An International Case-Control Study of Glutathione Transferase and Functionally Related Polymorphisms and Risk of Primary Adult Brain Tumors

Glutathione transferase (GST) is important for the detoxification of exogenous and endogenous substances, including protecting cells from the toxic effects of reactive oxygen. A number of polymorphic GST enzymes are expressed in the human brain, including GSTP1 and GSTM3 (1). The GSTM3*B allele of the GSTM3 *A/*B single nucleotide polymorphism (SNP), in intron 6, may affect the regulation and ultimately the amount and activity of GSTM3 (2). Although results from a small study of risk factors for astrocytoma did not indicate an effect of this GSTM3 SNP on high-grade astrocytoma (3), findings from a recent large hospital-based case-control study (n = 489 glioma cases, 197 meningioma cases, and 799 controls) suggest increased risks of glioma [odds ratio (OR), 2.3; 95% confidence interval (95% CI), 1.0, 5.2] and meningioma (OR, 3.6; 95% CI, 1.3, 9.8) among patients with the GSTM3 *B/*B genotype compared with those with the *A/*A genotype (4). The investigators report even higher risks associated with both tumors among those with the GSTM3 *B/*B genotype who had ever smoked cigarettes. In addition, a novel finding of a polymorphism in the promoter region of GSTM3 was recently reported (5). This GSTM3 −63A/C SNP also seems to have a major impact on the regulation of GSTM3 expression in vivo, with the C allele associated with considerably lower expression of GSTM3.

The GSTP1 protein is the most highly expressed GST enzyme in the brain (1), and this enzyme is overexpressed in several precancerous and malignant tumors (6), including malignant glioma. There are five known studies that have evaluated the association between the GSTP1 105 SNP and adult brain tumors (7-11) and three that have examined the association between the GSTP1 114 SNP and adult brain tumors (8, 9, 11). Results for both SNPs are inconsistent; for example, using data from a hospital-based case-control study, De Roos et al. (8) reported an association between the GSTP1 105 Val/Val genotype and glioma (OR, 1.8; 95% CI, 1.2-2.7). In contrast, in a large population-based case-control study, Wrensch et al. (11) found no evidence for an association between either the GSTP1 105 Val/Val genotype (OR, 0.68; 95% CI, 0.43, 1.10) or the GSTP1 114 Ala/Val or Val/Val genotypes and glioma (OR, 1.24; 95% CI, 0.84, 1.80). Differences between studies may arise from sampling variation or control selection procedures. One previous study of GSTP1 used control values from the published literature rather than concurrent controls (9), and a second used hospital controls with diagnoses (i.e., “a variety of non-neoplastic conditions”) that may not be as likely to cause hospital admission as is a diagnosis of glioma (8). As Schwartzbaum et al. (12) explain, different case-control hospital selection probabilities may produce biased effect estimates. There is conflicting evidence for an association between cigarette consumption and brain tumors (4); however, two recent cohort studies have provided some support for a link between cigarette consumption and brain tumor risk (13, 14).

In view of both the findings and inconsistencies of the previous literature, we decided to reevaluate potential associations between GSTP1 105, GSTP1 114, and GSTM3 *A/*B genotypes and adult brain tumors. We also examined a potential association between the GSTM3 −63 polymorphism and adult brain tumors for the first time. In addition, we evaluated the interaction between effects of these polymorphisms and cigarette smoking on adult brain tumor risk. Haplotypes consist of polymorphisms on the same chromosome that may be inherited together, and we considered whether haplotypes based on the GSTP1 105 and 114 polymorphisms or GSTM3 −63 and GSTM3 *A/*B polymorphisms might be related to adult brain tumor risk.

Genotyping was first conducted in a preliminary study restricted to Swedish cases and controls, where, in addition to the above-mentioned genes, we also included analyses of germ line variants in the GSTM1, GSTT1, NQO1, and CYP1A1 genes in relation to adult brain tumor risk. The NQO1 and CYP1A1 enzymes may alter susceptibility to oxidative damage in the brain (15). Variants on these four genes were not further analyzed in the extended material because high-throughput genotyping methods were not available (GSTM1, GSTT1) or because both preliminary results and previous literature (16, 17) indicated no associations with brain tumors (NQO1, CYP1A1).

Four population-based case-control studies of primary adult brain tumors were carried out in Sweden, southeast England, Denmark, and Finland. These studies were conducted in the Stockholm, Lund, Göteborg, and Umeå regions of Sweden, the Thames regions of southeast England, throughout Denmark, and all regions of Finland except Northern Lapland and Åland. All four studies followed the core protocol of the Interphone Study, coordinated by the IARC (18), but with extensions to the study design such as a wider age range of patients, an extended questionnaire, and collection of blood samples.

Adult brain tumor cases were identified through neurosurgery, neuropathology, oncology, and neurology centers. Lists of case patients were also obtained from the appropriate population-based cancer registries to evaluate and improve completeness of ascertainment. Eligible cases were patients with glioma, including glioblastoma, [International Classification of Diseases (ICD), 10th revision, code C71; International Classification of Diseases for oncology (ICD-O), 2nd ed., codes 9380-9384, 9390-9394, 9400-9401, 9410-9411, 9420-9424, 9430, 9440-9443, 9450-9451, 9460, 9480-9481, and 9505] and meningioma (ICD 10th revision, code C70; ICD-O, 2nd ed. codes 9530-9539). Eligible case patients were individuals diagnosed with primary brain tumors between September 1, 2000 to February 29, 2004 (the dates of case ascertainment within this period vary between centers), at ages 20 to 69 years in the Nordic countries and 18 to 59 years in England, and resident in the study region at the time of diagnosis.

Controls in the Nordic centers were randomly selected from the population register for each study area and frequency matched to all adult brain tumor cases on age, sex, and region. In England, frequency-matched controls were randomly selected from general practitioners' practice lists. Controls were subject to the same age and residence criteria as case patients and were excluded if they had ever been diagnosed with a brain tumor.

Eligible study participants were invited to take part in the study by letter. If no reply was received, a repeat letter was sent, or the potential participant was contacted by telephone. These procedures varied among countries; for example, in Finland, potential participants were contacted by telephone only if they did not respond to the second letter. Each study was approved by the local ethics committees, and informed consent was obtained from all study participants before the start of the interview. The median time from glioma diagnosis to interview was 2.6 months, whereas that from meningioma diagnosis to interview was 3.5 months.

Computer-assisted personal interviews were conducted in participants' homes or other convenient locations (e.g., hospital rooms, offices, etc.) by trained interviewers. Information was collected on various potential risk factors for brain tumors, such as diagnostic and therapeutic radiation, medical history, smoking habits, mobile phone use, and occupational history. In Sweden, southeast England, and Denmark, a blood sample was drawn from cases and controls who participated in the interview and agreed to give blood (Table 1). Finnish investigators did not attempt to collect blood from all interviewed participants; rather, they drew blood from a convenience sample, of predetermined size, from patients in three hospitals or controls living in the area served by these hospitals. During the data collection period, all eligible Finnish participants were asked to donate blood samples. The required number of samples was obtained by the end of the study period.

Dynamic allele-specific hybridization (DASH; refs. 19-23) was conducted to identify the GSTM3 *A/*B and GSTP1 105 and 114 genotypes. For this, two PCR primers and one DASH probe per target SNP were designed by means of custom software (24) supplied by DynaMetrix Ltd. (Leicester, England, United Kingdom). To verify successful PCR amplification, several randomly chosen samples were examined on a 3.0% low-melt agarose gel. DASH analysis of the PCR product was then conduced on membrane macroarrays, using the DASH-2 protocol (22). No samples were sequenced because DASH is an extensively validated method (19-23). Quality control data included in the preceding references confirm that the error rate of the method across many different SNPs is <0.1%.

Determination of GSTM3 −63 genotypes was done with matrix-assisted laser desorption/ionization time-of-flight (MALDI-TOF) mass spectrometry (Sequenom Inc., San Diego, CA) of allele-specific primer extension products. All assays for the PCR and associated extension reaction were designed using the SpectroDESIGNER software (Sequenom). Cleaned extension products were analyzed using a mass array mass spectrometer (Bruker Daltonik GmbH, Bremen, Germany), and peaks were identified using the SpectroTYPER RT 2.0 software (Sequenom).

Genotyping for the GSTM1 and GSTT1 null polymorphisms was done as previously described (25, 26). The presence of GSTM1 or GSTT1 was analyzed by multiplex PCR with β-actin as a positive control. This PCR method does not differentiate between the heterozygous and homozygous carriers of the functional genes. The NAD(P)H quinone oxidoreductase (NQO1) Pro¹⁸⁷Ser polymorphism was determined using the HinfI PCR-RFLP method of Traver et al. (27). The polymorphism in exon 7 of cytochrome P450 1A1 (CYP1A1), which arises from an A>G base change and results in the replacement of isoleucine by valine at residue 462 in the heme-binding region of the enzyme, was analyzed by allele-specific PCR, as previously described (25). The neighboring Thr⁴⁶¹Asp polymorphism does not interfere with the allele-specific method because the primers used do not cover this polymorphic region.

Associations between polymorphisms and brain tumors were assessed using unconditional logistic regression, adjusted for the matching variables sex and age at diagnosis for cases (or for the controls, their age at their interview date with adjustment for the interval between the diagnostic and interview date of the cases and the difference between the mean interview dates of cases and controls) and country of residence. We also examined whether the region within each country on which controls were also matched (except in Denmark) should be included in all models. To do so, we conducted within-country analyses adjusted for age, sex, and region and found that results were almost identical to within-country analyses adjusted for age and sex only. We therefore did not use region within country of residence as a potentially confounding variable.

To find out whether we could legitimately estimate the combined effects of polymorphisms across countries of residence, we conducted a global test for heterogeneity by country. To do so, for each polymorphism, we compared the model including main SNP and country effects and SNP-country interaction terms to a model with only SNP and country main effects and carried out a log likelihood ratio test between them. We found no evidence for heterogeneity using a significance level of 10%. We therefore do not present country-specific results.

To evaluate genotype trend, we coded each genotype as an ordinal categorical variable, included it in a logistic regression model, and then used the P value associated with its regression coefficient (Proc Logistic, SAS, Cary, North Carolina).

Haplotype probabilities for the GSTM3 (GSTM3 −63 and GSTM3 *A/*B) and GSTP1 (GSTP1 105, GSTP1 114) sequence variants were estimated using Proc Haplotype's (SAS Genetics) implementation of the expectation-maximization (EM) algorithm. Each individual in the sample is assigned a probability of having each haplotype or combination of SNPs (e.g., for the GSTP1 105 and 114 SNPS, the haplotypes are A-C, G-C, G-T, A-T, the corresponding amino acids are: Ile/Ala, Val/Ala, Val/Val, Ile/Val). The probability assigned by the program is based on the individual's genotype together with the prevalence of the SNPs in the population. These probabilities are then used as weights to estimate haplotype frequencies. We excluded haplotypes from the analysis with frequencies of <5% (e.g., GSTP1 105/114 A-T) because they are unstable and more likely than are more prevalent haplotypes to produce false-positive results (28).

We used Proc Allele (SAS Genetics) to test for Hardy-Weinberg equilibrium among controls (an assumption required for Proc Haplotype results to be valid) and found that genotype frequencies were consistent with those expected under Hardy-Weinberg equilibrium. We next estimated sex-, age-, and country-adjusted odds ratios for each haplotype using unconditional logistic regression.

We estimated the biological interaction between the effects of the GSTM3 and GSTP1 SNPs and smoking based on an additive model (29). To do so, we used the adaptation of Proc Logist (SAS), a logistic regression program provided by Andersson et al. (30), for obtaining odds ratios and confidence intervals for additive interaction models.

In the three centers where all subjects were asked to give a blood sample total, blood samples were collected from 72% of interviewed controls, 64% of glioma cases, and 70% of meningioma cases. In Finland, blood sample collection was limited to a predetermined number of samples from participants at three hospitals and residents living in areas served by these hospitals. This sampling strategy resulted in collection of blood samples from 13% of Finnish controls, 38% of Finnish glioma, and 23% of Finnish meningioma cases. Participants from all four countries who had their blood drawn did not differ from the total sample interviewed with respect to sex or age (Table 1).

Table 2 shows results of analyses of GSTP1 and GSTM3 genotypes and is based on analyses of DNA from 725 people diagnosed with glioma, including 329 diagnosed with glioblastoma, and 546 diagnosed with meningioma in addition to 1,612 controls. Table 3 is restricted to the Swedish sample and includes analyses of GSTM1, GSTT1, NQO1, and CYP1A1 genotypes in 231 people diagnosed with glioma, 112 of whom were diagnosed with glioblastoma, 184 people diagnosed with meningioma, and 441 controls.

All odds ratios in Tables 2 and 3 are close to unity and all confidence intervals, with the exception of that for the GSTP1 105 Ile/Val genotype for glioma, include the null. However, results in Table 3 are based on small numbers of observations and, therefore, must be considered preliminary. In further analyses, we did not find an elevated risk of glioma among people with the GSTM3 *B/*B genotype compared with those with the *A/*A genotype as had De Roos et al. (8). Our glioma result for this De Roos et al. analysis was OR, 1.13; 95% CI, 0.65, 1.96 and our meningioma finding for the same genotypes was OR, 1.02; 95% CI, 0.53, 2.00.

Table 4 shows results of the haplotype analyses. A weak association is indicated between the G-C (Val-Ala) GSTP1 105/114 haplotype and glioma (OR, 0.73; 95% CI, 0.54, 0.99), whereas all other ORs are close to unity.

In Table 5, there is no evidence of an interaction between the effects of either of the GSTP1 or GSTM3 SNPs and cigarette smoking. Table 6 shows results of genotype combination analyses. Again, all ORs are near the null value, but some deviation exists. The combined GSTP1 105 Ile/Val and GSTP1 114 Ala/Ala genotypes are inversely related to glioma (OR, 0.74; 95% CI, 0.59, 0.91) and to glioblastoma (OR, 0.75; 95% CI, 0.56, 1.01), but there are no other noteworthy associations among the remaining genotypes and glioma or meningioma.

In the largest reported population-based case-control study of the association between GSTP1 and GSTM3 polymorphisms, we found no strong evidence for an association between GST polymorphisms and adult brain tumors. We did, however, observe some evidence of a slightly decreased risk of glioma among people with the GSTP1 105/114 G-C (Val-Ala) haplotype and glioma. Whether this small effect is merely a false-positive finding is not known. A recent meta-analysis (31) of the GSTP1 105 and GSTP1 114 SNPs found no evidence for associations between either SNP and adult brain tumors; however, none of the authors whose work was summarized had conducted haplotype analyses.

We did not confirm the recent findings of De Roos et al. who observed an elevated risk of glioma and meningioma among individuals with the GSTM3 *B/*B genotype relative to the *A/*A genotype (8), nor did we find the even stronger association among individuals with the GSTM3 *B/*B genotype who ever smoked cigarettes that these authors observed. Although the De Roos et al. sample was based on a relatively large number of glioma (n = 489) and meningioma cases (n = 197) and hospital controls (n = 799), the number of people with the *B/*B genotype in their data set was relatively small (n = 14 glioma cases, n = 10 meningioma cases, and n = 12 controls), and the numbers of smokers with this rare genotype were, of course, even smaller. Our data set contained a slightly larger number of cases with the *B/*B genotype for glioma (n = 21) and about the same for meningioma (n = 12), but we had a larger number of controls (n = 43). We also found no association between the newly identified GSTM3 −63 A/C polymorphism and adult brain tumor risk. As the −63C allele has been shown to be associated with lower GSTM3 expression (5), these results suggest that low GSTM3 levels are of minor or no importance for risk of developing brain tumors.

We were also unable to confirm associations between GSTP1 105 and GSTP1 114 genotype combinations initially observed in a small study conducted by Ali-Osman et al. (32) and later found in a larger study conducted by De Roos et al. (8). We did, however, find a weak association between the genotype combination GSTP1 105 Ile/Val and GSTP1 114 Ala/Ala and glioma risk (Table 6). However, the component genotypes (Table 2) and related G-C haplotype (Table 4) are inversely related to glioma. Therefore, findings in Table 6 may simply represent a repetition of these genotype and haplotype results. Our study of the Swedish data did not confirm the meta-analytic findings of Lai et al. (31), who observed associations between GSTT1 null and meningioma, but they did confirm the absence of associations between the NQO1 and CYP1A1 SNPs and adult brain tumor risk found in three previous studies (4, 16, 17). However, our preliminary sample was relatively small, and our results are therefore not definitive.

Although our analysis of the GSTM3 and GSTP1 SNPs is the largest reported to date, we did not have sufficient precision in our estimates to exclude small increases or decreases in risk. In spite of this limitation, we may have identified a small decrease in glioma risk associated with the G-C GSTP1 105/114 haplotype. This finding is not consistent with experimental results of Hu et al. (33) that suggest that the G-T GSTP1 105/114 haplotype would be more effective than the G-C or A-C haplotypes in providing protection against the carcinogen benzo(a)pyrene. However, levels of protection of different combinations of GSTP1 105/114 alleles may depend on the specific chemical structure of each carcinogen. In addition, haplotype analysis is an estimation procedure that uses methods similar to those used to estimate the effects of missing data (34) and, as such, is subject to error. However, unlike missing data analysis, the genotypes that form the basis of haplotype analysis provide information about the probability of occurrence of specific haplotypes. Some evidence for the validity of our haplotype findings comes from the consistency of these findings with results from our individual genotype analyses. A final point to consider in evaluating our results is that in the main body of our study we examined, only four variants of two GST genes but detoxification enzymes produced by GST genes are unlikely to work in isolation (6). An understanding of whether GST polymorphisms and two functionally related polymorphisms affect brain tumor risk is not a question that can be restricted to a few GST genetic variants or even exclusively restricted to GST genetic variants. Therefore, in addition to variables that represent environmental and endogenous toxins, a valid analysis of effects of GST polymorphisms on brain tumor risk should account for effects of additional genes on GST detoxification pathways.

Wacholder et al. (28) argue persuasively that prior knowledge determines the probability of false-positive findings. Prior findings from the previous meta-analysis of the GSTP1 105 and GSTP1 114 SNPs and brain tumors suggest that these SNPs are unrelated to adult brain tumor risk, but more important, are the general failure of study results of genetic variants to be replicated. However, the present study is the largest reported to date and may, therefore, have been able to identify patterns that other smaller studies were not able to detect. In addition, none of the previous studies reported results of haplotype analyses, so it is possible that because haplotypes more closely reflect the functional effects of genetic variants than do individual SNPs, we were able to observe associations that others were not.

Overall, our results did not indicate an association between GST or functionally related polymorphisms and adult brain tumor risk, with the possible exception of a weak effect of the G-C (Val-Ala) GSTP1 105/114 haplotype on glioma risk. In view of the inherent risks of drawing conclusions based on results of one study, particularly a study of the effect of SNPs on cancer risk, we reserve judgment on the meaning of our findings until others have attempted to replicate them.

The Nordic-United Kingdom collaborative group thanks all participants and those who assisted with data collection for their valuable contribution to this study. We thank the International Agency for Research on Cancer (IARC) team, in particular Elisabeth Cardis, Isabelle Deltour, and Lesley Richardson, for their input in this study, and James Doughty and Jan Ivar Martinsen for programming work. The Swedish center thanks the Swedish Regional Cancer Registries and the hospital staff, especially the following key persons at the hospitals: Dr. J. Boethius, Dr. O. Flodmark, Prof. I. Langmoen, Dr. A. Lilja, Dr. T. Mathiesen, Dr. I. Olsson Lindblom, and Dr. H. Stibler (Karolinska University Hospital); Dr. J. Lycke, Dr. A. Michanek, and Prof. L. Pellettieri (Sahlgrenska University Hospital); Prof. T. Möller and Prof. L. Salford (Lund University Hospital); Dr. T. Bergenheim and Dr. L. Damber (Umeå University Hospital). The Finnish center thanks Ph.D. Sirpa Heinävaara (STUK); Dr. J. Jääskeläinen (Helsinki University Hospital); Dr. S. Valtonen (Turku University Hospital); Prof. J. Koivukangas (Oulu University Hospital); Prof. M. Vapalahti (Kuopio University Hospital); Dr. T. Kuurne and Dr. H. Haapasalo (Tampere University Hospital); and Prof. R. Sankila (Finnish Cancer Registry). The southeast England center thanks Deborah Hogben, who was responsible for study administration, and the research nurses Alison Butlin, Alison Hart, Margo Pelerin, Rebecca Knight, Caroline Parsley, Jennifer Owens, Karen Sampson, and Maureen Swanwick. They also thank Prof. H. Møller, B. Plewa, and S. Richards from the Thames Cancer Registry and the following neuropathologists, neurosurgeons, neuro-oncologists, clinical oncologists, neurologists, administrators, and secretaries for the help they provided: D.G. Hardy, P.J. Kilpatrick, R. Macfarlane (Addenbrooke's Hospital); M. Cronin, T. Foster, S. Furey, Dr. M.G. Glaser, F. Jones, N.D. Mendoza, Prof. E.S. Newlands, K.S. O'Neill, D. Peterson, F. Taylor, Prof. J. van Dellon (Charing Cross Hospital); Dr. J.J. Bending (Eastbourne District Hospital); P.R. Bullock, C. Chandler, B. Chitnavis, L. Doey, R.W. Gullan, Prof. C.E. Polkey, R. Selway, M.M. Sharr, L. Smith, Prof. A.J. Strong, N. Thomas (King's College Hospital); Dr. G.M. Sadler (Maidstone Hospital); Dr. S. Short (Mount Vernon Hospital); Prof. S. Brandner, G. Brookes, A.D. Cheesman, Prof. M.J. Gleeson, J.P. Grieve, W.J. Harkness, Dr. R. Kapoor, N.D. Kitchen, T. Pearce, M.P. Powell, Dr. J. Rees, Prof. F. Scaravilli, Prof. D.T. Thomas, L.D. Watkins (National Hospital for Neurology and Neurosurgery); A.R. Aspoas, S. Bavetta, J.C. Benjamin, K.M. David, J.R. Pollock, Dr. E. Sims (Oldchurch Hospital); J. Armstrong, J. Akinwunmi, G. Critchley, L. Gunasekera, C. Hardwidge, J.S. Norris, Dr. P.E. Rose, P.H. Walter, P.J. Ward, Dr. M. Wilkins (Princess Royal Hospital); Prof. T.Z. Aziz, Prof. D. Kerr, P.J. Teddy (Radcliffe Infirmary); M. Allen, T. Dale, R. Bradford, Dr. C. Collis, Prof. A.P. Dhillon, N.L. Dorward, D. Farraday-Browne, Dr. D.J. McLaughlin, R.S. Maurice-Williams, Dr. K. Pigott, B. Reynolds, C. Shah, C. Shieff, Dr. E.M. Wilson (Royal Free Hospital); F. Afshar, H.E. Ellamushi, Prof. P.M. Richardson, H.I. Sabin, J. Wadley (Royal London Hospital); Prof. M. Brada, D. Guerrero, Dr. F.H. Saran, D. Traish (Royal Marsden Hospital); Dr. S. Whitaker (Royal Surrey County Hospital); Dr. P.N. Plowman (St. Bartholomew's Hospital); Carole Bramwell, Prof. A. Bell, F. Johnston, H. Marsh, A. Martin, P.S. Minhas, A. Moore, S. Stapleton, Dr. S. Wilson (St. George's Hospital); Dr. R.P. Beaney (St. Thomas' Hospital).