Inhibition of Glioblastoma Growth in a Highly Invasive Nude Mouse Model Can Be Achieved by Targeting Epidermal Growth Factor Receptor but not Vascular Endothelial Growth Factor Receptor-2 Free

Glioblastomas infiltrate the brain diffusely and can therefore not be resected completely. Typically, glioblastomas have already invaded also the contralateral hemisphere when they become clinically symptomatic. Therefore, any efficient therapeutic approach must not only target the bulk of the tumor mass but also reach cells that have invaded far beyond radiologically and intraoperatively visible borders.

A major deficit of traditional glioblastoma animal models used for preclinical testing of novel therapeutics is that tumors in these models do not invade. Orthotopic models are usually based either on human glioblastoma cell lines xenografted into the brains of immunodeficient mice or rats, or on syngeneic or heterogeneic models using rodent cell lines. Typically, tumors in these models form solid nodules at the injection site, which compress rather than invade the surrounding brain. In addition to these widely used models, some more sophisticated animal models exist, in which human glioma xenografts can show various degrees of true invasion. These models include (a) direct implantation of patient surgical specimens into the brains of nude mice (1), (b) transplantation of patient surgical material s.c. in nude mice followed by dissociation and orthotopic reinjection of these xenotransplants (2), (c) engraftment of glioblastoma-derived spheroids after short-term culture into rat brain (3), and (d) engraftment of glioblastoma stem cell–enriched cultures into mouse brain (4, 5). Thus far, no studies were done that specifically addressed the question whether diffusely invasive tumor growth in such models can be inhibited by antibodies or other agents that have proven effective in conventional noninvasive xenograft models.

Another major disadvantage of the standard models using xenografted human glioblastoma cell lines is that genetic alterations present in the original tumor are often not maintained. Especially amplification of the epidermal growth factor receptor (EGFR) gene that is present in approximately 40% to 50% of human glioblastomas (6–10) is typically not preserved in glioblastoma cell lines and xenografts derived thereof. In addition, cell lines can acquire additional genetic alterations in culture. The problem of EGFR amplification loss has previously been overcome by direct implantation of tumor specimens into the flanks of nude mice (11). Moreover, it was shown that subsequent orthotopic reinjection of s.c. grown tumor cells into the brains of nude mice can combine the advantage of maintaining EGFR amplification with the advantage of producing a brain-invasive phenotype (2). However, it cannot be excluded in this model that during serial passaging under selection pressure in the s.c. space tumor cells may acquire new genetic or phenotypic alterations before their intracranial establishment.

The inhibition of tumor invasion and angiogenesis is a major goal in the development of novel therapeutic strategies against malignant gliomas. It was shown that antiangiogenic antibody therapy can inhibit tumor growth in conventional orthotopic human glioblastoma xenograft models (12), and it may also be beneficial clinically at least for alleviating brain edema (13, 14). However, it is unknown whether antiangiogenic treatment is also effective against diffusely invading malignant gliomas, which grow less angiogenesis dependent than the solid parts of the tumor (12). Previous studies further showed that EGFR signaling can not only stimulate glioma cell proliferation but can also promote glioma invasion (15–19), suggesting that inhibition of EGFR signaling might be able to inhibit invasive glioblastoma growth. In the present study, we adapted a highly invasive glioblastoma model, which was originally established for nude rats (3), to nude mice and analyzed the effect of monoclonal antibodies (mAb) against vascular endothelial growth factor receptor-2 (VEGFR-2) or EGFR on tumor growth, invasion, and angiogenesis.

Cell culture. Tumors were obtained from patients who underwent surgical resection of a glioblastoma in the Department of Neurosurgery, University Medical Center Hamburg-Eppendorf (Hamburg, Germany). All patients consented to the use of their tissue for research. Biopsy spheroids were prepared as described by Bjerkvig et al. (20). Briefly, freshly obtained tumor tissue was minced into fragments with a diameter <0.5 mm. Fragments were seeded into T75 flasks precoated with 10 mL of 0.75% agar (Sigma-Aldrich) in DMEM. Culture medium was DMEM with 10% FCS, 4× nonessential amino acids, 2% l-glutamine, 100 IU/mL penicillin, and 10 μg/mL streptomycin (all from Invitrogen).

Antibodies. All antibodies used for in vivo experiments were provided by ImClone Systems. DC101 is a rat mAb against mouse VEGFR-2 (21). Cetuximab is a mouse/human chimeric antibody against human EGFR, in which the constant region of a mouse mAb was chimerized with human IgG1 (22). The rabbit polyclonal antibody against von Willebrand factor (vWF) as well as mouse mAbs against Ki-67 and p53 were obtained from Dako. Rabbit polyclonal antibody against cleaved caspase-3 and rabbit mAb against phosphorylated Akt (p-Akt; Ser⁴⁷³) were from Cell Signaling Technology. Mouse mAb against PTEN was from Cascade BioScience.

Orthotopic mouse glioblastoma model. After 3 to 11 days of in vitro culture, spheroids were prepared for implantation. Six- to 8-week-old nude mice (NMRI-nu/nu) were anesthetized, and a burr hole was drilled into the skull as described previously (23). Ten spheroids sized 200 to 300 μm were picked using a 5 μL Hamilton syringe with a diameter of 350 μm and were stereotactically injected into the right caudate/putamen. Three weeks after tumor cell implantation, two animals from each series were sacrificed, and brains were embedded in paraffin and cut up in serial sections, which were stained with H&E to analyze whether tumors had developed. If tumors had developed, treatment was initiated; if no tumors had established, another two animals were sacrificed 2 to 3 weeks later. This screening procedure was repeated until tumor cells became histologically detectable.

Treatment consisted of either i.p. injections of 800 μg DC101 every 3 days or i.c. infusion of cetuximab (C225, Erbitux). Infusion was done using osmotic minipumps (Alzet mini-osmotic pump, model 2004, Durect Corp.), which maintain a constant flow of 0.25 μL/h over 28 days. Pump reservoirs were filled with the antibody at a concentration of 9.2 mg/mL so that mice received 55 μg of the antibody per day. The reservoir was connected to an intracranial catheter (Alzet Brain Infusion Kit II, Durect). Posterior to the site of tumor cell implantation, a s.c. tunnel was created, and the pump was pushed forward until it came to lay on the back of the mice. The catheter tip was inserted through the same burr hole that had been created to inject the tumor cells and was placed into the center of the tumor. Control groups for DC101 received i.p. injections of PBS every 3 days, and control groups for cetuximab received minipumps loaded with PBS. After 4 weeks of treatment, animals were killed using CO₂. Institutional guidelines for animal welfare and experimental conduct were followed for all in vivo experiments.

Determination of tumor burden. Mouse brains were removed from the cranial cavity, fixed in formalin, bisected coronally, and embedded in paraffin. Serial sections, 7 μm thick, were stained with H&E. To quantify tumor cells in representative areas that typically displayed diffuse infiltration by tumor cells in control animals, we selected six different areas at six different coronal levels. These 36 regions were scanned using a 20× objective and Leica IM50 software (Leica). Images were analyzed as TIF files using the ImageJ program (version 1.36b). Pictures were transformed into 8-bit grayscale images, and the threshold was adjusted so that nuclei became white on a black tissue background. The percentage of white pixels in a representative rectangular field measuring 0.1 mm² was determined. To estimate the total tumor burden, three adult mice were analyzed by the same method so that for all 36 regions the area occupied by nuclei in normal brain could be subtracted from that in treated and control mice, resulting in the excess area taken up by tumor cell nuclei. All morphometric analyses were done in a blinded fashion.

Immunohistochemical analyses. Paraffin sections were dewaxed using standard histologic procedures. Sections were incubated with primary antibodies against vWF (1:200), Ki-67 (1:50), cleaved caspase-3 (1:200), p53 (1:200), p-Akt (1:50), or PTEN (1:300) overnight at 4°C. Detection of bound antibodies was done using the EnVision System (Dako) or Vectastain kit (Vector Laboratories) following the manufacturers' instructions. Vessel density was determined by counting the number of blood vessels stained with the anti-vWF antibody in six high-power fields (one high-power field = 0.031 mm²) in the most densely vascularized tumor “hotspot” area on sections most adjacent to the site of tumor cell injection. To analyze the proliferative activity of the tumor cells, the percentage of Ki-67 immunoreactive nuclei was determined in three high-power fields in the most actively proliferating tumor area. The fraction of apoptotic tumor cells was determined in apoptotic hotspot areas by counting tumor cells that expressed cleaved caspase-3 in three high-power fields. Immunoreactivity for p-Akt was scored as described by Haas-Kogan et al. (24), and immunoreactivity for PTEN was scored as described by Mellinghoff et al. (25) and Choe et al. (26).

Reverse transcription-PCR. Total RNA was extracted from frozen tumor specimens using Trizol reagent (Invitrogen) or from paraffin-embedded tumor xenografts using the RNeasy FFPE kit (Qiagen). Following DNase digestion, cDNA was synthesized from 1 to 3 μg of total RNA using random primers (MWG Biotech) and SuperScript II reverse transcriptase (Invitrogen). PCR amplifications were done using 1 μL of cDNA in a total volume of 20 μL with 10 pmol of each primer, 0.2 mmol/L deoxynucleotide triphosphates, 0.5 unit Taq polymerase in 1× buffer [10 mmol/L Tris-HCl (pH 8.4), 50 mmol/L KCl, 1.5 mmol/L MgCl₂, 0.001% (w/v) gelatin; all from Invitrogen]. Primers specific for EGFR and EGFRvIII transcripts were used, generating amplification products of 1,044 and 243 bp, respectively, as described previously (25). PCRs were initially denatured at 95°C for 2 min followed by 40 cycles of 30 s at 95°C, 30 s at 56.5°C, and 80 s at 80°C in a thermal cycler. PCR products (10 μL) were analyzed by electrophoresis on 2% agarose gels stained with ethidium bromide.

Real-time PCR analysis. Genomic DNA was extracted from frozen tumor specimens using the DNeasy Blood & Tissue kit (Qiagen). Real-time PCR analysis was done using a 7500 Fast Real-Time PCR System (Applied Biosystems) with standard conditions for the 9600 emulation run mode, including a dissociation stage. Amplification reactions contained 10 ng of genomic DNA and 10 pmol of forward and reverse primers for EGFR or IFN-γ (MWG Biotech), generating fragments of 110 and 85 nucleotides, respectively, as described previously (27). Reactions further contained 10 μL of 2× QuantiTect SYBR Green PCR Master Mix (Qiagen) in a total volume of 20 μL. Relative amounts of amplified DNA were determined using the ΔΔC_T method, calculating relative quantity values with normal human leucocyte DNA as calibrator.

Fluorescence in situ hybridization. The probe for EGFR detection was derived from Homo sapiens PAC clone RP5-1091E12 containing a 180-kb insert with almost the entire EGFR gene sequence (Genbank accession no. AC006977). The probe was labeled with Spectrum Orange-dUTP (Abbott Molecular, Inc.) using the BioPrime DNA Labeling System (Invitrogen). Briefly, 1 μg DNA was dissolved in 19 μL H₂O, boiled for 5 min, and cooled on ice. Subsequently, 5 μL of 10× deoxynucleotide triphosphates, 20 μL of 2.5× random hexamers, 1 μL Klenow fragment, and 5 μL of 0.5 mmol/L Spectrum Orange-dUTP were added, and the mixture was incubated at 37°C for 2 h. DNA was purified using a Bio-Spin 30 Tris column (Bio-Rad). Sections (4 μm thick) of paraffin-embedded mouse brains were deparaffinized and pretreated with 0.25 mg/mL proteinase K for 45 min at 45°C. Probe (1 μL), 50 μL Cot-1-DNA (1 mg/mL; Invitrogen), 1 μL hairpin-loop DNA (Sigma-Aldrich), and 1 μL blue dextran (100 mg/mL) were mixed with 7 μL sodium acetate and precipitated with 2.5 volume ethanol. The pellet was dissolved in 10 μL of 50% formamide, 2× SSC, and hybridized to paraffin sections overnight at 37°C. Posthybridization washes were done at 45°C in 50% formamide, 2× SSC, and 2× SSC/0.1% Tween 20, counterstained with 4′,6-diamidino-2-phenylindole (Sigma-Aldrich), and mounted using Citifluor antifading solution (Agar Scientific).

Methods of data analysis. Statistical comparisons of tumor burden in the 36 different areas analyzed (unpooled and pooled) as well as comparisons of vessel density, proliferation, and apoptosis between different treatment groups were done using the unpaired t test for normally distributed samples or the Mann-Whitney rank-sum test for not normally distributed samples and the SigmaStat program (version 2.0).

Patient characteristics, xenograft model, and assessment of tumor burden. Glioblastoma xenografts were established from short-term spheroid cultures of nine different tumors. Patients were aged between 25 and 68 years, and six were male and three were female (Table 1). Six patients underwent primary surgery for a newly diagnosed glioblastoma, and three cases were recurrences.

Spheroids were implanted into the caudate/putamen of nude mice. Treatment was initiated when tumors became histologically detectable, on average 6 weeks after spheroid implantation. Treatment and control groups comprised 5 to 11 animals, depending on the initial amount of tumor material available, the capacity of primary cultures to form spheroids, and on the time xenografts took to establish because repeated sacrifice for histologic screening reduced the number of animals available for treatment. We could not use magnetic resonance imaging to screen for xenograft establishment because the subtle tumor manifestation eludes the resolution of the magnetic resonance imaging (data not shown). In four experiments, mice received i.p. injections of DC101 (cases 1-4; Table 1), and in 7 experiments, animals received continuous i.c. infusion of cetuximab (cases 3-9). In two experiments (cases 3 and 4), enough tumor material was available to treat mice with both antibodies in parallel. Control groups received either vehicle injections i.p. or vehicle infusion i.c. Treatment was continued over 4 weeks.

Tumor xenografts grew diffusely invasive (Fig. 1A and B), and tumor cells were often highly pleomorphic, resembling the morphology of the original tumors (Fig. 1C). The highest tumor cell density was typically observed around the injection site in the center of the caudate/putamen. From there, tumors extended into the entire striatum and invaded the thalamus, globus pallidus, corpus callosum, anterior commissure, as well as tracts and nuclei at the base of the brain (Fig. 1D).

Due to the diffuse and extensive tumor invasion, conventional histologic techniques for estimating tumor volume, which can be applied to standard xenograft models with a nodular growth pattern, could not be used in our system. Neither could we use nonhistologic methods, such as quantification of total human DNA, which are applied in models of metastasis to estimate tumor load in affected organs, because we wished not only to quantify total tumor burden but also to evaluate histologic growth patterns and assess treatment effects on tumor cell invasion. We therefore devised a novel method for quantifying tumor burden in regions containing a mix between normal cells and tumor cells, rather than solid tumor only. Initially, we attempted to distinguish tumor cells from murine tissue by immunohistochemistry using human-specific antibodies against vimentin, mitochondria, or human nuclei. However, staining was often patchy, and background reactivity of murine cells could never completely be avoided despite extensive optimization of blocking and staining protocols (data not shown). Thus, it was impossible to reliably distinguish tumor cells from murine cells by immunohistochemistry, let alone by conventional histologic techniques. We therefore decided to assess tumor burden by subtracting cells in defined areas of normal reference brain from cells in identical regions of brains that had received tumor xenografts. Selecting areas that typically displayed tumor invasion, such as the corpus callosum, thalamus, or globus pallidus, we defined 36 different landmark points on six different coronal levels with six points of analysis for each level (Fig. 2A). The total area taken up by cell nuclei was quantified in these regions using image analysis (Fig. 2B), and the area taken up by nuclei in normal adult murine brain was subtracted, resulting in the excess area taken up by tumor cell nuclei.

Effects of cetuximab and DC101 on tumor growth. A representative analysis of tumor burden for all 36 landmark points is shown in Table 2. To compare total tumor burden between different treatment groups, results for the 36 landmark points were pooled. This cumulative analysis showed that tumor growth was inhibited in three of seven xenograft cases treated with cetuximab compared with controls that had received vehicle i.c. (Fig. 3, and representative example in Table 2 and Fig. 2B). Mean inhibition of tumor burden was 78.6 ± 13.7% in case 3, 72.4 ± 17.9% in case 4, and 80.3 ± 10.2% in case 5. In four cases, cetuximab had no significant effect. Treatment with DC101 did not inhibit tumor growth in any of the four cases that were treated, including two (case 3 and 4) that were effectively growth inhibited by cetuximab (Figs. 2B and 3A and B).

Mechanisms of tumor growth inhibition. Signaling through EGFR can mediate antiapoptotic, mitogenic, proinvasive, and antiangiogenic effects (28). We therefore analyzed whether cetuximab treatment affected tumor cell apoptosis, proliferation, invasion, and angiogenesis in vivo using quantitative immunohistochemical techniques. In tumors that responded to cetuximab, the proportion of apoptotic cells was significantly increased (2.4- to 4.4-fold; P < 0.05) compared with controls (Fig. 4A and B). Although the overall rate of apoptosis was low, this difference is particularly striking because we had to determine the proportion of apoptotic cells relative to total cell counts in 3 hps instead of being able to score apoptotic tumor cells as percentage of a pure tumor cell population. This was due to the fact that we could not reliably distinguish human from murine cells as described above. Because tumor cell density and extension were lower in cetuximab-responsive tumors than in controls, fewer tumor cells were scored so that the true rate of apoptosis would even be higher in cetuximab-responsive tumors.

Analyzing tumor cell proliferation, we detected a significantly reduced proportion of Ki-67 immunoreactive cells (56.7-99.6%; P < 0.05; Fig. 4A and B) in tumors that were successfully treated with cetuximab. However, in inverse analogy to the evaluation of apoptosis, this result also has to take into account the decreased tumor cell density in cetuximab-responsive tumors so that the true difference in proliferation between cetuximab-responsive tumors and controls tends to be overestimated by our analysis.

To assess tumor cell invasion, we compared the cell density in central areas close to the injection site with cell densities in distant areas. Stereotactic tumor cell injection was done in the central striatum at level 2, point 4 (Fig. 2A). The most distant areas were thus located most caudally and basally at level 6, points 5 and 6. When comparing tumor cell densities in central and distant areas, we found a significantly greater inhibition by cetuximab in distant areas than centrally. In case 3, the mean inhibition at the injection site was 60.5%, whereas the mean inhibition at level 6, points 5 and 6 was 84.0% (P < 0.05). In case 4, central inhibition was 48%, whereas distant inhibition was 85.9% (P < 0.05). In case 5, central inhibition was 71.3%, whereas distant inhibition was 91.6% (P < 0.01). Taken together, these findings suggest that cetuximab may have inhibited tumor cell invasion; however, it is impossible to differentiate to what degree distant tissue infiltration is due to actual tumor cell migration/invasion or rather to continuous diffuse growth of the proliferating tumor.

Some studies reported stimulatory effects of EGFR activation on tumor angiogenesis, which seem to be mediated mostly indirectly through up-regulation of other angiogenic effectors such as VEGF (29). In addition, two studies reported that cetuximab had antiangiogenic effects (30, 31). Therefore, and also to determine whether DC101 had antiangiogenic effects in the invasive glioblastoma model despite lacking growth-inhibitory effects, we quantified the intratumoral microvessel density (MVD). Cetuximab had no significant effect on the MVD, neither in responsive nor in nonresponsive cases (Fig. 4A and B), neither did DC101 reduce the intratumoral MVD in treated versus control cases (Fig. 4A and B). Comparisons with normal age-matched murine brain even showed that the MVD in tumor xenografts was slightly reduced (MVD in normal striatum at level 2: 35.8 ± 2.9, n = 5 mice). In addition, we detected no dividing endothelial cells by immunohistochemistry for Ki-67 or combined immunohistochemistry for vWF and Ki-67 (data not shown), indicating that the invasively growing tumors do not depend on neovascularization.

Response to cetuximab is associated with EGFR amplification and expression of EGFRvIII. Because only three of seven cases responded to the cetuximab treatment with significant inhibition of tumor growth, we hypothesized that molecular differences affecting EGFR signaling might account for this divergence. We first analyzed whether EGFR gene amplification was present in the original tumors. Using real-time PCR, EGFR amplification was detectable in five of the nine tumors from which xenografts were derived, including all three cases that responded to cetuximab but no nonresponsive cases (Fig. 5A; Table 3). Amplification levels ranged from at least 5-fold to >35-fold compared with normal human leucocyte DNA. Second, we determined whether EGFR amplification was also present in xenograft tumors using fluorescence in situ hybridization analysis. Amplifications were maintained in all xenografts derived from tumors in which EGFR amplifications were identified by real-time PCR (Fig. 5B and C) but were absent in xenografts derived from tumors without amplification. Third, we analyzed whether EGFRvIII, a common and constitutively active genomic deletion variant of EGFR (6–10), was expressed in the original tumors. Using reverse transcription-PCR, we detected the EGFRvIII variant in four of nine tumors (Fig. 5D; Table 3), all of which also displayed EGFR amplification by real-time PCR analysis (Fig. 5A). Only one tumor with amplified EGFR (case 2) lacked the EGFRvIII variant, and this tumor showed the lowest amplification level. Fourth, we studied whether EGFRvIII expression was maintained in xenografted tumors. Using reverse transcription-PCR, we found that xenografts derived from tumors expressing EGFRvIII retained expression of the truncated receptor variant (data not shown).

Other molecular determinants of the response to cetuximab. It has been suggested by clinical correlative studies that, in addition to the EGFR status, downstream effectors of the phosphatidylinositol 3′-kinase pathway determine the response to EGFR kinase inhibition (24, 25). We therefore analyzed whether expression of p-Akt and PTEN was associated with the response of xenografted tumors to cetuximab. Using immunohistochemistry, we detected strong expression of p-Akt in all three tumors that gave rise to cetuximab-responsive xenografts (Fig. 5E; Table 3). p-Akt was further detected in two of the four nonresponding cases. Immunoreactivity for PTEN was found in two of the three responding cases, but one case lacked PTEN expression (Fig. 5F; Table 3). Of the four nonresponding cases, three expressed detectable levels of PTEN, whereas one was negative. Taken together, there was no clear association between p-Akt or PTEN status with response to cetuximab treatment.

The goal of this study was to analyze whether treatment with antibodies directed against EGFR or VEGFR-2 can inhibit the growth of invasively growing glioblastomas. To establish a genuinely invasive xenograft mouse model, we adapted a technique that was originally developed for nude rats (3) to nude mice. As our study shows, another major advantage of this model, besides the highly invasive growth of the tumors, is that xenografts maintain amplification of EGFR and expression of the EGFRvIII variant in vivo. These alterations are typically lost in conventional glioma cell lines and xenografts derived thereof (11). Disadvantages of the invasive model are, however, that tumors grow slowly, requiring long observation periods, that sequential sacrificing of test animals is necessary to find out whether tumors have at all established before initiating treatment, and that the quantification of total tumor burden requires a more elaborate technique than estimating the size of a solitary tumor nodule.

The main finding of our study is that constant interstitial application, analogous to convection-enhanced delivery in patients (32), of a mAb against EGFR can strongly inhibit progression of a subset of invasively growing glioblastomas; in contrast, treatment with an anti-VEGFR-2 antibody is ineffective in this model. This lack of an effect of antiangiogenic treatment was not entirely unexpected, as tumors in the invasive model grow largely independent of neovascularization. In line with observations in the rat model (33), we detected no dividing endothelial cells, and tumors even displayed a reduced MVD compared with normal murine brain due to the expansion of neoplastic cells between the preexistent host vasculature. This growth pattern is also characteristic for the invasive components of gliomas in human brain, where intratumoral vessel densities are similar or even lower than those in normal brain, and only when the tumor cell density exceed a critical level angiogenesis occurs (34–36). We previously showed that treatment with DC101 can even increase perivascular tumor cell invasion in a conventional xenograft model, most likely as an escape mechanism to compensate the insufficient vascular supply of the main tumor mass (23). Taken together, our current findings and those of previous studies suggest that antiangiogenic treatment is effective against solid components of glioblastomas but is largely ineffective against the invasive, angiogenesis-independent tumor component.

Interestingly, we found that treatment with cetuximab was only effective against tumors showing EGFR amplification and expression of EGFRvIII but was ineffective against tumors expressing normal levels of wild-type EGFR. Amplification of EGFR is present in approximately 40% to 50% of all human glioblastomas, and EGFRvIII is expressed in about half of these cases but rarely in tumors lacking EGFR amplification (6–10). EGFRvIII represents the most common EGFR mutation in human gliomas and is created by an in-frame deletion of 801 bp spanning exons 2 to 7. EGFRvIII is unable to bind any known EGFR ligand but is constitutively activated and can enhance the growth of glioma xenografts in vivo (37). Cetuximab recognizes both EGFR and EGFRvIII with similar affinity and inhibits ligand binding to EGFR as well as autophosphorylation of EGFRvIII, causing internalization of the cetuximab-EGFRvIII complex (38). In addition, cetuximab is likely to also block heterodimerization between EGFRvIII and EGFR as it sterically prevents the receptor from adopting an extended conformation required for dimerization (39). Notably, inhibition of heterodimerization was found to be the most likely mechanism of action for the anti-EGFR antibody mAb528, which was isolated at the same time as the murine version of cetuximab and displays similar properties (40). mAb528 inhibited the in vivo growth of transfected U87 cells overexpressing EGFRvIII and expressing wild-type EGFR endogenously but was ineffective against xenografted fibroblasts overexpressing EGFRvIII without wild-type EGFR (41, 42). The findings of our present study, in which spheroids were not genetically engineered to overexpress EGFRvIII but were directly derived from tumors with a different EGFR/EGFRvIII status, suggest that inhibition of EGFRvIII autophosphorylation and/or inhibition of EGFRvIII/wild-type EGFR heterodimerization but not inhibition of ligand binding to wild-type EGFR alone can explain the growth-inhibitory effect of cetuximab.

In previous studies using EGFR tyrosine kinase inhibitors, controversial findings were reported as to whether EGFRvIII expression sensitizes glioblastomas to treatment with the EGFR tyrosine kinase inhibitors gefitinib and erlotinib or whether it confers resistance to such treatment. Initially, an in vitro study showed that U87 cells became resistant to treatment with gefitinib on transfection with EGFRvIII (43). A subsequent clinical study showed that expression of EGFRvIII or EGFR amplification in malignant gliomas was irrelevant for predicting response to gefitinib (44). However, more recently, coexpression of EGFRvIII and PTEN was found to sensitize glioblastoma patients to EGFR tyrosine kinase inhibitor treatment and also to render U87 cells susceptible to erlotinib in vitro (25). This was supported by a study using a panel of intracranial glioblastoma xenografts derived from serially passaged s.c. xenografts, showing that the expression of wild-type PTEN combined with amplification of aberrant EGFR were key characteristics of erlotinib-sensitive tumors, although this molecular constellation did not obligatorily confer sensitivity to erlotinib in all cases analyzed (45). It was further described that cetuximab inhibited the growth of EGFR-amplified glioma cell lines in vitro but not of unamplified ones and that it was effective against two EGFR-amplified cell lines in vivo (46). However, in contrast to our present study, the EGFR-amplified cell lines did not express EGFRvIII and also no direct in vivo comparison with non–EGFR-amplified xenografts was done; furthermore, the study contained no histologic analysis so that the tumor growth pattern remains unknown (46). Our findings significantly extend those previous observations, showing that EGFR amplification and EGFRvIII expression render invasively growing i.c. glioblastoma xenografts susceptible to cetuximab as opposed to conferring treatment resistance.

Previous studies suggested that not only EGFR and PTEN but also the activation status of Akt influences the response of gliomas to EGFR-targeting agents. Haas-Kogan et al. (24) identified EGFR amplification and low levels of p-Akt as predictors of a clinical response to erlotinib. Our observation that all cetuximab-responsive tumors showed activation of Akt is somewhat in contrast to these findings. However, none of the erlotinib-responsive tumors in the clinical study expressed EGFRvIII, whereas all of the cetuximab-responsive tumors in our study did, in line with reports showing that EGFRvIII constitutively activates Akt even in the presence of wild-type PTEN (26, 41). Our findings thus suggest that activation of the Akt pathway through EGFRvIII may be associated with a different response to EGFR-targeting agents than activation of the Akt pathway through wild-type EGFR or other mechanisms. Of course, it can also not be excluded that the different mechanisms by which erlotinib and cetuximab target EGFR may also affect the extent to which tumors respond to treatment.

Activation of Akt is positively regulated by EGFR and negatively regulated by PTEN. Our finding that all tumors lacking PTEN immunoreactivity showed activation of Akt is in line with studies showing that loss of PTEN is highly correlated with Akt activation in vivo (26, 47). However, the lack of PTEN in one cetuximab-responsive tumor was unexpected, given that loss of PTEN is believed to promote cellular resistance to EGFR-targeting treatment (25, 45). Our findings thus indicate that expression of wild-type PTEN combined with amplification of aberrant EGFR may represent a positive predictor but not an obligatory constellation for response to EGFR-targeting treatment. The most important predictive factor seems to be the presence of EGFR amplification and/or EGFRvIII expression.

We identified several potential mechanisms for the effect of cetuximab against invasively growing glioblastomas, including (a) increased tumor cell apoptosis, (b) inhibition of tumor cell proliferation, and (c) possibly decreased tumor cell invasion. The proapoptotic effect is most striking, especially as we calculated the fraction of apoptotic tumor cells based on total cell counts, so that when accounting for the greatly reduced tumor cell density, the true rate of apoptotic tumor cells in cetuximab-responsive tumors is even higher. Proapoptotic effects of EGFR-targeting therapies have been reported previously (28), and at least in vitro cetuximab was found to only have a proapoptotic effect on glioblastoma cell lines harboring EGFR gene amplifications but not on nonamplified cell lines (46). Our findings confirm and extend this observation to the in vivo situation. We further detected a reduced proliferation rate in tumors treated with cetuximab. However, also here the decreased tumor cell density in cetuximab-responsive tumors has to be considered so that the net decrease in tumor cell proliferation is actually less than estimated by our method and is less striking than the proapoptotic effect. To exactly determine the rate of tumor cell apoptosis and proliferation, it will be essential in the future to develop a reliable method to distinguish between tumor and murine cells.

Activation of EGFR and EGFRvIII is further known to stimulate glioma cell migration and invasion (15–18). In human gliomas in situ, EGFR amplification is preferentially found at the invasive edge (19), and different EGFR-targeting agents were found to inhibit glioblastoma cell invasion at much lower concentrations than required for growth suppression (16). In a conventional xenograft model with a non–EGFR-amplified cell line, we even found that cetuximab inhibited the compensatory increase in perivascular cuffing and invasion along the host vasculature induced by DC101, although it had no effect on growth of the main tumor mass (48). Notably, in that previous study, invasion presented only as perivascular cuffing, whereas in the present study we observed genuinely diffuse single-cell invasion. In the present study, we detected a significantly greater reduction of the tumor cell density at distant sites compared with the tumor center, suggesting that cetuximab may have inhibited tumor cell invasion. However, with currently available techniques, it is impossible to assess to what extent distant tissue infiltration is due to actual tumor cell migration/invasion or is a function of the total tumor burden.

To conclude, we showed that the invasive growth of glioblastomas in vivo can be inhibited by interstitial delivery of an anti-EGFR antibody, but the response of individual tumors depends on the presence of amplified and/or mutated EGFR. In contrast, antiangiogenic treatment is ineffective against diffusely invading glioblastomas.

Z. Zhu and L. Witte are employees of ImClone Systems.

We thank Dorothea Zirkel, Regina Fillbrandt, and Hildegard Meissner for expert technical assistance and Sker Freist and Monika Thiel for help with the illustrations.