Associations Between ERCC2 Polymorphisms and Gliomas¹

Several rare cancer-prone syndromes caused by highly penetrant single gene defects have been described, including Li-Fraumeni and Bloom syndromes and XP.³ Diverse environmental factors have been implicated in epidemiological studies of glioma. These include radiation, chemicals, and viral or infectious exposures, but only therapeutic radiation appears to be a generally accepted risk factor (1). Certain rare genetic syndromes (e.g., neurofibromatosis, tuberous sclerosis, and Li-Fraumeni family cancer syndromes) are known to predispose to glioma, but collectively these account for a small proportion of cases (2, 3). Because substantial genetic heterogeneity exists even within tumors of the same histological subtype, it is generally believed that there are multiple pathways of genetic alterations leading to glioma (4). In addition, the same chromosomal region that harbors the ERCC2 gene and other NER proteins was found to be altered by gain and loss of 19q13.2–13.3 in a study of familial gliomas (5, 6) and nonfamilial gliomas (7). Recently, a study (8) showed that loss of 19q and 1p was significantly related to the overall survival of patients specifically with oligodendroglioma but not with astrocytomas, and loss of heterozygosity of 19q13.3 occurred most frequently in oligodendrogliomas (73% in the sample set; Ref. 6). Studies (9) on glioma cell lines have revealed translocations in the 19q13 region. Thus, an important glioma gene(s) may reside in the same region as ERCC2.

Patients with XP exhibit an increased sensitivity to sunlight and are prone to skin cancer from UV-induced DNA damage. Individuals with XP are also at risk for other cancers that result from unrepaired damage (10), including brain cancer (11). The studies of Dabholkar et al. (12) and Liang et al. (13) showed different tissue expression patterns and alterations in genomic copy number for ERCC2 in malignant brain tumors when compared with nonmalignant brain tissue. Both studies postulated that because of its role in correcting transcription-coupled DNA damage, the ERCC2 helicase of the DNA NER pathway may be involved in glial cell tumorigenesis.

Several different complementation groups have been characterized in these patients. The XP complementation group D (XPD)/ERCC2 (14, 15) gene functions as a helicase (16) in the NER pathway and transcription initiation by binding to the transcription factor IIH via p44 (17). Whereas ERCC2 protein restores NER in deficient cells (18), it does not seem to be required for in vivo transcription (19) but may enhance it (20). ERCC2 protein defects are also responsible for two additional (21) clinically distinct diseases including trichothiodystrophy and Cockayne syndrome. Individuals with trichothiodystrophy suffer from mental retardation, ichthyosis, brittle hair, and dwarfism. Cockayne syndrome patients with ERCC2 defects have XP-type neoplasia plus retinal degeneration, mental retardation, dwarfism, and hyperreflexia.

In addition to studies identifying disease-causing mutations in XP (17, 22, 23, 24, 25, 26), sequence analysis of the ERCC2 (GenBank accession no. L47234) gene revealed several DNA polymorphisms (27, 28). It is possible that genetic alterations in ERCC2 result in subtle effects on the efficiency of the NER pathway and may be involved in damage caused by environmental factors, including oxidative damage. To investigate potential relationships between the ERCC2 sequence alterations and their role in the etiology of gliomas, several polymorphisms were studied in a series of patients with glioma and controls to determine their prevalence in these populations.

The study population has been described previously (29, 30). Cases and controls for the study were drawn from the San Francisco Bay Area adult glioma study and discussed in detail elsewhere (3). Incident glioma cases (n = 492; age >20 years) were ascertained from August 1991 to April 1994 in six San Francisco Bay Area counties through the rapid case ascertainment service of the Northern California Cancer Center. Uniform neuropathology review indicated 4 cases were not glioma, and specimens could not be reviewed for 12 subjects. Thus, the parent study included 476 cases. A random digit-dialing technique was used to contact 462 controls, and these samples were frequency matched to the cases for gender, ethnicity, and age (3). Blood specimens were collected part of the way through the study (29, 30), and thus samples available for genotyping included 187 cases with pathology review and 171 controls. Of these, 161 cases and 164 controls were white. The protocol was reviewed and approved by the Committee for Human Subject Research at the University of California, San Francisco.

DNA was isolated from heparinized whole blood or tumor tissue using Chelex or Qiagen column purification. DNA was insufficient for any PCR amplification for two controls. The amplification primers for each polymorphism are given in Table 1. Each reaction included 0.2–0.6 μm primers mixed with 50 ng of genomic DNA, 1.0–2.0 mm MgCl₂, 0.8 units of Taq DNA polymerase, and 200 μm deoxynucleotide triphosphates in 10 mm Tris, 50 mm KCl (pH 8.3) in a final reaction volume of 50 μl. Amplification was carried out by initial denaturation at 95°C for 5 min followed by 35 cycles of denaturation at 95°C for 15 s, annealing at 57–59°C for 15 s, and extension at 72°C for 30 s. A final extension step was carried out for 5 min at 72°C. Restriction digestion or allele-specific oligonucleotide hybridization was used to detect the polymorphisms (Table 1). Digestion products were electrophoresed for 30 min at 150 V on 3% NuSieve:1% LE agarose (FMC Biochemicals) and stained with ethidium bromide (10 μg/ml). Hybridizations were carried out using tetramethylammonium chloride at 52°C using [γ-³²P]ATP-labeled probes.

Genotype and allele frequencies among cases and controls were compared, and χ² tests were conducted to assess similarity of genotype frequencies. In addition, genotype frequencies were compared in three histological subgroups of cases for which there were sufficient numbers for separate analyses [glioblastoma multiforme, astrocytoma (including anaplastic astrocytoma), and oligoastrocytoma]. Next, the frequencies of all of the cases, case by histological subgroup (n = 3 subgroups), and controls were computed for each polymorphism. This analysis was done by comparing one versus no variant alleles first and then by comparing two versus one or no variant alleles. Logistic regression was used to estimate unadjusted, age-adjusted, and gender-adjusted ORs for having the indicated genotype in all of the cases versus controls and then in each histological category versus controls. Kendall correlations were used to assess associations among ERCC2 genotypes. Wilcoxon tests and ANOVA were used to compare median and mean ages of subjects with different genotypes, respectively. Goodness of fit to Hardy-Weinberg equilibrium expectations was assessed by χ² test for each polymorphism.

Table 2 shows the demographics of the overall population, the population genotyped, and whites genotyped. There was a higher proportion of men among genotyped subjects than among the parent population. Genotyped cases were significantly younger (P = 0.003) than the overall case population (48.3 ± 1.1 years versus 54.2 ± 0.8 years) and less likely to have aggressive tumors (i.e., glioblastoma) than those in the parent study (44% versus 59%, respectively). This is most likely because of the decreased survival of older individuals with more aggressive tumors during the lag time between diagnosis, ascertainment, and follow-up venipuncture. Because the preponderance of genotyped study participants (86% of cases and 96% of controls) were white, the results presented here are restricted to this group.

A total of eight polymorphisms were studied (Table 1). Three were silent changes that did not result in amino acid substitutions. These included R156R, A575A, and D711D. The remaining five nucleotide changes included: I199M, H201Y, D312N, R616C, and K751Q. The variant allele frequencies for I199M, H201Y, A575A, and R616C were very low (none detected, 0.7%, 0.5%, and 0.4%, respectively) in this population and were excluded from additional statistical analyses. Genotype and allele frequencies for the remaining four variants are given in Table 3. The genotype frequencies of each polymorphism appeared to be in Hardy-Weinberg equilibrium in all of the controls. Among cases, the R156R (χ² = 7.4; P = 0.01), D312N (χ² = 4.7; P = 0.03), and D711D (χ² = 4.1; P = 0.04) polymorphisms appeared not to be in Hardy-Weinberg equilibrium. Analyses in histological subgroups of cases revealed significant differences from Hardy-Weinberg expectations for polymorphisms in R156R in subjects with either glioblastoma or oligoastrocytomas, in D312N in subjects with glioblastoma, and in D711D in patients with astrocytoma. Discrepancies in the numbers of study subjects shown in Table 3 and Table 4 (see below) compared with the summary (Table 2) were because of technical difficulties with the amplification of some of the DNA samples.

Because the R616P mutation resulted in a null allele (30) and was not found to be causative for XP, it was chosen originally for inclusion in this study. A BsaHI RFLP assay was designed to detect the R616P variant (see Table 1). The consensus for the restriction endonuclease is 5′ G[A,G]CG[C,T]C 3′. When a detected positive sample was sequenced for confirmation, the expected CGT to CCT change was not identified, but rather a CGT to TGT change was observed resulting in the rare, newly identified R616C polymorphism. Previously, two other sequence variations, R616P and a double mutant, R616W (CGT to TGG), have been reported (22, 24).

Cases were more likely than controls to have the variant nucleotide at R156R (P = 0.07) but were less likely than controls to carry variant nucleotides at D312N, D711D, and K751Q. χ² tests for each of three genotypes in subgroups of patients versus controls revealed that patients with oligoastrocytomas had a significantly different distribution of genotypes than controls at codon positions R156R (P = 0.0002), D312N (P = 0.03), and D711D (P = 0.01).

Cases were significantly more likely than controls to be homozygous variant AA at R156R, regardless of histology (see Table 4; age- and gender-adjusted OR, 2.3; 95% CI, 1.3–4.2). The association was strongest for patients with oligoastrocytoma versus controls (see Table 4). Cases were less likely than controls to carry a variant nucleotide at each of the codons D312N, D711D, and K751Q (see Table 4), but none of the differences achieved statistical significance after adjustment for age and gender. Within each category of subjects, cases and controls, median or mean ages of people with different genotypes did not differ significantly.

Variants at D312N, D711D, and K751Q were highly significantly correlated with each other and were each strongly negatively correlated with the presence of the variant nucleotide at R156R (Table 5). That is, an individual carrying a variant at D312N, D711D, and K751Q was more likely to carry the variant at the others and less likely to carry the variant at R165R. This was true regardless of case status. The positive correlation was particularly striking between D711D and K751Q, which are only 605 bases apart on the chromosome. The smallest correlation was observed for the two polymorphisms in which the genetic distance is greatest (R156R and K751Q; 13.4 kb).

The evolving field of SNP analysis is becoming more and more widely used for association studies to map and study genetic variation (31, 32, 33). Several recent studies (34, 35, 36) have shown that in many instances it is not appropriate to determine the frequency of one given SNP to determine its association with a complex disease. The effects of a given polymorphism considered separately may be hampered by haplotype differences and linkage disequilibrium (36). The study reported here shows that when eight different polymorphisms are studied in the ERCC2 gene, one variant is more frequently associated with cases. However, all of the remaining variant alleles (three of which occurred with sufficient frequency for study) occur at higher frequencies in the control population, raising the possibility that these three additional variants may be inversely associated. These variants are also positively correlated in the population, irrespective of case status, such that carriers of any one of these are more likely to carry the others. The utility of SNP data varies depending on the goals of the study (31, 34) because linkage disequilibrium may occur at very large distances in some instances and small distances in other genomic regions. Thus, data examining multiple SNPs in a gene become difficult to analyze, unless haplotypes can be established.

Observed significant ORs for the R156R variant for the overall case group (OR, 2.3) ranging to (OR, 3.2) for cases with oligoastrocytoma are somewhat surprising, because this variant most associated with case-control status is a silent polymorphism in exon 6 of the gene. Dybdahl et al. (37) also found that the AA variant was associated with basal cell carcinomas in psoriatics. The significance of these findings will have to await functional analysis of this as well as the other polymorphisms. Additionally, in this same series of gliomas, Chen et al. (38) studied an A to C polymorphism in the 3′ untranslated region of the ERCC1 gene, which curiously resides in the coding region of a nucleolar gene on the opposite strand (ASE-1/CAST). ERCC2 and ERCC1 reside in the 19q region, raising the possibility that multiple polymorphisms in multiple NER genes increase the risk of disease. An alternative explanation is that both of these genes are in linkage disequilibrium with another nearby gene. One such potential candidate, the human glia maturation factor-γ gene, is located on chromosome 19q13.2. Mutational analysis of this gene did not reveal any sequence variations in glioma patients, whether or not they demonstrated loss of heterozygosity (39). It is also known that the kinesin light chain (KLC2) gene lies tail to tail with ERCC2 (40).

The different frequencies of the R156R polymorphism among glioma cases and controls could be attributable to either an influence of ERCC2 or a nearby gene on gliomagenesis, progression, or survival from glioma. The latter possibilities are suggested by the fact that this polymorphism is a silent nucleotide variation and by the observation that it and two other polymorphisms were not in Hardy-Weinberg equilibrium. Despite using a population-based rapid case ascertainment program that on average identified cases within 2 months of diagnosis, 44% of cases were deceased by the time of study contact because of the very poor survival associated with glioblastoma. This required interviews with proxies and precluded specimen collection for such subjects. The problem was further compounded because we did not begin blood collection until part of the way through the study. Studies in which specimens for constitutive DNA analyses are collected at the time of diagnosis will be necessary to distinguish whether ERCC2 or a nearby gene is involved in tumor formation or progression and treatment response. Another potential weakness of our study was that our response rate among controls in the parent study was only 63%, but it seems unlikely that ERCC2 genotype would influence control participation. This study had 77% power to detect the observed 2.1-fold OR for the R156R variant at the 0.05 level of significance but substantially less power for the weaker associations observed for the more prevalent variants at the other loci. Additional studies of these polymorphisms in other populations will be necessary to confirm these results, and larger sample sizes will be needed to examine potentially relevant gene-environment interactions.

We thank the Molecular Genetics Core at the Wadsworth Center for oligonucleotide synthesis and sequence analysis. We thank Salvatore Duva for expert technical assistance.