1Laboratorio de Oncogenética Molecular y Epigenética del Cáncer, Unidad de Investigación, Departamento de Cirugía Experimental, and ⁴Departamento de Neurocirugia, Hospital Universitario La Paz, 28046 Madrid, Spain.

Key Words: RB1, p14^ARF, p16^INK4a, CpG island, aberrant methylation, epigenetics, gliomas, astrocytoma, oligodendroglioma, ependymoma, glioblastoma

Abbreviations: central nervous system, (CNS); methylation-specific polymerase chain reaction, (MSP); neurofibromatosis 1, (NF1); polymerase chain reaction, (PCR); World Health Organisation, (WHO)

Primary brain tumors are neoplasms that originate from various intracranial tissues. About 17,000 new cases occur annually and primary cancer of the central nervous system (CNS) is the cause of death of approximately 13,000 individuals per year (Surawicz et al, 1998). More than 60% of all brain tumors have a glial origin, including pilocytic astrocytoma, low-grade astrocytoma, anaplastic astrocytoma, glioblastoma, oligodendroglioma, anaplastic oligodendroglioma, mixed oligoastrocytoma and low-grade and anaplastic ependymomas (Kleihues and Cavenee, 2000).

Pilocytic astrocytoma (a slow-growing tumor with a World Health Organization (WHO) grade I) is considered to be the most common glioma in children, accounting for 10% of cerebral and 85% of cerebellar astrocytomas. It constitutes the principal CNS neoplasm in neurofibromatosis 1 (NF1) (Burger et al, 2000). Cytogenetic analysis of pilocytic astrocytoma has revealed normal karyotypes or a variety of aberrations, primarily involving gains of chromosomes 7 and 8 (Rey et al, 1987; Karnes et al, 1992; White et al, 1995). Allelic losses at 1p36 or at 17q have been identified in a few cases (von Deimling et al, 1993; Bello et al, 1995), and comparative genomic hybridization analysis identified gains of chromosomes 19, 22 and 9q34.1-qter, and losses of chromosome 19 (Sanoudpu et al, 2000). Regarding TP53 gene mutations discordant data are available; early studies identified sequence changes in a few tumors (von Deimling et al, 1993), whereas 35% of samples (7 of 20) analyzed by Hayes et al. (1999) displayed mutations of this gene. The only consistent gene alteration described in this astrocytoma subtype is a loss of NF1 alleles that occurs in up to 90% of informative NF1-associated cases, in contrast to only 4% of sporadic tumors (Burger et al, 2000). LOH analysis at 1p, 10, 17 and 19q, and mutation detection at TP53, p16^INK4a and EGFR has been performed on 12 samples, including three NF1-associated tumors (Tada et al, 2003). None of the genetic abnormalities commonly detected in higher-grade astrocytomas were found in the sporadic cases. In contrast, LOH 10 and 17q (including the PTEN and NF1 regions, respectively) and homozygous deletion of p16^INK4a were identified in the NF1-associated samples. These data support the hypothesis that some NF1-associated pilocytic astrocytomas would differ genetically from sporadic cases.

Diffuse astrocytic gliomas are the most common primary neoplasm occurring in the CNS and are histologically classified as WHO grade II astrocytomas, WHO grade III anaplastic forms, and WHO grade IV glioblastoma (Kleihues and Cavenee, 2000). Low grade (WHO grade II) tumors and anaplastic grade III astrocytomas usually occur in adults and show a strong tendency toward progression. Glioblastoma, the most malignant subtype of glioma, may develop either from diffuse or anaplastic tumors (secondary glioblastoma) or de novo (primary glioblastoma) without a defined prior tumor lesion. Multiple genetic alterations have been identified in these astrocytic neoplasms; these alterations primarily involve inactivation or amplification/over-expression of TP53, p16^INK4a, RB1, PTEN, MDM2, and EGFR genes (Kleihues and Cavenee, 2000). Several other non-random anomalies are also characteristic features of these gliomas, including loss of heterozygosity at 1p, 10p, 10q, 11p, 19q and 22q, although the putative tumor suppressing genes remain unidentified. A distinct pattern of involvement of these genes and chromosomal regions characterizes both forms of glioblastoma. The main differences consist of EGFR gene amplification and TP53 mutations, which respectively characterize primary and secondary glioblastomas (Kleihues and Cavenee, 2000).

Tumors with a major oligodendroglial component account for 4% of all primary brain tumors and represent between 5% and 18% of all intracranial gliomas, including oligodendroglioma (WHO grade II), anaplastic oligodendroglioma (WHO grade III) and mixed oligoastrocytoma (Kleihues and Cavenee, 2000). They arise preferentially in the cerebral hemispheres of adult patients with a mean age at diagnosis of ~40 years. Low-grade oligodendrogliomas are characterized by a high incidence of loss of chromosome arms 1p and 19q, and anaplastic forms accumulate allelic losses on the short arm of chromosome 9 and on chromosome 10 (for review see Kleihues and Cavenee, 2000).

Ependymomas represent 3-9% of all intracranial brain tumors and about 60% of spinal tumors, and commonly arise in children (Kleihhues and Cavenee, 2000). Cytogenetic and molecular biology studies have demonstrated a preferential involvement of chromosome 22 (by losses), parallel to the inactivation of the NF2 gene (located at 22q12), primarily in sporadic cord tumors. Additional genomic abnormalities include chromosome 7 gains and overrepresentation of chromosomes 2, 5, 9, 12, 15, 18, 20q and X, and proportional losses of 13q. Losses of 6q and 9p, with gains of 1q, have primarily been found in intracranial ependymomas (Weremowicz et al, 1992; Rubio et al, 1994; Ebert et al, 1999; Hulsebos et al, 1999; Rousseau-Merk et al, 2000; Kraus et al, 2001; Alonso et al, 2002). These findings, thus, suggest that intracranial and spinal cord ependymomas progress along different genetic pathways that may influence differences in the clinical behavior of these gliomas.

Tumorogenesis of gliomas seems to be a multi-step process composed of genetic and epigenetic alterations involving tumor suppressor genes, cell cycle regulatory genes, oncogenes, and as yet unidentified genes located at specific chromosomal regions (Kleihues and Cavenee, 2000). Transcriptional silencing by hypermethylation of CpG islands located in the promoter regions is considered a common epigenetic mechanism for inactivation of tumor-related genes (Esteller, 2003). CpG islands are 0'5-2 Kb regions rich in cytosine-guanine dinucleotides, present in the 5' region of about half of all human genes (Baylin et al, 1998). Little information is available on the CpG island methylation status of neurogenic neoplasms. Isolated previous studies focus on high-grade astrocytomas, primarily the anaplastic forms and glioblastoma multiforme (Costello et al, 1996; Park et al, 2000; Nakamura et al, 2001a; 2001b; Yin et al, 2002; Gonzalez-Gomez et al, 2003a, 2003b; Uhlmann et al, 2003) and less frequently on low-grade astrocytomas (Costello et al, 2000; Gonzalez-Gomez et al, 2003a; 2003b; 2003c; Uhlmann et al, 2003), oligodendrogliomas (Watanabe et al, 2001a;Wolter et al, 2001; Yin et al, 2002; Alonso et al, 2003; Hong et al, 2003; Uhlmann et al, 2003), pilocytic astrocytoma (Gonzalez-Gomez et al, 2003c; Uhlmann et al, 2003) and ependymomas (Rousseau et al, 2003; Alonso et al, 2004).

In the present study we determined the frequency of methylation of three genes: RB1, p14^ARF and p16^INK4a in a series of 198 gliomas, including astrocytic, oligodendroglial and ependymal tumors, and in two normal brain tissue samples, using polymerase chain reaction (PCR)-based techniques involving sodium bisulfite modification of DNA (MSP) and sequencing of the PCR products.

Fresh tumor tissues were obtained from 198 patients with malignant gliomas, including 16 WHO grade I pilocytic astrocytomas; 26 WHO grade II diffuse astrocytomas; 23 WHO grade III anaplastic astrocytomas; 53 WHO grade IV glioblastomas multiformes (43 primary and 10 secondary); one giant cell astrocytoma; 24 WHO grade II oligodendrogliomas; 16 WHO grade III anaplastic oligodendrogliomas, six WHO grade II-III mixed oligo-astrocytomas; two WHO grade I ependymomas, 24 WHO grade II ependymomas, five anaplastic (WHO grade III) ependymomas, and two ependymoblastomas (WHO grade IV) as well as two nonneoplastic brain samples. Tumors were diagnosed according to the WHO guidelines (Kleihues and Cavenee, 2000), and the tumor cell content was estimated by histologic examination to be approximately 75-90%. DNA was prepared from frozen tissues using standard methods, as described (Rey et al, 1992).

Examples of methylation of all three genes, sequences from methylated alleles aligned against the corresponding wild type sequence, and bar diagrams that illustrate the methylation frequency of each gene in the different histological glioma subtypes are shown in Figure 1.

All but 92 study samples displayed CpG island hypermethylation in at least one gene (54%). The methylation frequency of the RB1, p14^ARF and p16^INK4a genes was 13%, 21%, and 37%, respectively. In contrast, methylation of these genes did not occur in either of the two normal brain samples. The methylation frequencies for the three genes in every tumor type studied are shown in Table 1.

Table 1: Methylation frequencies of cell-cycle control genes in histological subtypes of gliomas

PA: Pilocytic Astrocytoma; A: grade II Astrocytoma; AA: Anaplastic Astrocytoma; GB: Glioblastoma; GCA: Giant Cell Astrocytoma; O: grade II Oligodendroglioma; AO: Anaplastic Oligodendroglioma; OA: Mixed Oligo-astrocytoma; E-I: grade I Ependymoma; E: grade II Ependymoma; AE: Anaplastic Ependymoma; EB: Ependymoblastoma. N.D. Not done

Figure 1. Examples of MSP analysis for RB1, p14^ARF, and p16^INK4a in gliomas. Lane U, amplified product with primers recognizing unmethylated sequence; lane M, amplified product with primers recognizing methylated sequence. NC: negative control of methylation; PC: positive control of methylation; L: ladder. The forward sequences of representative methylated tumors are shown. Sequences are aligned against the wild-type (Wt) sequences showing a change of C to T by bisulfite treatment. C in CpG dicnucleotides that were methylated remains unaffected. The frequency of aberrant methylation for all three genes in every glioma subgroup is shown on the right. (The tumor nomenclature is the same described in Table 1).

RB1 was primarily methylated in astrocytic and mixed tumors, with frequencies ranging from 13% (in anaplastic astrocytoma) to 40% in secondary glioblastoma. On the other hand, pure oligodendrogliomas and ependymomas presented 0-4% aberrant RB1 methylation. In contrast, p14^ARF hypermethylation rates were higher in oligodendroglial tumors (46% and 50% in low-grade and anaplastic oligodendrogliomas, respectively), and values of about 20-30% characterized p14^ARF in anaplastic astrocytomas and ependymomas. On the other hand pilocytic astrocytomas and secondary glioblastoma displayed a 0-6% methylation rate for this gene. The p16^INK4a gene was hypermethylated by >35% of each subtype of astrocytic tumor, including pilocytic astrocytomas. However, the methylation rate for this gene was lower than 25% in pure oligodendrogliomas and ependymal tumors.

As shown in Table 2, simultaneous promoter methylation of two or three genes was demonstrated in some tumors as follows: RB1 and p16^INK4a in 7% (14/198) of cases, p16^INK4a together with p14^ARF in 8% (15/198) of cases, and RB1 plus p14^ARF in 0.5% (1/198) of cases. Three gliomas showed hypermethylation of all three genes (1.5%). Promoter methylation of either RB1 or p16^INK4a was present in 41% of samples (82/198).

In most tumors, gene hypermethylation was always accompanied by amplification of the unmethylated reaction (Figure 1). This finding was expected, since the tumor specimens were macroscopically isolated samples that contained both tumor and a small fraction of non-malignant tissues. Alternatively, cytosine hemimethylation, that is the cytosine on one chromosome is methylated but its homologue on the other chromosome is not, might also explain the finding. Sequencing of the corresponding methylated PCR products demonstrated the presence of invariable CpG as was expected (Figure 1).

The results presented here clearly demonstrate that CpG island hypermethylation of cell-cycle control genes is a frequent event in gliomas, as methylation of one or more genes was observed in 54% of the analyzed samples (106 of 198). Our study shows that hypermethylation rates vary by gene, and that it occurs in most instances at early stages of gliomagenesis, because it is already present in the low-grade forms. This was primarily evident for p16^INK4a, which presented methylation rates >40% in low-grade astrocytic tumors such as pilocytic astrocytoma and WHO grade II tumors. For its part p14^ARF was hypermethylated in <15% of these low-grade astrocytomas.

RB1 is located on the long arm of chromosome 13 (13q14) and is the classical example of a tumor suppressor gene. It encodes a nucleoprotein (pRB) that plays a key role in the cell cycle regulation complexes that govern the G1-S transition of cells, thus allowing mitosis and cell division (Friend et al, 1986; Lee et al, 1987).

Table 2: Isolated and concurrent methylation frequencies of cell-cycle control genes in gliomas

Loss of RB1 function has been described in a variety of tumor types, and hypermethylation of the promoter is recognized as an important RB1 silencing mechanism (Ohtani-Fujita et al, 1993). RB1 methylation was most frequent in astrocytic neoplasms in our tumor series and, in agreement with previous reports (Nakamura et al, 2001a; Gonzalez-Gomez et al, 2003a), we found RB1 promoter methylation more frequently in secondary (40%) than in primary (12%) glioblastomas. The non-random frequencies (13-35%) of aberrant RB1 methylation that we detected in the low-grade and anaplastic astrocytomas might be indicative of a subgroup of astrocytic tumors that further develop towards secondary glioblastoma, thus displaying a more aggressive biological behavior. In fact, loss of RB1 expression has been associated with a higher grade of malignancy in several human neoplasms (Cryns et al, 1994; Tsuda et al, 2000). In oligodendroglial and ependymal tumors we found very low rates or no methylation of this gene. Since RB1 inactivating mutations are also infrequent in these tumors (Gonzalez-Gomez et al, 2003a), the inactivation of the RB1 cell cycle control pathways should occur through the silencing of another alternative genes. However, oligo-astrocytomas presented a 17% rate of aberrant RB1 methylation, probably corresponding to the astrocytic component in these mixed tumors.

Low rates of p14^ARF hypermethylation were found in the pilocytic astrocytoma group (6%) and low-grade astrocytomas (15%) and this frequency increased slightly in anaplastic astrocytomas (22%). Loss of p14^ARF expression has been shown to be an important event in the genetic pathways for the development of both primary and secondary glioblastoma (Nakamura et al, 2001b; Ghimenti et al, 2003). Although promoter hypermethylation of this gene has been reported as more frequent in secondary tumors (Nakamura et al, 2001b), we found 21% p14^ARF methylation in primary glioblastoma and no secondary glioblastoma in our series displayed this anomaly. This finding may be due to the small number of secondary tumors included in our study; p14^ARF promoter methylation has been proposed as an early event in a subset of low-grade astrocytomas (as occurs in our tumor series) that may undergo malignant progression to secondary glioblastoma. Some of these highly malignant (secondary) tumors display homozygous deletion of p14^ARF (Nakamura et al, 2001b). With regard to oligodendroglial tumors, previous studies on p14^ARF methylation have shown variable results: Dong et al (2001) found 2%, whereas Wolter et al (2001) found 41%, and Watanabe et al (2001b) detected p14^ARF methylation in 21% of grade II oligodendrogliomas and 15% of the anaplastic forms. Our data show 46% and 50% in grade II and grade III forms, respectively, whereas 17% of our mixed tumors showed epigenetic alteration in this gene; these figures are closer to those we detected in astrocytomas. The epigenetic inactivation of p14^ARF is a frequent alteration in ependymal tumors, since about 30% of the samples (low-grade and anaplastic forms) displayed this alteration in our series. These findings agree with the data recently provided by Rousseau et al (2003), who detected 21% epigenetic change in this gene in this glioma subtype and, in agreement with their data, aberrant p14^ARF hypermethylation was more frequently identified in the intracranial ependymal tumors. An inverse correlation has been reported for p14^ARF and TP53 mutations in glioblastomas (Ichimura et al, 2000). Since oligodendrogliomas and ependymomas rarely present TP53 inactivating mutations (for review see Kleihues and Cavenee, 2000), silencing of the p14^ARF gene through aberrant promoter methylation may be a mechanism to inactivate the p14^ARF/MDM2/TP53 cell-cycle signaling pathway in these glioma subtypes. Recent data have demonstrated a high percentage of p14^ARF inactivation in glioblastomas with classical, astrocytic or oligodendroglial differentiation areas (Ghimenti et al, 2003), suggesting again that silencing this gene is an important step in tumorogenesis and/or progression of distinct glioma subtypes Aberrant methylation was the molecular gene silencing mechanism in some of those cases.

Oligodendrogliomas and ependymomas showed a lower rate of p16^INK4a promoter hypermethylation than astrocytic tumors. We found a 25% rate of p16^INK4a aberrant hypermethylation in oligodendrogliomas and a 12.5% rate in the anaplastic forms, as well as a 25% rate in the low-grade ependymomas. Dong et al, (2001), Walter et al (2001) and Watanabe et al, (2001) also showed variable results in methylation studies of this gene. They respectively report rates of 12%, 32% and 0%. Rates of about 21% aberrant methylation in the p16^INK4a gene were identified in ependymomas (Rousseau et al, 2003; Alonso et al, 2004). Variable frequencies of p16^INK4a hypermethylation have also been reported in secondary glioblastoma, and range from 70% to <5% (Costello et al, 1996; Fueyo et al, 1996; Hegi et al, 1997; Burns et al, 1998; Nakamura et al, 1998; Schmidt et al, 1998; Park et al, 2000; Yin et al, 2002; Gonzlaez-Gomez et al, 2003). Our results corroborate the proposal that a high rate of p16^INK4a CpG island hypermethylation is characteristic of both primary and secondary glioblastoma subtypes and also of the lower grade astrocytic forms. A 44% rate for p16^INK4a CpG island hypermethylation was found in the pilocytic astrocytomas included in our series. The mixed oligo-astrocytoma group was different from the pure oligodendrogliomas, perhaps due to their astrocytic component. Interestingly, an association between a worse prognosis and p16^INK4a inactivation either by deletion/mutation or aberrant promoter methylation has been recently reported in oligodendrogliomas (Bortolotto et al, 2000).

Concurrent aberrant hypermethylation was identified in several cases; for instance methylation occurred in both p16^INK4a and either RB1 or p14^ARF in 7-8% of cases, whereas only one sample (0.5%) displayed concurrent RB1 plus p14^ARF methylation. Accordingly, promoter hypermethylation of the p14^ARF gene seems to be independent of the methylation status of p16^INK4a even though their promoters are very close to each other on 9p21 and they share two exons, albeit in different reading frames (Sherr et al, 1996; Kamijo et al, 1998), and they are frequently co-deleted in glioblastoma (Newcomb et al, 2000). The transcriptional activity of p14^ARF is regulated independently of p16^INK4a and participates in a regulatory feedback loop with p53 and MDM2 (Kamijo et al, 1998). The p16^INK4a protein function is closely related to RB1 in cell cycle regulation; it regulates G1-S phase transition by inhibiting the activity of cyclin-dependent kinases CDK4 and CDK6. This cell-cycle control pathway was inactivated in 41% of the gliomas we studied, since methylation of both or either the RB1 or p16^INK4a gene occurred. Some cases in our series displayed a concurrent aberrant hypermethylation of both RB1 and p16^INK4a that might represent a redundant epigenetic alteration of this cell-cycle control pathway.

In conclusion, this study demonstrates that CpG island methylation of cell-cycle control genes is a common event in gliomas. It generally occurs at early stages of carcinogenesis, since it is already present in the low-grade forms. Some differences in the pattern of gene methylation were observed: hypermethylation of p14^ARF occurred more frequently in oligodendrogliomas and ependymomas than in astrocytic tumors. On the other hand p16^INK4a and RB1 were frequently methylated in astrocytic gliomas. In agreement with previous data (Gonzalez-Gomez et al, 2003c; Uhlmann et al, 2003) our findings suggest an important role for epigenetic changes in the development of pilocytic astrocytoma, a glial tumor in which no consistent genetic alteration has been identified previously. Finally, epigenetic inactivation of the cell-cycle control genes in some glioma subtypes might be indicative or predictive of an aggressive behavior: RB1 or p14^ARF in astrocytic tumors (Nakamura et al, 2001a; 2001b), or p16^INK4a in oligodendrogliomas (Bortolotto et al, 2000). Accordingly, therapies addressed to promoting re-expression of these genes (Swanton, 2004) might be useful in the management of glioma patients. An accurate glioma-subtype histological diagnosis together with an inequivocal identification of samples displaying promoter hyperemethylation, in combination with gene mutational/expression analyses, will contribute to optimizing the clinical application of molecular characterization of gliomas; this should lead to a firm establishment of predictive prognostic factors and specific therapies.

This study was supported by Grants 02/0669, 03/0235 from Fondo de Investigaciones Sanitarias (FIS), Ministerio de Sanidad, and Grant 08.1/0040/2003.1 from Consjeria de Educacion, Comunidad de Madrid, and grant from Fundación MAPFRE-Medicina. M. Eva Alonso is supported by a predoctoral felowship from Consejeria de Educacion, Comunidad de Madrid. C. Casartelli is supported by Fundaçao de Amparo a Pesquisa de estado do Sao Paolo (FAPESP) and Coordenaçao de Aperfeiçoamento de Pesoal de Nivel Superior (CAPES). We are indebted to Dr J.L. Sarasa and Dr. M. Gutierrez for histological diagnosis of tumor samples.