stratagene). Only primer pairs with amplification efficacy between 90 and 110% were used. For quantification of relative expression levels the ΔΔCt method was used (normalization genes: ribosomal protein S16 or alpha-tubulin). Each gene was tested in at least two independent experiments. For patient tumor sample studies, normal human brain (Stratagene) was used as reference tissue.
VEGF proteins levels
Control and VEGF siRNA tumors at day 2 and 4 (pool of 3–5 tumors/group) were homogenized in lysis buffer (CelLytic-MT, Sigma-Aldrich, Lyon, France) in presence of a protease inhibitor cocktail (Sigma). VEGF-A protein concentration was measured using the human VEGF-A DuoSet ELISA kit (R&D Systems, Lille, France) and normalized to tissue weight. In vitro, number of viable cells was used for normalization (WST-1 assay Roche, Neuilly sur Seine Cedex, France). The assay was repeated twice with similar results, one representative experiment is shown. For determination of VEGF levels over time, experiments were performed in duplicate wells; all values are shown (Fig. 1d).
In vitro assays: proliferation, migration, hypoxia and serum deprivation
Cell proliferation was measured with the WST-1 assay (Roche), according to the manufactures indications. Cells were grown under the same conditions as described above and proliferation was determined after 24, 48, 72 or 96 hr.
Migration of U87 cells across a denuded cell culture dish area (“wounding assay”) was performed as described earlier.20
U87 cells were exposed to moderate hypoxia (3% O2) or 0% serum and RNA was isolated after indicated time points for qPCR analysis.
In vivo growth parameters of VEGF siRNA tumors
Sixteen control and 15 VEGF siRNA tumors were photographed under a biomicroscope and tumor volumen was estimated using established criteria described.18
Tumor angiogenesis was measured using a semi-quantitative approach. Each tumor was scored for red color appearance (coincidences with vascularization) from 0 to 2. Statistical analysis was performed using the Mann–Whitney U-test. Implantation frequency of control and VEGF siRNA tumors was determined from several independent experiments at day 4 and 7 after tumor cell implantation.
Histology and immunochemistry
Tumors were fixed at indicated days in vivo with 4% paraformaldehyde, and processed for cryo-sectioning. Ten micrometer sections were stained with hematoxylin and eosin. For immunohistology studies, the following primary antibodies were used: mouse anti-human vimentin Ab-2 (1:400, clone V9, NeoMarkers Ab, Interchim, Montluçon, France), anti-human Ki-67 (clone MIB-1, 1:100, Dakocytomation, Trappes, France), mouse anti cytochrome c (clone 6H2.B4, BD Pharmingen, 1:200, Erembodegem, Belgium) and fluorescein-coupled Sambucus nigra lectin SNA-1 (1:500, AbCys, Paris, France). Cell nuclei were visualized by Hoechst dye. Corresponding secondary antibodies were from Molecular Probes (Invitrogen). Elafin was detected in paraffin-embedded sections of GBM and low-grade glioma by immunohistochemistry using a rabbit polyclonal elafin antibody (1:80, sc-20637, Santa Cruz Biotechnology, Tebu-Bio, Boechout, Belgium). Sections were stained using Vectastain ABC peroxidase Kit (Vector Laboratories, Burlingame, CA and counterstained with hematoxylin. For negative controls, nonimmune rabbit serum was used.
In vivo cell proliferation and mitochondrial morphology
Ki-67-positive cells were counted in 3 fields of 6 different VEGF siRNA and control tumors on Ki-67/Hoechst/S. nigra (SNA) stained cryosections at 20× magnification and divided to the total number of cell nuclei in the field to account for differences in cell number on a given section (Lucia G/F software/Nikon fluorescence microscope). Cytochrome c (cyt c) signal distribution was quantified by setting the contrast of immunohistochemistry images to 100% (Adobe Photoshop CS2), distributing all pixels into 4 channels (RGB and black). Pixels of a given channel are correlated to the stained area covered by a given antigen. Eight fields of 2 typical control tumors were compared to 29 fields from 7 typical VEGF siRNA tumors at 40× magnification. Statistical comparison was done using the Unpaired t test (two-tailed) after successful testing for normality of the data.
Complementary RNA (cRNA) preparation and microarray hybridization
Pooled RNA from VEGF siRNA (n = 19) and control tumors (n = 7) tumors at day 4 were extracted using the RNeasy mini kit (Qiagen). Tumors as shown in Figures 3f and 3g are typical examples of material used for comparison in the microarray experiment.
RNA quality was assessed on an Agilent 2100 Bioanalyzer (Agilent, Massy Cedex, France) and a Nanodrop ND-1000 spectrophotometer (Nyxor Biotech, Paris, France). cRNA was prepared according to the manufacturer's protocol “sample protocol I,” and hybridized to HG-U133 A 2.0 GeneChip oligonucleotide arrays representing 18,400 genes (Affymetrix, UK, High Wycombe, UK). Image analysis was performed using GeneChip Operating Software 1.2 (GCOS, Affymetrix) and data scaled to a target value of 100 using the “global scaling” method.
Microarray data filtering and bioinformatics analysis
Samples were analyzed using a pair wise comparison with GCOS 1.2 software (Affymetrix). A gene was considered as over- or under-expressed when all three possible pair wise comparisons showed a significant change p-value (p < 0.05), when the gene was indicated as “P” (present) in the reference gene set, when the difference in expression ratio was ≥1.5 (fold change) and when the fluorescence detection signal was >50. Gene annotation was performed using the NetAffx™ Analysis Center. Ambiguous probe sets (those which map to more that one gene) were excluded. Expression of 21 selected genes was validated by qPCR using the same mRNAs as used for the microarray.
Gene lists were uploaded to the WebGestalt analysis tool21 and KEGG tables were generated for up- and down-regulated gene lists. Each gene set was compared to all genes present on the HG-U133 A 2.0 GeneChip and KEGG pathways associated with the query gene set were extracted. Enrichment of a given pathway was considered significant if p < 0.05 and if at least 5 genes were associated with the pathway (Hypergeometric test).
Patients and tumor samples
Glioblastoma (GBM) Samples have been obtained during surgery (time of initial diagnosis) from the CHR Bordeaux, University Bordeaux 2 (provided by H. Loiseau); low grade (LG) glioma samples patients (9 World Health Organization (WHO) grade II oligodendroglioma, 1 WHO grade II astrocytoma) were from the University of Milan, Ospedale Maggiore Policlinico (provided by L. Bello) in compliance with institutional ethical guidelines. Median survival of GBM patients was 343.5 days (31–1,395 days), 2 patients are still alive; 1 patient died 1 month after surgery (censored subjects). Median age of GBM patients was 62.5 years (21 women, 21 men) and 38.5 years (5 women, 5 men) for patients with LG glioma. For immunohistological studies, paraffin-embedded tissue sections of 20 diffuse gliomas were obtained from the Laboratory of Neuropathology, Department of Pathology, University of Liège, Belgium including 10 GBM, 4 anaplastic astrocytoma (WHO grade III), 1 low grade astrocytoma (WHO grade II), 2 anaplastic oligodendroglioma (WHO grade III) and 3 low grade olidendroglioma (WHO grade II).
Gene expression correlation and survival analysis
Expression levels of 6 highly up-regulated genes in VEGF siRNA tumors (CHI3L1, CHI3L2, IL1B, EGLN3, APOC1 and PI3, for primers sequences see supplementary Table II) were determined in the tumor samples and compared to their levels in normal brain (Stratagene). Samples, for which no threshold cycle could be determined, were excluded for that particular gene. Values of CHI3L2, IL1B, EGLN3, APOC1 and PI3 were compared with those of CHI3L1, which encodes for the poor prognosis marker YKL-4022 using nonparametric Spearman's correlation. To determine whether expression of the selected genes differs between GBMs and LG tumors, mRNA levels were compared using the nonparametric Mann–Whitney U-test. Relation to survival of selected genes was determined using Kaplan and Meier curve analysis. A cut-off value based on the relative expression level was assigned for each gene (CHI3L1 < or > 10, CHI3L2 < or > 10, IL1B < or >20, APOC1 < or >10, PI3 < or > 100) and accordingly, GBM patients were divided into two groups. The log-rank test was used to evaluate if survival curves differed significantly and the hazard ratio (HR) was calculated for each gene expression. Odds derived from the HR were calculated according to Spruance et al.23
To determine if PI3 and CHI3L1 were independent markers from other prognostic factors, a Cox regression for survival analysis was performed. Survival analysis was adjusted for age (continuously) and sex for PI3 or CHI3L1 using the same cut-off values as for Kaplan–Meier analysis, and HRs were calculated (SAS V9.1, Cary, NC).
Vascular endothelial growth factor siRNAs efficiently reduce target mRNA and protein levels over 4 days
Three siRNAs specific for human VEGF-A were designed and tested for their efficacy to reduce VEGF mRNA production. Transfection of U87 glioma cells with either siRNA leads to reduction of mRNA levels to more than 60% compared to controls (Fig. 1a). When tested for effects on VEGF protein production, siRNA 2 showed greatest reduction (73.6% and 82.5%, 2 experiments) (Fig. 1b). This siRNA was further used in this study. VEGF mRNA and protein levels were determined each day after transfection. mRNA levels stayed below 55% of controls for the first 3 days and were 60% of controls at day 4. Protein levels were around 20% of controls the first 2 days, 45% on the third day and 55% after 4 days; Figures 1c and 1d. Transient transfection with VEGF siRNA thus leads to suitable strong and prolonged VEGF reduction for short-term in vivo experiments and had no effects on U87 cell proliferation or migration in vitro (data not shown).
Characterization of VEGF siRNA tumors
We next determined levels of VEGF protein in VEGF siRNA tumors in vivo. VEGF siRNA tumors contained no detectable amounts of VEGF compared to control tumors, which had about 15 ng VEGF per mg tissue at day two of development and more than 30 ng/mg at day 4 (Fig. 2a). Tumors, which had not been treated at all, produced similar amounts of VEGF protein as control siRNA-transfected tumors. We evaluated general growth characteristics of experimental glioma by estimating vascularization and tumor volume. VEGF siRNA tumors appeared white by biomicroscopy examination (angiogenesis score 0.4) whereas controls had a marked red color (score 1.875). Control tumors had also about 10 times the volume of VEGF siRNA tumors (Figs. 2b and 3).
Striking differences were observed concerning the implantation success of VEGF siRNA cells compared to controls (Fig. 2c). 15.5% (11/71) of chick CAMs implanted with VEGF siRNA cells did not show any sign of tumor growth after 4 days compared to only 5.6% (2/36) of controls. After 1 week, more than 50% (20/39) of CAMs implanted with VEGF siRNA cells had not formed tumors, compared to only 20% (5/24) of control CAMs. Topically applied recombinant human VEGF protein at day 1 and 2 after VEGF siRNA tumor cell deposition increased tumor implantation frequency after 4 days to 100% (11/11) and to 89.5% (17/19) after one week.
We determined the proliferation index of VEGF siRNA tumors on cryosections stained with Ki-67 antibody and Hoechst dye. VEGF siRNA tumors had about 50% less cells in proliferation (mean: 13.43 vs. 24.68) than control tumors (p = 0.0002; Fig. 2d). Permeabilization of mitochondrial membranes during apoptosis leads to the release of cytochrome c from the intermembrane space.24 Whereas glioma in control tumors showed a punctuate perinuclear distribution of cytochrome c (Fig. 2e and Ref.18)—consistent with absence of apoptosis—VEGF siRNA tumors showed a diffuse staining pattern of cyt c, often extending into the cytoplasm. We quantified the area covered by cyt c immunoreactivity and found a significant increase in VEGF siRNA tumors (p = 0.0241). These results suggest an increase in apoptosis in VEGF siRNA tumor cells, consistent with the consequence of anti-angiogenesis.
Morphological characteristics of VEGF siRNA tumors
Untreated controls and siRNA control tumors showed similar development with an apparent angiogenic switch during the second day after implantation as revealed by biomicroscopy (Fig. 3). Even though tumor size varied sometimes during the first days in the control groups, all tumors were fully vascularized at day 4 and vascular morphology did not change further until day seven, when the experiment was ended. In contrast, VEGF siRNA tumors had a whitish appearance from the second day on, a constant feature, which became even more evident the following days. When VEGF siRNA tumors were treated with recombinant human VEGF, new blood vessels invaded the tumor after a short delay compared to control tumors, and vessels were clearly distinguishable at the surface by day 7.
Standard histology confirmed the absence of vessels in VEGF siRNA tumors. Controls and VEGF siRNA tumors treated with VEGF protein were filled with typical, tortuous tumor blood vessels and capillaries. Immunohistology of cryosections showed S. nigra (SNA)-positive blood vessels and tumor cells positive for vimentin in all controls and VEGF siRNA tumors treated with VEGF. In VEGF siRNA tumors, only isolated, SNA-positive structures were seen, but their morphology did not correspond to functional blood vessels. Furthermore, these SNA-positive cell structures did not contain nucleated chick erythrocytes as seen in controls. However, VEGF siRNA tumors contained numerous vimentin-positive tumor cells.
Gene regulation in VEGF siRNA tumors
We then compared gene expression profiles of VEGF siRNA tumors to tumors transfected with the control siRNA. One hundred eighty-seven genes were up-regulated and 703 genes were decreased in vivo (cut-off 1.5-fold change, supplementary Table I). When the cut-off was set to 5-fold change, 18 genes were up-regulated and 42 showed decreased expression (Table I). We successfully validated expression tendency of 12 down-regulated and 9 up-regulated genes of interest by qPCR using human-specific primers (supplementary Table II). The VEGF knock down status of the tumors was confirmed by the microarray, all 4 probe sets specific for VEGF detected down-regulation (mean: 3.11). qPCR also confirmed down-regulation of VEGF mRNA (5.62-fold) in VEGF siRNA tumors compared to controls (supplementary Table II). Several up-regulated genes are induced by hypoxia (e.g. EGLN3, CHI3L1, BCL6, FOSL2 and HGFL), which suggests that VEGF siRNA tumors are in a defined hypoxic/serum-deprived stress environment. The reduced proliferation state of cells in VEGF siRNA tumors is further highlighted by down-regulation of MKI67, which encodes for Ki-67 (3.15-fold), and PCNA (2.63-fold).
Table I. Most Highly Regulated Genes after VEGF siRNA (>5-Fold)
Regulated genes were ranked according to siRNA control versus siRNA VEGF fluorescence ratio and cut off level was set to 5-fold. In vivo, 18 genes were upregulated in VEGF siRNA tumors compared to 42 down-regulated genes. The complete list of regulated genes (>1.5-fold) is available as online (supplementary table 1).
collagen, type III, alpha 1 (Ehlers-Danlos syndrome type IV, autosomal dominant)
progesterone receptor membrane component 1
pirin (iron-binding nuclear protein)
TRAF and TNF receptor associated protein
epithelial cell transforming sequence 2 oncogene
COMM domain containing 8
SAC3 domain containing 1
adhesion molecule with Ig-like domain 2
SNARE protein Ykt6
replication protein A3, 14kDa
Kyoto encyclopedia of genes and genomes analysis reveal biological pathways enriched within up- and down-regulated genes in vivo
Three biochemical KEGG pathways (“apoptosis,” “complement and coagulation cascade” and “insulin signaling”) were enriched amongst 180 up-regulated genes referenced in the database, whereas 15 pathways were over-represented in the 701 down-regulated genes. Pathways enriched within the down-regulated genes were mainly related to cell growth and maintenance (e.g., oxidative phosphorylation and cell cycle, see supplementary Table III). Down-regulation often occurred in functionally related groups of genes: minichromosome maintenance deficient genes (MCM2, −4, −5, −6), mitotic checkpoint control genes (BUB1, BUB3, BUB1B, MAD2 and CDC20) as well as several aminoacetyl-tRNA synthetases.
CHI3L1 and PI3 expression profiles are correlated, differ between glioblastoma and low-grade glioma and are associated with glioblastoma survival
One of the most up-regulated genes in VEGF siRNA tumors were CHI3L1 (8.48-fold) and PI3 (12.77-fold) (Table I). Regulation pattern of other up-regulated genes correlated significantly with those of CHI3L1: CHI3L2 (0.5491, CI 95% 0.2857 to 0.7354), IL1B (0.4287, CI 95% 0.1342 to 0.6535) and PI3 (0.7591, CI 95% 0.5736 to 0.8705; Spearman r; Fig. 4a). EGLN3 and APOC1 did not show any significant correlation with CHI3L1. Up-regulation of these genes might reflect adaptation of malignant cells to an environment with insufficient oxygen and/or serum supply. Expression of CHI3L1 (p < 0.0001), CHI3L2 (p < 0.0001), IL1B (p = 0.0134) and PI3 (p = 0.0002) also varied significantly between GBMs and LG glioma, which are less vascularized and aggressive and exhibit a much slower growth rate. No significant difference could for EGLN3 and APOC1 (Fig. 4b).
We next looked whether mRNA expression levels of these 4 genes could reflect enhanced aggressiveness of the tumor, measured by survival time after diagnosis. For IL1B and APOC1, no significant correlation with survival could be established (data not shown). CHI3L1 had a significant (p = 0.036) association with survival in a group of 42 typical GBM patients. Median survival in the group with low (<10) CHI3L1 levels was 516 days compared to 333.5 days in the group with higher levels (>10), which was close to the median survival of all patients (343.5 days; Figs. 4c and 4e).
For PI3, much higher expression differences between normal brain and GBMs were observed, so that the cut-off to separate low and high expression was set to 100. Patients with low PI3 values had a mean survival of 500 days, whereas the median in the high-level group was only 275 days. The survival in the 2 groups differed significantly (p = 0.0004), low expressers lived 165.5 days longer and high expresser 68.5 days shorter than the median of all patients. Thus, low expression of CHI3L1 and PI3 identifies a subgroup of patients with good prognosis, and in addition, high PI3 levels are associated with shorter survival (Figs. 4d and 4e). HR for CHI3L1 was 2.134 and 2.858 for PI3. The odds calculated from these values indicate that the probability of death for a given patient with high CHI3L1 (PI3) levels is 68% (74%) higher than for one with lower levels. After adjustment for age and sex, the Cox regression model still showed a higher risk of death for patients with elevated PI3 levels with a HR of 4.0 (p = 0.0011) and 2.2 (p = 0.0405) for elevated levels of CHI3L1 (Fig. 4e), indicating that expression of these genes are independent markers of survival for GBMs.
Elafin is expressed in GBM, around necrotic areas, but not in grade II/III diffuse glioma
Immunohistochemistry for elafin was performed on 10 GBM and 10 WHO grade II/III diffuse glioma. Eight out of ten GBM contained nests of elafin-positive tumor cell, mainly localized nearby necrosis areas as shown in representative cases (Figs. 5a–5d) at different magnifications. No elafin immunoreactivity could be detected in grade II or III glioma (Fig. 5e).
PI3 induction by metabolic stress
In vitro, exposure of malignant glioma to serum deprivation or moderate hypoxia leads to strong upregulation of PI3/elafin mRNA. After 24 hr of serum deprivation or hypoxia, about 10-fold induction occurs. After 48 hr of serum deprivation, levels of PI3 were still up more than 10-times, but 48 hr of hypoxia caused more than 35-fold induction (Fig. 5f).
Determination of tumor gene expression profiles by microarrays in response to treatment with new therapeutic molecules can facilitate the discovery of markers for therapy success or failure.25 Gliomas are amongst the highest vascularized tumors and depend on a functional VEGF system, suggesting anti-VEGF as a prime therapy target for this disease.2 Here we describe formation of small, avascular tumors after transfection of glioma cells with siRNA against all VEGF isoforms. In spite of the nearly complete absence of blood vessels and capillaries, these tumors formed in 85% of the cases after four days, suggesting that VEGF is not essential for the initial implantation of the tumors. It might be that pro-invasive molecules such as MMP926 (up-regulated 3.11-fold in VEGF siRNA tumors) allow tumor cells to access the CAM tissue and that pro-survival programs enable tumor cells resist to increasing hypoxia and serum starvation. Attractive mediators for such a response are CHI3L1 (YKL-40 protein), PI3 (protease inhibitor 3/elafin) and IL1B (interleukin-1 β), amongst others.
In U87 cells, hypoxia and other types of environmental stress increase expression of YKL-40.27 YKL-40 is a secreted glycoprotein, which plays important roles in tissue remodeling, migration and angiogenesis. CHI3L1 is strongly up-regulated in GBM compared to low-grade astrocytoma28 and to normal brain.29, 30 High YKL-40 levels in serum have been reported to be closely associated with shorter survival in various types of solid tumors, including glioblastoma,31 suggesting that YKL-40 could be useful as a prognosticator for cancer survival.32 Interestingly, high YKL-40 levels are also predictive of second-line chemoresistance in ovarian cancer.33
In VEGF siRNA glioma, CHI3L1 and the related CHI3L2 were both amongst the highest up-regulated genes, suggesting a critical role of chitinase-like proteins in regulating response of tumor cells to hypoxia. It may be that YKL-40 is induced by hypoxia in vivo and exerts pro-survival and/or pro-angiogenic effects, which promote malignant progression.
Our results suggest that CHI3L1 mRNA levels from tumor samples have prognostic value; especially low CHI3L1 levels are associated with better survival, which suggests a growth support role for glioblastoma progression. YKL-40 protein also accumulates in serum from patients with other diseases such as rheumatoid arthritis34 or type 2 diabetes.35 Determining levels of CHI3L1 mRNA from cancer tissue together with its concentration in serum might enhance sensitivity of this promising biomarker.
PI3, the gene encoding for elafin, a whey-acidic-protein (WAP) motif protein, is induced >12-fold in VEGF siRNA tumors. Elafin is an antileukoproteinase that potently inhibits proteinases released from polymorphonuclear leukocytes. This inhibition preserves tissue integrity in the epidermis by preventing cell detachment.36 Strong neutrophil infiltration is also a typical of GBM, especially around necrotic areas37 and immune cells of undefined nature are also frequently found in experimental glioma (data not shown).
Transcripts of PI3 are extremely abundant in patient GBM samples and significantly related to survival. Like for CHI3L1, low levels of PI3 are associated with better survival but in addition, higher PI3 levels also correspond to poorer outcome. PI3 is located on chromosome 20q12 and gain of 20q is significantly associated with shorter survival in GBM38 and enhanced lymph node metastasis in gastric cancers.39 Gain of 20q12-q13 is also frequently observed in ovarian cancer.40 PI3 and other genes of the WAP protein family are clustered on chromosome 20q12-13.1 and are frequently up-regulated in several solid tumors.41
Interestingly, we could reproduce up-regulation of PI3/elafin in vitro, by both, serum deprivation and moderate hypoxia, suggesting a direct link between unfavorable growth conditions and PI3 expression.
A possible function of elafin in GBM might thus be to attenuate adverse effects of immune cells on malignant cells, especially around necrotic areas, where tumor cells are exposed to hypoxia. Interestingly, no elastase can be detected in central parts of glioblastoma (where most necrotic areas are), in contrast to peripheral, invading edges.42 Absence of elastase might be due to increased elafin activity in the tumor center.
There is also evidence that a functional connection exists between PI3 and IL-1β. IL-1β induces PI3 gene expression in pulmonary epithelial cells43 and elafin secretion in keratinocytes.44 IL1B and PI3 are also significantly coregulated in our cohort of GBM patients (Pearson r = 0.5; p = 0.0021, data not shown).
In VEGF siRNA tumors, IL1B was up-regulated more than 6-times together with its high affinity receptor IL1R1 (2.16-fold), suggesting an autocrine feedback loop. Interleukin-1 has been shown to enhance tumor angiogenesis and blocking of IL1 receptor leads to regression of xenografted IL-1-producing tumors.45 IL1B might also coordinate expression of some other genes downregulated in VEGF siRNA tumors, like ADAMTS1,46 which encodes for the disintegrin and metalloproteinase with thrombospondin motifs 1 protein. ADAMTS1 inhibits angiogenesis by blocking VEGF165 interaction with KDR.47
A possible role of IL-1β in hypoxic tumors might be initiation of a pro-angiogenic program by down-regulation of angiogenesis inhibitors and induction of VEGF.48 However, up-regulation of VEGF by IL-1β would have been masked by the knock down of VEGF by RNAi in our model.
Kyoto encyclopedia of genes and genomes (KEGG) analysis of down-regulated genes in VEGF siRNA tumors showed that many key players of cell cycle, oxidative phosphorylation and protein translation are coregulated. Decreased expression of these crucial genes might lead to prolonged survival of tumor cells by shutting down energy consuming processes, which can protect cells against hypoxia.49
The molecular mechanisms of CHI3L1 and PI3 induction in anti-angiogenic, presumably hypoxic experimental tumors and patient glioma remain to be elucidated. U87 cells constitutively produce HIF1A (hypoxia-inducible factor-1 alpha) protein under normoxic and hypoxic conditions,50 and control and VEGF siRNA experimental glioma on the CAM contain very high levels of HIF1A transcripts (data not shown). Interestingly, HIF target genes like carbonic anhydrase IX51 and VEGF52 also frequently colocalized in perinecrotic areas, comparable to elafin.
Genes regulated in the VEGF siRNA tumors represent a snapshot of the response of malignant tumor cells to a prototype of anti-angiogenic therapy. Some of the genes we identified might coordinate a temporary “vascular independency program” whose activation might render tumor cells less vulnerable to metabolic stress, classically associated with anti-angiogenesis. This hypothesis is confirmed by the fact that levels of CHI3L1 and PI3 are associated with poor survival in GBM patients and the presence of elafin-positive tumor cell around necrotic tumor areas. Our results demonstrate that experimental anti-angiogenesis in the CAM glioma model mimics a tumor microenvironment that favors over-expression of genes related to prognosis of glioma patients. Further studies should be conducted to explore regulation of PI3/elafin in other types of cancer.
The authors thank Ms. Aurélie Meyre and Ms. Pascale Heneaux for excellent technical assistance with histology and Ms. Véronique Pantesco for processing of microarrays and Dr. Luc Letenneur (ISPED, INSERM U593) for the Cox regression analysis. This work was supported by grants from the European Union (STROMA Consortium 2004–2007, LSHC-CT-2003-5032, to AB and VC), La Ligue Contre le Cancer, Comitée de la Dordogne (to SJ) and l'Agence Nationale de la Recherche, ANR (“Glioma Model”, JC05_44897, to MH). Dr. Akeila Bellahcène is a research associate of the National Fund for Scientific Research (FNRS, Belgium).