Glioblastoma (GBM) is the most aggressive brain tumor and among the deadliest cancers.1 GBM cells are characterized by their invasive abilities and a striking angiogenic potential, which distinguishes them from low grade glioma. Anti-angiogenic agents already used in the clinic such as humanized anti-VEGF antibodies (Bevacizumab, Avastin™) are currently tested in patients with brain tumors together with conventional anti-tumor agents (http://www.cancer.gov/clinicaltrials). Recently, it has been shown that the combination of bevacizumab and the topoisomerase-1 inhibitor irinotecan improves progression-free survival for 60% of GBM patients.2 Anti-VEGF alone already reduces vascular density in patients with colorectal cancer,3 and anti-tumor efficacy is likely to be enhanced in combination with other anti-angiogenic and/or anti-tumor agents.4 This needs to be evaluated in preclinical setups to define combinations that will bring enhanced clinical benefit.
We have previously shown that experimental glioma grown from U87 cells grafted on the chick chorioallantoic membrane (CAM) recapitulate key features of patient glioblastoma on both, morphological and molecular levels.5 Major advantages of the model are enhanced visual control of morphological characteristics of the tumor such as size and vascularization, as well as reproducibility of tumor growth under normal and treatment conditions. Because grafted tumor cells are of human origin, the use of human-specific GeneChips® favors identification of genes that are regulated in tumor cells rather than in the chick host tissue. This is especially appropriate to comprehend selective molecular regulation that occurs in tumor cells in response to anti-angiogenic treatment.6
In the initial characterization of this model, we identified a subset of genes that were upregulated during the angiogenic switch, including IL6, and found their level of expression to be predictive of gene regulation in patients with GBM5 compared to low-grade glioma. These genes are therefore likely to be instrumental during tumor progression. Indeed, IL6 appears as a prospective target for glioma therapy. It is produced by GBM cells in vitro and in vivo7 and elevated IL6 expression distinguishes GBMs from low grade glioma8 (and our own study). Moreover, transgenic mice that express the src oncogene in astrocytes fail to develop glioma on an IL6−/− genetic background.9
Anti-IL6/IL6R therapy using monoclonal antibodies (BE-8 and CNTO 328, Tocilizumab) has been included in clinical trial for numerous diseases and cancers, such as multiple myeloma, renal cell carcinoma and B-lymphoproliferative disorders (for review see10) and rheumatoid arthritis.11 First results indicate that targeting the IL6 pathway has beneficial effects in the treatment of IL6-dependent cancers.12
In this study, we evaluated the effect of anti-IL6 therapy by RNA interference in experimental glioma models, either alone or in combination with VEGF inhibition. Furthermore, we provide insights into the molecular pathways associated with inhibition. Our results reveal a significant advantage of a combinatory anti-IL6/VEGF strategy, encouraging future clinical trials using already available pharmaceutical inhibitors against these prime targets for anti-glioma therapy.
CAM, chorio-allantoic membrane; GBM, glioblastoma; IL6, interleukin-6; kd, knockdown; shRNA, small hairpin ribonucleotide acid; siRNA, short interfering ribonucleotide acid; SNA, Sambucus nigra lectin; VEGFA, vascular endothelial growth factor A.
Material and methods
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, 2 mmol/L glutamine, 100 units/mL penicillin and 0.1 mg/mL streptomycin. Fertilized chicken eggs (E.A.R.L. Morizeau, Dangers, France) were handled as described.13
Production of lentiviral vectors
The Pic20 plasmid containing H1 RNA pol III promoter and pTRIPDU3EF1a-EGFPMCSDU3 vector were obtained from Dr. François Moreau-Gaudry (University of Bordeaux 2, France). The target sequences for each of the genes were as follows: IL6 (NM_000600) 5′-GGAGAAGATTCCAAAGATG-3′, VEGF (NM_003376), 5′-GGAGTACCCTGATGAGATC-3′. Negative control (NC) oligonucleotide was purchased from Eurogentec. To construct the 19 nucleotide hairpin shRNA cassettes, 2 complementary DNA oligonucleotides were chemically synthesized (Sigma-Proligo) 5′-AGCTTCC-19-TTCAAGAGA-19-TTTTT GGAAG-3′ and 5′-TCGACTTCCAAAAA-19-TCTCTTGAA-19-GGA-3′, annealed, and inserted downstream of H1 promoter.
Lentiviral particles were produced by transient transfection of 293T cells as described previously.14 To determine the titer of each viral supernatant, serial dilutions were used for transduction of 293T cells. Following transduction, EGFP was measured by cytofluorimetric analysis (FACS Calibur, BD Biosciences). Typical titers were around 1 × 108 infectious particles/mL.
Transduction of U87 cells
U87 cells (3 × 104 cells/well) were incubated for 24 hr in complete medium. Before infection, medium was removed, and cells were incubated with viral supernatants for 24 hr at 37°C in the presence of 8 μg/mL protamine sulfate. After 5 days, EGFP-positive cells were enriched by cell-sorting. Cells were further termed IL6-sh, VEGF-sh and NC-sh and the transient cells VEGF-si and NC-si.
Experimental glioma models
U87 cells were washed with PBS, trypsinized, collected and resuspended in medium. Three and a half millions cells in 15 μL were deposited on the CAM at embryonic day 10.5 Digital photos were taken under a stereomicroscope (Nikon SMZ800).
Six-week-old nude mice (Charles Rivers Italia, Calco, Italy) were divided in 4 groups (n = 10): NC-sh, NC-sh/Avastin, IL6-sh and IL6-sh/Avastin. They were intracranially implanted as previously described15 with U87 cells (5 × 104 in 10 μL PBS) transduced with NC-sh or IL6-sh. Treatment was initiated 12 days after implantation, when control tumors are well established and have switched to the angiogenic phenotype as previously described.16 Where indicated, mice received subcutaneous injections of 800 μg Avastin™ (Roche) every 3 days for 20 days. All mice were sacrificed 36 days after implantation.
U87 cells were seeded in 24-well plate (10 × 104 cells per well), and counted daily (Coultronics, Margency, France). All proliferation assays were conducted in triplicates, and each set of experiments was performed at least 2 times.
Modified Boyden chamber invasion assay
Transwell inserts (BD Biosciences, France) precoated with Matrigel (10 μg/100 μL) were placed in a 24-well plate. Subconfluent cells were resuspended in serum-free DMEM at a final density of 2 × 105 cells/mL. To induce migration through the Matrigel layer, the lower chambers were filled with 750 μL of DMEM supplemented with 0.1% FCS. After 36 hr of incubation, cells on the outside surface of the transwell filter were fixed with methanol 30%/acetic acid 10% and stained with Coomassie blue. Inserts were photographed, and cells were counted using NIS-Elements AR 2.30 software (Nikon France).
Tumor tissues were homogenized in lysis buffer, CelLytic-MT (SIGMA) in presence of protease inhibitor cocktail (Sigma-Aldrich, Lyon, France), and centrifuged at 14,000 rpm at 4°C for 10 min. The protein concentration of the supernatant (or culture media) was assayed by the Bradford method and stored at −80°C until used.
Human VEGF-A and IL6 protein levels were measured using the human specific DuoSet ELISA kits (R&D Systems, Lille, France) according to the manufacturer's instructions. A microplate reader (Molecular Devices SpectraMax) was used to determine colorimetric densities at a wavelength of 450 nm with a reference wavelength of 540 nm.
Histology and Immunochemistry
Tumor tissue was fixed with 4% paraformaldehyde and 10-μm sections were cut and stained with hematoxylin and eosin. Same specimen sections were incubated with the following primary antibodies: mouse anti-Human Vimentin Ab-2 (1/400, clone V9, NeoMarkers Ab, Interchim, Montluçon, France), biotinylated SNA (1/500 Vector Laboratories, Berlingame, CA) and anti-Human Ki-67 (clone MIB-1, 1/100, Dakocytomation, Trappes, France), rat anti-mouse CD31 (1/50, B&D, Pharmigen, San Jose, CA). Corresponding fluorescent secondary antibodies and streptavidin were from Molecular Probes (Invitrogen). Photos were taken at indicated magnifications using a Nikon fluorescence microscope.
Follow-up of tumor growth and invasion assessment
CAM experimental glioma from each knockdown condition (n ≥ 24) were photographed under a biomicroscope and tumor volumes were calculated as described.5 Blood vessel density was determined by counting individual SNA-positive vessels and subsequent normalization to tumor surface of the cross-sections labeled with fluorescent SNA lectin, anti-vimentin and DAPI. Proliferation index was evaluated by calculating the percentage of KI-67 positive cells amongst tumors cells identified by DAPI staining. For quantitative assessment of invasion in the CAM, 3 fields (n > 4 tumors per condition) of sections stained with anti-vimentin antibody and DAPI encompassing the basement of the tumor and the supporting CAM tissue were photographed and tumor cells were scored. Statistical analysis was performed with Prism 4 (GraphPad Software). For experimental glioma grown in mice, volume vessel density and invasion were quantified from 4 mice per group with similar approaches as for CAM experimental glioma, except that anti-CD31 antibody was used to stain the vasculature, and tumor cell clusters physically separated from the primary tumor were manually counted on vimentin stained sections to determine the invasion index.
Complementary RNA preparation and microarray hybridization
RNA was isolated from typical experimental glioma on the CAM at day 4 and pooled (n = 4 tumors per condition), then extracted using the RNeasy mini kit Plus (Qiagen). RNA quality was assessed on an Agilent 2100 Bioanalyzer (Agilent, Massy Cedex, France) and a Nanodrop ND-1000 spectrophotometer (Nyxor Biotech, Paris, France). Complementary RNA (cRNA) was prepared according to the manufacturer's protocol “sample protocol I,” and hybridized to Human whole Genome U133 plus 2.0 oligonucleotide arrays representing over 47,400 transcripts (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
Tumor samples were analyzed using a pair wise comparison with GCOS 1.2 software (Affymetrix). All data were assembled into a FileMaker Pro 8 Database.
For each of the 6 paired comparisons (IL6-kd vs. NC; VEGF-kd vs. NC; IL6/VEGF-kd vs. NC; IL6-kd vs. VEGF-kd; IL6/VEGF-kd vs. VEGF-kd; IL6/VEGF-kd vs IL6-kd), overview of regulation was obtained after filtering probesets using different degrees of stringency. To statistically determine the degree of correlation of gene coregulation induced by different treatments, differentially hybridized probe sets were selected (fold-change p-value < 0.01), log transformed and compared using the Spearman test. Given the overall high number of regulated genes, a more stringent filter was used to display the percentage of shared and specific genes per condition, that is, > 2-fold change in at least one comparison. Finally, genes were then selected as significantly over- or under-expressed when difference in expression ratio was at least 5-fold with a fluorescence signal above 100 for “present calls.” This set of 777 genes was used for cluster analysis (cluster 2.11” and “treeview 1.6; Life Sciences Division, Lawrence National Laboratory, Department of Molecular and Cellular Biology, University of California, Berkeley, CA). Functional annotation and gene ontology assignment was performed by querying the DAVID Bioinformatics Resources (NIH, http://david.abcc.ncifcrf.gov).
Effect of IL6 knockdown in vitro
U87 cells were chosen for their high angiogenic potential, a common feature of human GBM despite their genetic heterogeneity. It is the best documented cell line in terms of cytogenetics17, 18 and IL6 expression. It expresses high amounts of IL6 like do highly aggressive glioblastomas.19, 20 A correlation between IL6 gene expression and shortened survival in glioblastoma patients has been recently demonstrated.21
Knockdown (kd) of IL6 in U87 cells was tested using 3 different siRNA sequences (data not shown), and the most efficient was stabilized in a lentiviral vector (IL6-sh). VEGF-kd was achieved with a sequence previously described6 either transiently (for combined kd on CAM experimental glioma) or after transduction (for single kd). For combined knockdown of both IL6 and VEGF (IL6/VEGF-kd), the IL6-sh clone was transiently transfected with VEGF-siRNA (IL6-sh + VEGF-si). As shown in Figure 1, IL6 synthesis was fully inhibited in cells transduced with IL6-sh. A slight accumulation of the cytokine could be detected after 96 hr of culture corresponding to 2% of the control concentration. U87 cells transfected with VEGF siRNA also showed a significant decrease in IL6 production (80%), whether transduced with NC-sh or not. VEGF protein synthesis was also verified in these cultures. All clones transfected with VEGF-siRNA were consistently inhibited (>90%). Knockdown of IL6 also resulted in strong reduction of VEGF production (60%). Combined knockdown led to almost complete inhibition of VEGF and IL6 protein synthesis.
These results demonstrate reciprocal cross-downregulation by IL6 and VEGF siRNAs, a finding that was anticipated for IL6 knockdown22, 23 and also recently described after VEGF knockdown in retinal pigment epithelial cells.24 VEGF released by U87 cells thus induces IL6 in tumor cells in an autocrine manner, in addition to paracrine triggering of IL6 release in neighboring inflammatory cells.25 Therefore, both cytokines are integrated in a positive bidirectional autocrine regulatory loop likely to involve VEGF receptors VEGFR-1/Flt-1, VEGFR-2/KDR, Neuropilin-1 (NRP-1) and Neuropilin-2 (NRP-2) that are expressed by U87 cells.26
The effect of IL6 knockdown on cell proliferation and invasion was next assessed in vitro (Fig. 2). Neither IL6 nor VEGF knockdown affected proliferation rate. However, all lentiviral transduced clones including controls grew faster than nontransduced cells. They also exhibited a more aggressive phenotype on the CAM, transduced negative controls giving rise to bigger tumors after 7 days than nontransduced cells and showed increased angiogenesis (data not shown).
Invasive behavior of transduced cells was examined in vitro in Boyden chambers coated with matrigel. Unexpectedly, migration was significantly enhanced in cells transduced with IL6-sh. There was no significant difference of migration in this assay between VEGF-sh cells and NC-sh controls.
Effect of IL6 and IL6/VEGF knockdown on growth of CAM experimental glioma
U87 cells under negative control (NC), IL6, VEGF or IL6/VEGF knockdown (kd) conditions were grafted onto the vascularized CAM as previously described.5 As illustrated in Figure 3a, IL6-sh transduced U87 cells form a tumor nodule after 2 days but were unable to grow further. This behavior is comparable to tumors derived from VEGF-kd cells. Blood circulation at the periphery of the tumors was reduced indicating impairment of the angiogenic switch required for vessel-dependent growth 2 days after grafting. Cells under combined knockdown still aggregated into a prominent nodule. However, after 2 days, there was a visible regression in tumor volume and no blood circulation could be seen at the surface of the remaining nodules. Efficacy of single and combined knockdowns was assessed by ELISA quantification of IL6 and VEGF within the tumor tissue (Fig. 3b). The effect of interference was persistent on the protein level and reciprocal cross down-regulation was also evidenced, comparable to the results obtained in vitro.
Tumors were further characterized by quantitative assessment of volume, proliferation index vascular density and invasion, 7 days after grafting (Fig. 4). Size of the CAM experimental glioma was effectively reduced (>80%) by knockdown of IL6 or VEGF (Figs. 4a and 3a). The nodules in combined knockdown tumors were reduced in size to merely 5% of the controls and also significantly smaller than in single knockdown gliomas.
The percentage of proliferating tumor cells was determined by KI-67 detection on tumor sections (Fig. 4b). In addition to reducing the number of live cells, combined knockdown leads to more than 3-fold reduction of cell proliferation compared to controls. It is of interest to note, that KI-67 labeling index in controls corresponds to values found in highly aggressive patient GBMs.27, 28
Vascular density was also measured in these samples. As anticipated from biomicroscopy, there was a significant reduction of blood vessels in both single knockdowns compared to control tumors (Fig. 4c). Double knockdown tumors, were fully avascular as illustrated by representative immunofluorescence analysis and only residual SNA-positive cells were found.
The behavior of tumor cells after grafting on the CAM was assessed by analyzing immunostained sections of tumors and supporting tissue (Fig. 4d). Glioma cells grown on the CAM induce sprouting and intususceptive angiogenesis and also coopt nearby vessels thereby increasing blood supply. However, tumor cells are rarely detected at distance from the primary implantation site in the supporting CAM. Under IL6 or VEGF knockdown, tumor cells invade massively the CAM tissue, with 50% more cells being detected when VEGF alone is silenced. There was also a clear difference in the nature of invasion in these 2 groups. As illustrated in Figure 4d, 75% of cells under VEGF knockdown showed islets of aggregated cells inside the host tissue, mostly connected to the primary tumor and forming an irregular invasive front, whereas under IL6-kd, the majority of cells (70%) showed single-cell migration. In combined treated tumors, the invasive phenotype was significantly reduced to a similar degree as seen in controls.
These results indicate that combined inhibition of IL6 and VEGF is significantly more efficient than single inhibition in reducing size, proliferation and neovascularization of CAM experimental glioma. In addition, the invasive ability of tumor cells that was found extensively enhanced by VEGF or IL6 knockdown invivo (albeit with different behavior), is restricted when both VEGF and IL6 pathways are inhibited.
Evaluation of IL6 knockdown combined with Avastin™ treatment in mice xenografted with U87 cells
The effect of inhibiting IL6 either alone or in combination with VEGF was next assessed in mice xenografted with IL6-sh U87 cells and treated or not with Avastin™. Brains were analyzed by histology 36 days after implantation.
As shown in Figure 5a, control mice untreated with Avastin™ developed a compact tumor mass occupying half of the brain with a smooth and regular margin. Treatment of control mice with Avastin reduced tumor volume to less than 25%, but also induced the emergence of several cell tumor clusters invading the injected and controlateral hemisphere. IL6-sh U87 cells formed a much reduced primary tumor (volume < 2% compared to NC) but a significant number of infiltrated tumor cell clusters were detected in the brain of the recipient mice. Finally, mice inoculated with IL6-sh U87 cells and treated with Avastin™ did not form a detectable tumor in the brain and only rare and small tumor cell clusters were visible. Effects of IL6-kd and/or Avastin™ were comparatively evaluated on tumor angiogenesis by counting vessels stained with anti-CD31 in the primary tumor area or residual tumor cell clusters. Avastin™ reduced significantly the number of vessels in tumors derived from NC cells as well as from IL6-sh U87 cells. The percentage of vessels in residual clusters of the IL6-kd/Avastin group was 10% compared to controls and <25% compared to IL6-sh tumors.
These data confirm that combined inhibition of IL6 and VEGF is more efficient in reducing tumor growth, angiogenesis and invasiveness, comparable to results obtained in the CAM model.
Gene profiling of CAM experimental glioma under IL6-, VEGF- and combined knockdown
Our analysis of xenografted glioma suggests that IL6 inhibition reduces primary tumor growth and angiogenesis but induces autonomous invasive properties of glioma cells confirming the in vitro data. VEGF inhibition promotes invasion of residual cells in vivo, but the combined blocking of IL6 and VEGF synergistically inhibits both, tumor growth and invasiveness. To gain insight into these effects at the molecular level, we performed a transcriptomic analysis on experimental glioma grown on the CAM. This model allows more precise isolation of tumor tissue than murine models and cross hybridization between chicken and human mRNA is much reduced. Overall analysis recapitulated in Figure 6 revealed several interesting features. A significant correlation of gene regulation induced by each single treatment was evidenced (n = 4,815 probe set pairs, Spearman r = 0.77, p < 0.0001, Fig. 6a). The global response to treatment was even more similar when IL6-kd was compared to the double knockdown (n = 7,268 pairs, r = 0.92, p < 0.0001, Fig. 6b). VEGF-kd compared to the double knockdown caused also a significantly correlated response, however lower than for IL6 (n = 6,257 pairs, r = 0.85, p < 0.0001, Fig. 6c). These data confirm phenotypic observation made in vitro and in vivo and illustrate the overlap between IL6 and VEGF signaling pathways on a molecular level. They also suggest that the enhanced tumor growth inhibition in the double knockdown may be attributed either to genes specifically regulated by this treatment or to a general concerted down regulation of a group of genes with vital functions (e.g., cell division, replication).
There were twice as many genes downregulated in IL6-kd tumors compared to VEGF-kd tumors (Fig. 6d) consistent with the hypothesis that IL6 has a cell autonomous protumoral effect in addition to promoting angiogenesis via VEGF. As expected from the correlation analysis, the majority of genes regulated in IL6 or VEGF single knockdown tumors were regulated in a similar way in the double knockdown tumors (70 and 60%, respectively, Fig. 6e). The percentage of genes similarly over or under expressed in IL6-kd and VEGF-kd tumors was around 30% (maximum coregulation value: 47% of genes down-regulated in VEGF knockdown tumors also down-regulated in IL6 knockdown tumors) On the other hand, as many as 40% of regulated genes in the double knockdown tumors (35% up, 46% down) were specific for this condition, possibly accounting for the important synergistic effect of the combined inhibition. This suggests that impairing biological functions of one of the 2 cytokines affects distinct pathways in addition to common ones. Also note that combined treatment increases general gene expression changes (up and down) by more than 50%.
Transcriptomic data were then searched for genes known to be associated with invasive behavior of glioma cells.29 A proinvasive gene signature with clinical significance consisting of 19 genes misregulated in fast migrating glioma cell lines and glioma patients was queried in our database. Interestingly, IL6 knockdown tumors exhibit up-regulation of genes belonging to the pro-migratory signature (C6orf145, CTGF, CYR61, FZD2, IL6ST, LRRK2, PPP3CB, PRMT2, RBMS3, SMURF2, TNPO1, TPM4). In addition, TGFB1, TGBR1, TGFBR2 and WNT5B that belong to pathways predicted to be instrumental in migrating glioma cells were also over expressed. Most of these genes were not activated in VEGF knockdown tumors suggesting that their invasive behavior is mediated by other factors. Moreover, in double knockdown tumors (combo), IL6ST, WNT5B, TGFB1 and TGFBR2 were not over expressed unlike in IL6-kd tumors, indirectly suggesting that they may participate in controlling autonomous invasive potential of glioma cells in vivo. We next searched for genes with documented functions related to migration amongst the top regulated genes (listed in Supporting Information Fig. 1). We found DCN strongly induced in IL6/VEGF-kd tumors (4 probe sets; mean fold-change +35), RARRES1 induced under all conditions with the highest fold changes for IL6/VEGF-kd versus control tumors (3 probe sets; mean fold-change +187), POSTN mostly induced under IL6 knockdown (2 probe sets; mean fold-change +640), EPHA3 and EPHA4, SLIT2 all with the highest induction in combo tumors (fold-change>+10), MMP1 and MMP9 both specifically induced in the combo tumors (fold-change, +7.51 and +19.74, respectively).
We then evaluated the level of repression of protumoral genes in combined knockdown tumors. Cluster analysis of 777 probesets filtered as described in Materials and Methods and listed in Supporting Information Figure 1, revealed 2 subsets of genes consistently downregulated in this group. Subset 1 (Supporting Information Fig. 2) consisted of 133 probe sets for which genes were under consistent inhibition in single treatments and even stronger silencing under IL6/VEGF silencing. Functional classification using DAVID bioinformatic tools uncovered that genes related to positive cell cycle control were highly significantly enriched in this subset (37/133, Fisher exact test p-value: 9.6 × 10−21). From these, 27 genes were directly functionally assigned to mitosis control, p-value: 3.6 × 10−23 (Supporting Information Fig. 3, panel A). Whether down regulation of cell cycle genes by IL6 and/or VEGF inhibition occurs autonomously in tumor cells or indirectly under the influence of a hypoxic, growth limiting microenvironment remains to be determined. However, the second hypothesis seems more likely given that NC-sh, IL6-sh and VEGF-sh clones proliferate equally well in vitro. In addition, glioma cells poorly respond to exogenous VEGF stimulation,30 and are not affected in their proliferation when VEGF or VEGFR2 is inhibited.26 The second subset of 118 probesets (Supporting Information Fig. 4) corresponds to genes with increased expression in VEGF-kd tumors and substantial repression under combined treatment. Genes enriched in this cluster were primarily related to chromatin assembly (Supporting Information Fig. 3, panel B), (10/118, p-value: 2.2 × 10−8). This same cluster also contains many known tumor progression genes such as CDK6, COL4A2, IL1R1, IL1R2, HES6, TAGLN, PDPN and CXCL12.
Here, we evaluated the effects of IL6 inhibition on experimental glioma growth, alone or in combination with anti-VEGF, and investigated molecular pathways associated with either treatment. We show that targeted inhibition of IL6 impairs the growth of CAM experimental glioma comparable to anti-VEGF. This observation correlates well with the fact that the 2 pathways show cross-regulation in vitro and in the CAM experimental glioma as revealed by protein synthesis measurement in each single knockdown condition. IL6 has been previously shown to upregulate VEGF via STAT3 activation in glioma cells22, 23 and VEGF is able to induce IL6 synthesis in monocytes and retinal epithelial cells.24, 31
But although IL6-kd and VEGF-kd efficiently impair primary tumor growth and significantly influences gene regulation in a similar way, they also exhibit differences on a cellular and molecular level. For example, the quality of the invasive behavior of cancer cells migrating into the CAM was significantly different, with IL6-kd inducing a strong increase in individual cell migration, whereas VEGF-kd causes massive collective invasion of tumor cell clumps, an observation that correlates with PDPN (podoplanin) expression, which is induced by VEGF-kd and repressed by IL6-kd. Podoplanin is a small mucin-like protein recently detected at the surface of tumor cells at the invasive front of carcinomas, mediating their collective migration.32, 33 It may therefore be that VEGF-kd alone promotes collective invasion of tumor cells by inducing PDPN. In the absence of PDPN, individual cell invasion may occur as observed in IL6-kd tumors, presumably stimulated by other regulators that are not active in the double knockdown condition.
The results from the transcriptomic analysis reveal a complex expression profile of migration controlling genes. Indeed, amongst the genes with highest regulation, we found potent inhibitors, but also stimulators of cell migration. However, there are several candidates which might be instrumental in decreasing cell migration in the double knockdown tumors. For instance, DCN encoding the small leucine-rich proteoglycan decorin is specifically upregulated. Decorin has been shown to bind to and inactivate TGF-β,34 to block migration of osteosarcoma cells35 and to reduce glioma growth through modulation of the immune system.36, 37 Another candidate, RARRES1 is highly expressed in spontaneously regressing melanoma and its expression is lost in prostate cancer. RARRES1 overexpression potently decreases invasion in this pathology.38, 39 The axonal guidance molecule SLIT2 is highly up-regulated in the double knockdown. SLIT2 inhibits cell migration via activation of the ROBO1 receptor,40 which is highly expressed in patient glioma as well as in the experimental glioma, including in the double knockdown (data not shown). The SLIT2-ROBO1 pathway could therefore activate autocrine repulsive signals and explain the absence of migrating cells in this condition. EPHA3 is a strong inhibitor of SDF-1 dependent migration in cancer cells41 and highly overexpressed in the double knockdown, but also in IL6 knockdown tumors. Other factors might have bi-phasic effects as has been shown for POSTN (periostin), which can inhibit or induce migration in a dose-dependent manner.42 Finally, genes, which are normally associated with a more invasive phenotype of tumor growth such as MMP1 and MMP9 are also induced in the double knockdown, but obviously this is not sufficient to promote cell migration in our model. Most likely, tumor cells need additional signals to engage in a migratory pathway together with a minimal fitness condition.
Altogether, these findings suggest the existence of a “migratory balance” in tumor cells exposed to anti-angiogenic therapy, for which many genes positively or negatively influencing cell migration are misregulated. Dosage of certain critical genes or family ofgenes might determine the migration phenotype. In addition, cell motility is most likely influenced by parameters such as the proliferative state of tumor cells and metabolites available for migration.
There is evidence that tumor hypoxia resulting from anti-VEGF therapy increases malignant potential of surviving cancer cells in patients. Tumors under anti-angiogenesis treatment may persist, and switch on alternative pro-angiogenic pathways43 or show enhanced vessel co-option and/or invasion properties.44 Persistance of glioma treated with anti-VEGF has been evidenced during our initial analysis of the CAM experimental glioma treated with VEGF siRNA6 and was also described in rat with human G55 glioma grafted in the brain and treated with anti-VEGF antibody,45 as well as in mice injected subcutaneously with U87 cells transduced with VEGF-siRNA.46 In glioblastoma patients as in experimental models, tumor cells that escape inhibition of VEGF-mediated angiogenesis produce high amounts of SDF1α that is thought to be instrumental in mediating vessel cooption and infiltration.47, 48
We identified 2 clusters of genes that are sturdily downregulated by IL6/VEGF-kd, one being enriched in genes encoding mitotic factors, the other in genes of chromatin components. In the first cluster, in addition to CDC-CDK components, we found important mitotic checkpoint intermediates, such as Bub1, Aurora-A and Aurora-B. The silencing of the mitotic kinase Aurora-A is of particular significance as we and others have reported a correlation between its accumulation at the protein level and glioma malignancy49, 50 suggesting that not only proliferation but also tumor cell migration and consequently patient survival are dependent on Aurora-A levels. The expression of Aurora-B, a chromosomal passenger protein, correlates well with aggressive behavior in glioblastoma multiforme,51 presumably driving chromosomal instability in tumors with loss of heterozygosity of p53. In addition, survivin, a member of the apoptosis-inhibiting gene family encoded by BIRC5 was also found in the same cluster and is over expressed together with Aurora-B in patients. Survivin promotes tumor growth by counteracting the p53 tumor suppressor, which is induced by Aurora-B in tumor cells retaining one or both alleles of p53.52 The down regulation of BIRC5 in the experimental glioma under IL6 knockdown is likely due to the repression of its transcription factor STAT3. There is evidence that interfering with STAT3 in astrocytoma cell lines results in apoptosis due to downregulation of survivin.53 This suggests that tumor cells challenged with anti-IL6/VEGF might be more sensitive to apoptosis in a long-term treatment. Interestingly, Aurora-A, Aurora-B and survivin are also linked by the fact that they are indirect targets of histone deacetylase (HDAC) inhibitors.54
In the second cluster of genes repressed by the combination of inhibitors but not by VEGF-kd alone, we found a significant enrichment of histone encoding genes. This silencing of chromatin assembly components might contribute to the strong tumor inhibition and suggests that living tumor cells may generally be impaired in their replication and transcription abilities, which correlates well with their decreased proliferative behavior evidenced by KI67 staining. We recently showed that inhibition of (HDAC4) reduces the growth of CAM experimental glioma by inducing p21WAF1/Cip1.55 However, this pathway is unlikely to be involved in tumor repression driven by IL6/VEGF-kd because neither HDAC4 nor CDKN1A encoding p21WAF1/Cip1 are regulated in this condition and the transcription level of CDKN1A remains very low (data not shown). Beside histone silencing, the analysis of this same cluster highlighted the specific repression of protumoral genes including PDPN (discussed above), TAGLN, and CXCL12/SDF1α. TAGLN encodes transgelin, recently described as promoting cell migration in a TGFβ-dependent fashion after induction of lung fibrosis.56 CXCL12 encoding SDF1α also belongs to this cluster of genes that are induced by VEGF-kd but repressed under combined inhibition. As mentioned above, SDF1α is likely to mediate infiltration of tumor cells in patients treated with anti-VEGF and has been recently proposed as a generic biomarker of glioblastoma relapse for patients treated with anti-angiogenic agents.48
Tumors under combined inhibition of VEGF or IL6 were essentially unable to induce the angiogenic transition and to grow further. Inhibition of both pathways was significantly more efficient at reducing all tumor parameters including propensity of residual cells to invade the host tissue either collectively or individually. These observations were validated in a mouse xenograft model. Importantly, humanized neutralizing antibodies are available for VEGF (Bevacizumab, Avastin™) and for the IL6 receptor (Tocilizumab, Actemra™). Bevacizumab is being assayed in numerous clinical trials for new and recurrent GBM in combination with various drugs (for review57). As for Tocilizumab, it should be soon authorized for patients with inflammatory disorders such as rheumatoid arthritis. Results of clinical trials are expected shortly for patients with epithelial cancers (breast and lung) which progression have recently been reported as IL6-dependent.58, 59 We anticipate that treating human GBM with Tocilizumab may have an even more important inhibitory effect on tumor development than the one we describe in this study. Here, only tumor cell-derived IL6 was inhibited, whereas Tocilizumab will block IL6 signaling in both, tumor and stromal cells. It should therefore bring an additional level of protection against immune suppression possibly driven by key downstream effectors such as STAT3.60 Using immuno-competent systems such as the syngeneic GL261/C57BL6 glioma model61 may also reveal additional inflammatory/immune molecular responses to the combined inhibition of IL6 and VEGF, which were presumably negligible in our models (chick embryos have no mature immune system but macrophages are found in the tumors).
In conclusion, our results indicate that combination of anti-IL6 and anti-VEGF is significantly more efficient than monotherapy at inhibiting the growth of aggressive experimental glioma and also control invasive behavior of resistant cells. In addition, this study identified distinct genes and pathways that have been shown to promote invasion and tumor cell proliferation. These new “anti-angiogenesis-response genes” merit further investigation as markers of treatment efficacy in glioma patients treated with angiogenesis inhibitors and might constitute new therapeutical targets in case of treatment failure. Furthermore, our results highlight that despite of similar morphological appearance of tumors treated with angiogenesis inhibitors, molecular responses can be assigned to each therapy.
This work was supported by the European Union (STROMA Consortium 2004-2007 to AB), La Ligue Contre le Cancer, comités de la Dordogne et des Landes (to SJ) and l'Agence Nationale de la Recherche, ANR (Glioma Model to MH). The authors thank Ms. Aurélie Meyre and Ms. Audrey Barrault for technical assistance with histology, Ms. Véronique Pantesco for processing of microarrays and Ms. Véronique Guyonnet for production of recombinant lentiviruses. This work was carried out in the context of a joint European laboratory co-administered by INSERM and the University of Milan (ELAT).