Pilocytic astrocytoma (PA) is the most common glioma of childhood. Despite their relatively high incidence, the molecular mechanisms responsible for tumorigenesis and growth of PA are poorly understood. Previous in vitro studies in our laboratory showed that despite the absence of ErbB1, PA was sensitive to ErbB1 tyrosine kinase inhibitor gefitinib. To identify alternative targets of gefitinib in PA, we studied other members of the ErbB receptor tyrosine kinase family that have been identified in brain tumors. Using gene expression microarray and Western blot analyses, we found that ErbB3 is highly overexpressed in PA compared with other pediatric brain tumors (glioblastoma, ependymoma, medulloblastoma, atypical teratoid/rhabdoid tumor, and choroid plexus papilloma). Developmental biology studies have identified Sox10 as a regulator of ErbB3 expression during development of the neural crest. Investigation of Sox10 in PA revealed that it is highly overexpressed relative to other pediatric brain tumors, lending support to the theory that Sox10-regulated overexpression of ErbB3 may be driving growth in PA. Sox10-regulated ErbB3 overexpression is a novel insight into the biology of PA, suggests possible recapitulation of developmental pathways in tumorigenesis, and presents possible targets for therapeutic intervention that might be used for hypothalamic variants not amenable to surgical cure.
Pilocytic astrocytoma (PA), the most common glioma of childhood, is a World Health Organization (WHO) grade I tumor with distinctive histologic and biologic features (1, 2). Like other WHO grade I lesions, the tumor is biologically indolent and amenable to surgical cure if located in a favorable anatomic site such as the cerebellar hemisphere (3). Unfortunately, the tumor often occurs in vital areas of brain such as the hypothalamus, which not only precludes surgical excision, but also results in devastating clinical symptomatology as the tumor slowly progresses. Although both chemotherapy (carboplatin/vincristine) and radiation therapy are standard treatment regimens used by neurooncologists for PA that cannot be cured by surgery alone, neither is very effective in the long term (4-6). Hence, new, and preferably targeted, therapies are needed to improve survival and decrease the cumulative toxicities of multiagent chemotherapy and radiotherapy.
The development of targeted therapies mandates better understanding of the biologic factors responsible for initiation and growth of PA. Unlike diffuse astrocytomas, mutation in the TP53 gene does not play a role in tumorigenesis for PA (7, 8). Although neurofibromin, a gene product of NF1, has been found to be overexpressed in PA, it is also expressed in normal, reactive, and neoplastic astrocytes (9, 10). Our laboratory has recently shown that gefitinib (ZD1839, IRESSA; AstraZeneca, Boston, MA), a reportedly selective ErbB1 (epidermal growth factor receptor [EGFR]) tyrosine kinase inhibitor, blocked proliferation of in vitro PA cell cultures despite a lack of ErbB1 expression (11). Gefitinib was found to inhibit a number of tyrosine kinases other than ErbB1, albeit with less potency (12). Identification of the target(s) of gefitinib in PA would help elucidate the mitogenic pathways driving proliferation in PA. These findings prompted our interest in whether gefitinib blocks growth in PA by inhibiting other ErbB receptor tyrosine kinase family members.
The ErbB family consists of ErbB1, ErbB2, ErbB3, and ErbB4, assembled into a network of interacting transmembrane homo- and heterodimers, which respond to numerous ligands, including the neuregulins (13). They drive a number of basic biologic processes, including growth and differentiation, and have been shown to be critical during development of the nervous system (14). Aberrant ErbB activity has been implicated in the phenotype of a number of central nervous system malignancies, in particular ErbB1 amplification in adult glioblastoma (GBM) (15). ErbB2 and ErbB4 coexpression has been correlated with aggressiveness in ependymoma (EPN) (16), and ErbB2 overexpression has been observed in meningioma (17). The role of members of the ErbB family in PA tumorigenesis is uncertain. We investigated ErbB family receptors in PA using a diverse panel of pediatric tumor specimens as a comparison. The results of our analyses also led us to further investigate Sox10, a transcription factor that controls the expression of ErbB3 in embryonic neural crest development (18).
Materials and Methods
RNA Preparation From Tumor Samples
This was a retrospective study carried out at the Children's Hospital and the University of Colorado Health Sciences Center Denver (Denver, CO) in compliance with internal review board regulations (COMIRB #95-500). Neurosurgical tumor resections were snap-frozen in the operating room at the time of surgery. RNA was extracted from a total of 38 specimens obtained from children under the age of 18 years with primary brain tumors. Specimens were classified according to WHO international histologic tumor classification as the following: 5 PA (with NF1 and pilomyxoid variants excluded), 4 pediatric GBM, 16 EPN, 6 medulloblastoma (MED), 5 atypical teratoid/rhabdoid tumor (AT/RT), and 2 choroid plexus papilloma (CPP). Total RNA was extracted from snap-frozen tumor using a Qiagen RNeasy Mini Kit (Qiagen, Valencia, CA). Approximately 30 mg of frozen tumor specimen was added to 500 μL lysis buffer provided in the RNeasy Kit. Frozen tumor specimens were homogenized using a Rotor-Stator homogenizer (Kinematica, Lucerne, Switzerland) and then processed according to the standard RNeasy protocol.
Sample labeling, microarray hybridization, washing, and scanning were performed according to the manufacturer's protocols (Affymetrix Inc., Santa Clara, CA). Before processing, the quantity and quality of each tumor RNA sample was assessed by spectrophotometric comparison of the 260- and 280-nm wavelengths and by visualization of intact 28S and 18S ribosomal RNA bands on agarose gel electrophoresis. Five micrograms of total RNA was reverse-transcribed using a T7-(dT) 24 oligomer and Superscript II Reverse Transcriptase (Invitrogen Corp., Carlsbad, CA). Subsequently, biotin-labeled cRNA was generated from the double-stranded cDNA template by in vitro transcription using T7 RNA polymerase and a BioArray High Yield RNA Transcript Labeling Kit (Enzo Diagnostics, Farmingdale, NY). Biotinylated cRNA (20 μg) was fragmented to an average size of 35 to 200 bases with fragmentation buffer (40 mM Tris-acetate, pH 8.1, 100 mM KOAc, 30 mM MgOAc, for 35 minutes at 94°C). The fragmented, biotinylated cRNA (10 μg) was then hybridized to the Affymetrix Human Genome U133 plus 2.0 GeneChip array. The arrays were hybridized for 16 hours at 45°C. After hybridization, the arrays were washed and stained according to the standard Amplification for Eukaryotic Targets Protocol (Affymetrix Inc.). The GeneChip arrays were subsequently scanned at 488 and 570 nm with a confocal G2500A GeneArray Scanner (Agilent Technologies, Palo Alto, CA). After data acquisition, the scanned images were quantified according to algorithms of Microarray Suite 5.0 software. The scans from each array were globally scaled by setting the average signal intensity to a target signal of 500. The Affymetrix HG-U113 Plus 2 chip contained 54,675 probe sets with some redundancy for certain genes or splice variants. Pairwise comparison between the treated and untreated samples was performed, and the fold change derived from the signal log ratios was averaged for the replicated genes that were deemed "present" on the detection call. Routine quality control parameters, including visual array inspection, scaling factor, background noise, 3″/5″ housekeeping gene ratios, and percent "present" calls were evaluated for each microarray analysis.
Data Transformation and Analysis
Data was exported to GeneSpring 7.0 bioinformatics software (Silicon Genetics, Redwood City, CA), which normalized data from different experiments using multifilter comparisons. GeneSpring has 2 methods of normalizing data: a "per-chip" method, in which each measurement is divided by the 50th percentile of all measurements in the same array; and a "per gene" method in which samples are normalized against the median of the control samples. From this data set, we obtained normalized hybridization intensity values for selected genes for individual tumor specimens.
We performed 2 types of data analysis on our gene-expression profiles to find genes that were overexpressed in PA. In the first analysis, we compared PA with other individual types of pediatric brain tumor. Each comparison involved the normalized hybridization intensity of selected genes, specifically ERBB family and SOX10 genes, from individual tumor gene expression profiles. For these 5 comparisons, normalized hybridization intensity values for a selected gene were then averaged for each type of pediatric brain tumor (i.e. PA, GBM, EPN, MED, AT/RT, and CPP) to identify genes that were overexpressed in PA relative to other individual pediatric brain tumors.
The second analysis performed on our microarray data was a rapid screen designed to identify those genes that were uniquely and highly overexpressed in PA versus other pediatric brain tumors in a nonbiased manner. This list of genes could then be ranked according to the magnitude of overexpression in PA. To achieve this comparison, we divided our gene-expression profiles into 2 groups. The first group contained pooled PA gene-expression profiles and the second group contained pooled gene-expression profiles for all pediatric brain tumors other than PA (i.e. GBM, EPN, MED, AT/RT, and CPP). By pooling tumors other than PA, we created a baseline against which significantly overexpressed genes in PA could be identified. The 2 groups were compared using GeneSpring, and a list of differentially expressed genes with a fold change >2 in PA and "present" in 4 of 5 PA (p < 0.05) was generated. This list of genes was then ranked according to the highest modified centroid value. The modified centroid allows us to measure the significance of differentially expressed genes by applying equal weighting to fold change and p values. It is similar to the centroid value, which is obtained by dividing the fold change value by the p value for differentially expressed genes. Because the numerical range of the p value is usually much greater than the fold change value, the centroid value is biased heavily toward the p value. To overcome this bias, the modified centroid method applies logarithm-10 to the p value to reduce its numeric range, converts the value from negative to positive, and then multiplies this value by the fold change value to obtain a single value. This single value (the modified centroid) can be used to rank genes according to their degree of overexpression (fold change) and statistical significance (p value) with roughly equal weighting. Calculation of modified centroids and ranking of gene lists was performed using Excel 2003 (Microsoft Corp., Redmond, WA).
Western Blot Analysis
Tumor protein lysates were obtained from a number of snap-frozen tumor specimens (COMIRB #95-500) classified by the WHO international histologic tumor classification as the following: 8 pediatric PA, one adult anaplastic astrocytoma (AA), one adult anaplastic oligoastrocytoma (AOA), 3 pediatric GBM, 6 adult GBM, and 8 pediatric EPN. Approximately 50 mg of frozen tumor specimen was homogenized using a Rotor-Stator homogenizer (Kinematica) on ice in 500 μL 1x lysis buffer (50 mM HEPES pH 7.5, 150 mM NaCl, 10 mM EDTA, 10% glycerol, 1% Triton X-100, 1 mM sodium vanadate, 1 mM sodium molybdadate with a complete Mini Protease Inhibitor Cocktail [Roche, Indianapolis, IN]). The resulting protein lysates were then incubated with a loading dye at 95°C for 1 minute according to the method of Laemmli (19), separated on precast 8% SDS Tris-glycine gel (BioRad, Hercules, CA), and transferred to PVDF membrane (Millipore, Billerica, MA). The membranes were blocked with 5% nonfat milk in TBST (10 mM Tris-HCL, 150 mM NaCl, 0.2% Tween 20) for 5 hours, followed by an overnight incubation at 4°C with rabbit polyclonal anti-ErbB3 (sc-285; Santa Cruz Biotechnology, Santa Cruz, CA), rabbit polyclonal anti-Sox10 (CeMines, Golden, CO), and goat polyclonal anti-β-actin (Santa Cruz Biotechnology) sera at dilutions 1:200, 1:2,000, and 1:1,500, respectively. After incubation with secondary horseradish peroxidase-conjugated donkey anti-rabbit antibody (1:3,000 dilution) (Jackson Laboratories, West Grove, PA), the membrane was added to a blocking buffer for 30 to 40 minutes at room temperature followed by 3 washes in TBST. The immunoreactive signals of the protein bands were then detected using enhanced chemiluminescence (Perkin-Elmer, Boston, MA) detection solution and exposed to X-Omat Blue XB-1 imaging film (Eastman Kodak, Rochester, NY). Images of Western blot bands were acquired by white light illumination flat fielding performed with a ChemiDoc XRS gel documentation system (BioRad). Band intensities were detected and measured in optical density units using BioRad Quantity One 1-D analysis software.
Data for Affymetrix gene expression comparisons are presented as mean values ± standard error of the mean. Comparisons between protein and cDNA microarray expression levels were achieved using Pearson's correlation test. A p value of < 0.05 was considered statistically significant. Statistical analysis was performed using Graphpad Prism 4 statistic software (Graphpad Software, San Diego, CA).
ERBB Family Gene Expression in Pediatric Brain Tumors
We studied ERBB gene expression by microarray analysis in a panel of pediatric brain tumors consisting of 5 PA, 4 GBM, 16 EPN, 6 MED, 5 AT/RT, and 2 CPP. We found that the average normalized hybridization intensity of ERBB3 is significantly higher in PA than any in other pediatric brain tumor types we tested (Fig. 1). A list of genes overexpressed in PA versus a panel of other pediatric brain tumors was generated (Table 1) (for the complete list of 1,215 genes, see online supplemental table). The average ERBB3 expression in PA was 27.5-fold higher than in the other pediatric tumors combined. When genes overexpressed in PA versus other brain tumors (>2-fold in PA; "present" in 4 of 5 PA; p < 0.05) were ranked according to greatest significance using the modified centroid value, which applies equal weighting to fold change and p value, we found that ERBB3 was the seventh most significantly overexpressed gene in PA (Table 1). As previously described by other groups, we also observed that ERBB1 and ERBB2 were overexpressed in pediatric GBM (15) and EPN (16), respectively (Fig. 1).
SOX10 Gene Expression in Pediatric Brain Tumors
It has previously been demonstrated that transcription factor Sox10 controls the expression of ErbB3 in developing embryonic neural crest cells (18). Because ERBB3 was found to be overexpressed in PA, we hypothesized that SOX10 would also be overexpressed in PA. We studied SOX10 gene expression by microarray analysis in the same panel of pediatric brain tumors and found that the average normalized hybridization intensity of SOX10 is significantly higher in PA than any of the other pediatric brain tumor species (Fig. 2). The average SOX10 expression in PA was 29.4-fold higher than in the other pediatric tumors combined (Table 1). When genes overexpressed in PA versus other brain tumors (>2-fold in PA; "present" in 4 of 5 PA; p < 0.05) were ranked according to greatest significance using the modified centroid value, we found that SOX10 was the fourth most significantly overexpressed gene in PA (Table 1).
Western Blot Analyses of ErbB3 and Sox10 in Pediatric Brain Tumors
Using Western blot analyses, we investigated ErbB3 and Sox10 protein levels in tissue lysates from frozen specimens of 8 pediatric PA, 3 pediatric GBM, 6 adult GBM, one AA, one AOA, and 8 pediatric EPN. We found that both Sox10 and ErbB3 protein expression is highest in PA relative to other pediatric brain tumors investigated (Table 2). High ErbB3 protein expression (≥++) was observed in 7 of 8 pediatric PA, as opposed to 2 of 11 of the high-grade astrocytoma, and 2 of 8 EPN. Similarly, high Sox10 protein expression (≥++) was seen in 6 of 8 pediatric PA as opposed to 2 of 11 high-grade astrocytomas, and a low level of Sox10 expression was observed in the majority of tumors studied.
Relative protein levels for both Sox10 and ErbB3 correlated well with relative gene expression levels in tumors for which both Western blot and microarray analyses were performed, namely PA, pediatric GBM, and EPN. Together, the protein and gene expression data provide strong evidence that Sox10 and ErbB3 are overexpressed in PA compared with other brain tumors.
Correlation of Sox10 and ErbB3 Expression in Pilocytic Astrocytoma
Because Sox10 has been shown to control ErbB3 expression, we studied the correlation of expression levels of these genes in our pediatric brain tumors. Statistical analysis of SOX10 and ERBB3 using microarray data in a panel of 38 pediatric brain tumors (5 PA, 4 GBM, 16 EPN, 6 MED, 5 AT/RT, and 2 CPP) showed a strong correlation of SOX10 with ERBB3 gene expression (Pearson's R = 0.93, p < 0.0001) (Fig. 3). This strong correlation is also observed between in Sox10 and ErbB3 protein levels, as measured in a panel of brain tumors consisting of 8 pediatric PA, 3 pediatric GBM, 6 adult GBM, one adult AA, one adult AOA, and 8 pediatric EPN (Table 2). Statistical analysis showed a strong correlation between Sox10 and ErbB3 protein level (Pearson's R = 0.83, p = 0.011) using densitometry data from a selected Western blot analysis of 3 pediatric PA, one adult AA, one adult AOA, and 3 pediatric GBM (Fig. 4). Together, these data led support to the hypothesis that Sox10 is controlling ErbB3 overexpression in PA.
Gene expression microarray and Western blot data analyses show that Sox10 and ErbB3 are overexpressed in PA relative to other pediatric brain tumors, suggesting that this overexpression may be responsible for growth in PA.
Sox10 and ErbB3 Overexpression in Pilocytic Astrocytoma
In this study, we observed paired, significantly elevated overexpression of Sox10 and ErbB3 in PA compared with other pediatric brain tumors. A recent study by Bannykh et al also found Sox10 to be uniformly overexpressed in PA contrary to their assumption that Sox10 would only be expressed in oligodendrogliomas (20). The same study also found that Sox10 was present in a small number of higher-grade astrocytomas, which parallels our findings. Although we found that SOX10 and ERBB3 gene expression was uniformly low in pediatric brain tumors other than PA, Western blot analyses showed occasional high expression of ErbB3 protein expression in pediatric GBM, which again correlated with Sox10 expression. This latter finding differs from the results in a study by Schlegel et al who found no evidence of ErbB3 expression in adult grade II through IV diffuse astrocytomas (21). There are no published data on ErbB3 expression in PA with which to compare our results.
Sox 10 and ErbB3 in Normal Brain and the Origin of Pilocytic Astrocytoma
Insights into how Sox10 and ErbB3 might play a role in PA tumorigenesis can be gained by briefly reviewing the role of these factors in nervous system development. By controlling ErbB3 expression, Sox10 acts as a major regulator in the embryonic development of astrocytes, Schwann cells, and oligodendrocytes from pluripotent neural crest cells (18, 22). After neuregulin-ligand activation, ErbB3 drives cell growth and transformation by heterodimerizing with ErbB1 or ErbB2, thus activating a phospho-inositide 3-kinase (PI3K) and mitogen-activated protein kinase (23). ErbB3 preferentially forms heterodimers with ErbB2, inducing the most potent mitogenic signal among ErbB family members (24). The strong correlation between Sox10 and ErbB3 overexpression in PA suggests that, like in developing embryonic astrocytes, Sox10 is controlling ErbB3 expression in this tumor.
It is possible that in normal astrocytes, Sox10-controlled ErbB3 expression is switched off once embryonic astrocyte development is completed. PA could potentially arise from developing embryonic astrocytes in which Sox10 expression has failed to be switched off after astrocyte maturation and the cells later undergo oncogenic induction. Alternatively, neoplastic astrocytes and/or stem cells giving rise to PA may accrue specific oncologic alterations that allow Sox10 to again be reexpressed. Based on the results of this experiment, a plausible hypothesis is that aberrantly regulated expression of Sox10 in PA is responsible for overexpression of ErbB3 in PA and that subsequent overexpression of ErbB3 may be responsible for driving growth in PA. Unfortunately, it is difficult to identify corroborating evidence for this hypothesis at present, because data on Sox10 and ErbB3 expression during and after embryonic astrocyte development is scarce. Kulbrodt et al showed continued expression of Sox10 in adult rats after development, especially in Schwann cells, oligodendrocytes, and some glial lineages (25). Evaluation of SOX10 expression in normal brain was performed using the Serial Analysis of Gene Expression (SAGE) Anatomic Viewer (http://cgap.nci.nih.gov/SAGE/AnatomicViewer) (26). SOX10 was shown to be low to undetectable in normal human adult cerebellum, the common presentation site of PA. In human postmortem material, ErbB3 immunoreactivity was observed in oligodendrocytes and microglial cells, but not astrocytes (27). The SAGE Anatomic Viewer demonstrated that ERBB3 was absent in normal human adult cerebellum. Direct comparison between Sox10 and ErbB3 expression levels obtained in this study, and those outlined previously in this article, cannot be made. As a consequence, further studies are required to determine whether Sox10 are ErbB3 are aberrantly overexpressed in PA, or whether expression of these genes normally persists after embryonic astrocyte development and is therefore conserved in PA.
Sox10 and ErbB3 Overexpression as a Distinguishing Feature of Pilocytic Astrocytoma
PA differ from diffuse astrocytomas, grade II-IV, in that 1) the peak incidence is in children and young adults rather than middle to older aged adults; 2) they are generally more circumscribed and less infiltrative of surrounding tissues; 3) they are located in cerebellar, hypothalamic, optic nerve, and dorsal exophytic brainstem sites less frequented by diffuse astrocytomas; and 4) they display little or no tendency to upgrade and undergo malignant progression. These features collectively underscore fundamental differences in this class of astrocytomas from diffuse astrocytomas and also suggest the possibility that tumorigenesis in PA is related to, or recapitulates, developmental pathways. Because Sox10 and ErbB3 expression has been shown to be critical in development of the nervous system, we find it reasonable to suppose that PA might be caused by aberrations in developmental pathways. Pediatric tumors in general have strong links between embryogenesis and oncogenesis; examples of this include Wilms tumor, neuroblastoma, and hepatoblastoma. In contrast, although unlike some adult systemic tumors, the etiology of diffuse astrocytomas is unknown; these predominantly adult tumors probably arise from environmental mutagenesis. Not only does a relationship with developmental origin explain the significantly younger age at presentation of PA (because tumors displaying aberrations in development pathways usually manifest in childhood rather than adulthood), it also might explain their inability to upgrade. The infrequency of malignant transformation in PA may be the result of aberrant Sox10/ErbB3 overexpression maintaining astrocytic differentiation at near-normal/nonneoplastic levels.
Preclinical drug testing of gefitinib using in vitro PA cell cultures suggests that the drug inhibits the proliferation of PA by inhibition of a tyrosine kinase(s) other than its intended target ErbB1 (11). A number of clinical trials conducted on adult tumor types have sought to identify predictors of gefitinib sensitivity other than ErbB1 to better select patients who will be responsive to therapy. Recently, gefitinib was seen to be effective in treating nonsmall cell lung cancer, which expressed ErbB3, partly because of its ability to inhibit ErbB3 and thereby the downstream PI3K/Akt pathway (28, 29). We suggest that gefitinib may also be exerting its antiproliferative effect in PA cell cultures to some extent through the inhibition of ErbB3 activity. This possibility warrants prospective studies that might establish ErbB3-targeted inhibitors as effective drugs for treatment for PA located in unfavorable anatomic sites such as the hypothalamus and not amenable to surgical cure.
Despite the fact that PA is a WHO grade I tumor, treatment of recurrent/progressive PA remains a major challenge in pediatric neurooncology. The toxicity of the current chemotherapy and radiotherapy regimens, coupled with the lack of known molecular targets that might invite targeted therapies, impedes quality of life and diminishes prognoses as the tumor continues to progress and grow. Indeed, the standard chemotherapy regimen for progressive PA has not changed in the past 20 years. Part of the reason for this is the lack of biologic information on PA. The finding that both Sox10 and ErbB3 are overexpressed in PA helps to characterize this tumor and provides potential new targets for therapeutic intervention.