Meta- and Pooled Analyses of GSTM1, GSTT1, GSTP1, and CYP1A1 Genotypes and Risk of Head and Neck Cancer

Sequence variation in genes coding for phase I and phase II enzymes, such as members of the cytochrome P450 (CYP) and glutathione S-transferase (GST) families may potentially alter individual susceptibility to cancer. Polymorphisms that confer a modest disease risk (relative risk <2) can be a substantial public health burden if they are common (1). A review on GSTM1 and GSTT1 deletion genotypes and head and neck cancer summarized the results of case–control studies as inconclusive (2). Of the 21 studies reviewed for the GSTM1 deletion genotype, 13 studies reported odds ratios (ORs) between 0.9 and 1.3, whereas 8 studies reported ORs between 1.4 and 3.9 (2). For the GSTT1 deletion genotype, eight studies reported ORs from 0.5 to 1.2, whereas six reported ORs from 1.4 to 2.6 (2). A meta-analysis that identified 25 studies on the GSTM1 null genotype and the risk of head and neck cancer reported a summary OR of 1.20 [95% confidence interval (95% CI), 1.08–1.33 (3)]. Because these carcinogen-metabolizing enzymes may be among numerous genes involved in the multistage pathway of cancer, they are expected to be modest to moderate risk factors that may be difficult to detect. However, even modest single gene effects on cancer risk are of biological and medical importance because of the possibility of identifying, under multigenic models, high-risk individuals for target prevention activities.

Tobacco smoke contains a range of different carcinogens, including polycyclic aromatic hydrocarbons, aromatic amines, and nitrosamines (4). The extent of exposure of the upper aerodigestive tract to carcinogens may depend on whether the carcinogen is activated by phase I enzymes and whether it is detoxified by phase II enzymes. An individual’s exposure to tobacco carcinogens may therefore be altered by sequence variation in genes coding for these enzymes.

The CYP1A1 gene codes for a phase I enzyme that activates tobacco procarcinogens, such as benzo[α]pyrene and aromatic amines, into their carcinogenic forms (5). An A→G base substitution at nucleotide 2455, which is strongly linked to 3801T>C in the 3′-flanking region, encodes for an amino acid replacement of isoleucine by valine at codon 462 and has been reported to be associated with increased enzyme activity (6, 7). The variant genotype is suggested to be harmful, possibly by increasing carcinogen activation and generating reactive oxygen species (8). Moreover, smokers with the CYP1A1 variant genotype may have elevated DNA adduct levels (9).

The GST family includes phase II enzymes that detoxify carcinogens and reactive oxygen species (10). Individuals who have homozygous deletions for the GSTM1 and GSTT1 gene have no GSTM1 and -T1 enzyme activity. Lack of these enzymes may potentially increase cancer susceptibility because of a decreased ability to detoxify carcinogens such as benzo[α]pyrene-7,8-diol epoxide, the activated form of benzo[α]pyrene. The missense substitution Ile105Val results from an A→G base substitution at nucleotide 313. The Val105 form of the GSTP1 enzyme may be 2–3 times less stable than the canonical Ile105 form (11) and may be associated with a higher level of DNA adducts (12).

A previous review on GSTM1, GSTT1, and head and neck cancer included journal articles written in English and published between 1993 and 2000 (2), whereas a published meta-analysis on GSTM1 included publications up to May 2001 (3). Because published reports on additional study populations not included in the review and meta-analysis are available, we conducted an updated meta-analysis of case–control studies evaluating the relationship between head and neck cancer and GSTM1, as well as GSTT1, GSTP1, and CYP1A1, to assess whether the available evidence supports these associations and to determine the sources of heterogeneity among the study results. In addition, we pooled the raw datasets from 11 case–control studies on the relationship between these genes, which encode carcinogen-metabolizing enzymes, and head and neck cancer to explore the main effect of the genes as well as gene–gene and gene–environment interactions.

A MEDLINE search was conducted for case–control studies reported up to August 2003 on GSTM1, GSTT1, GSTP1, and CYP1A1 and the risk of head and neck cancer, including oral, pharyngeal, and laryngeal cancers. We focused on the null alleles of the GSTM1 and GSTT1 genes, the Val105 allele of the GSTP1 gene, and the Val462 allele of the CYP1A1 gene. The following keywords were used in the Medline search: “Glutathione,” “GSTM1,” “GSTT1,” “GSTP1,” and “CYP1A1.” In addition, we reviewed the literature cited by each of the journal articles that we identified. When several articles were identified for the same population, we referred to the most updated information source. A total of 37 publications were identified, with 4635 cases and 5770 controls from 31 different populations (13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50).

We focused on studies that genotyped individuals by PCR and excluded ∼30 studies that assessed gene expression by measurement of protein levels. Positive controls for GSTM1 and GSTT1 genotyping were mentioned for the majority of studies. GSTP1 was genotyped in all studies by PCR combined with restriction fragment length polymorphism analysis. For CYP1A1 genotyping, three studies used allele-specific PCR (19, 28, 46), one study used PCR combined with single-strand conformational polymorphism analysis (42), and the other studies used PCR combined with restriction fragment length polymorphism analysis.

For each study, we abstracted the publication date, country where the study was conducted, site within the head and neck cancer studied, control source, numbers of cases and controls, and whether controls were matched to cases. Healthy subjects recruited from hospitals as controls were categorized as “hospital-healthy.”

We calculated for each study crude odds ratios (ORs) and 95% confidence intervals (95% CIs) for head and neck cancer when possible. For a study from France, we combined the data for cancers of the oral cavity, pharynx, and larynx that had been presented in separate publications (25, 26, 27). We abstracted crude ORs from the publications when they were available. When both the data and crude ORs were presented and there was a discrepancy between the two estimates (47), we retained the calculated crude OR because the OR presented in the original publication may have been adjusted for some factors. We did not perform a meta-analysis of the adjusted ORs because adjustment was not comparable among the studies. For GSTP1 and CYP1A1, we combined the heterozygous and homozygous genotypes because of the limited number of subjects who were homozygous mutant.

When possible we estimated or abstracted study-specific ORs separately by site within the head and neck (oral cavity, larynx) and by smoking status (never, ever smokers). For several studies, we were unable to separate cancer cases of the nasopharynx, nasal cavity, sinus, or salivary gland from other head and neck cancers (20, 32, 36, 40). The histology of head and neck cancer cases was squamous cell carcinoma in most studies, but two studies (23, 47) did not specify the histology, and one study (32) included 20 of 156 cases with “other miscellaneous histologies.”

Summary ORs were estimated with the statistical program STATA, version 8.0, by inverse-variance weighting, using a random-effects model that included a term for heterogeneity among studies (51). We estimated summary ORs when there were at least three risk estimates available. Thus, for some strata, summary ORs could not be estimated because of the small number of studies. Tests for heterogeneity among the studies were conducted for each analysis. Publication bias was assessed with the funnel plot of Begg and Mazumdar (52) and regression asymmetry test of Egger et al. (53).

We conducted influence analyses, in which each study was excluded one at a time to determine the magnitude of influence on the overall summary estimate. The influence analyses showed that the inferences for GSTM1, GSTT1, and GSTP1 did not change as a result of the exclusion of any one study. However, the summary estimate for CYP1A1 was statistically significant only when we included one specific study that reported a high OR relative to the other studies (28). We considered this study to be a possible outlier and thus excluded the study. Further, for GSTM1, we observed four studies of small sample size that had identified strong positive associations and led to asymmetry in the Begg funnel plot. We excluded these four studies in an attempt to minimize publication bias. Both summary estimates including all studies identified and excluding several studies are presented.

The data on GSTM1, GSTT1, GSTP1, CYP1A1, and head and neck cancers for the pooled analysis were extracted from the International Collaborative Study on Genetic Susceptibility to Environmental Carcinogens database (54, 55), which contains individual level data from case–control studies on genes that metabolize environmental carcinogens. Investigators who had published their results from case–control studies on genetic polymorphisms and cancers were identified through a MEDLINE search and requested to provide published and unpublished original data from their studies. Our data included 11 case–control studies, of which 9 had been published and included in the meta-analysis (13, 15, 16, 20, 25, 26, 27, 29, 30, 34, 35, 40, 42, 43, 44). Cases of cancer of the nasopharynx, maxillary sinus, and salivary glands were excluded from the analysis. Of the 2334 head and neck cancer cases included in the analysis, there were 840 oral cavity cancers, 501 pharyngeal cancers, 904 laryngeal cancers, and 79 cases with unspecified cancer within the head and neck, whereas the control group included 2766 subjects.

To assess the association of the genotypes with head and neck cancer, the logistic regression model was used to estimate study-specific ORs and 95% CIs. We estimated a crude OR and an OR adjusted for age, sex, and race for each study. ORs estimated for individual studies and numbers of cases and controls may not precisely match those reported in the publications. Heterogeneity among studies was assessed with the test for heterogeneity, whereas publication bias was assessed with the funnel plot of Begg and Mazumdar (52) and regression asymmetry test of Egger et al. (53). A summary OR was estimated by inverse-variance weighting with the random-effects model (51) because of the heterogeneity detected among studies. In the pooled analysis, we did not assess the effect of study characteristics because of the small number of studies available.

Study-specific crude ORs and overall summary ORs from the meta-analysis of the GSTM1, GSTT1, GSTP1, and CYP1A1 genotypes are shown in Table 1 . The distribution of genotypes at these four loci among the head and neck cancer cases and controls from the 31 populations are presented in Table 2. The frequencies of the genotypes varied among controls: 24.7–57.8% for the GSTM1 null genotype, 8.3–52.8% for GSTT1 null genotype, 23.8–64.6% for the GSTP1 valine genotype, and 5.8–39.5% for the CYP1A1 valine genotype. The percentage of Caucasians in the United States studies were as follows: 88.9% of cases and 87.9% of controls (14), 95.6% of cases and 93.3% of controls (36), 62% of cases and 86% of controls (39), 100% of cases and controls (42), 66.2% of cases and 67.3% of controls (43), 61.5% of cases and controls (44), or not specified (21, 50).

Of the 30 identified studies on GSTM1, 4 were excluded to minimize publication bias (15, 21, 31, 38). For the remaining 26 studies, the summary OR for the GSTM1 null genotype was modestly elevated (OR, 1.23; 95% CI, 1.06–1.42; Table 3). There appeared to be heterogeneity among the studies according to the test for heterogeneity. The ORs for the GSTM1 null genotype were higher for studies with smaller sample size than for larger studies (P = 0.07) and for studies from Asia relative to studies from Europe (P = 0.08). The summary OR for the GSTT1 null genotype and risk of head and neck cancer was 1.17 (95% CI, 0.98–1.40; Table 3) for 21 studies. None of the studies was excluded because there were no strong indications of excessive influence or publication bias. Geographic region may again be a source of heterogeneity in the GSTT1 studies (P = 0.03). For GSTP1, we estimated a summary OR of 1.10 (95% CI, 0.92–1.31) for carrying the Val105 allele, including nine case-control studies (Table 4). The risk of oral cancer may be higher than the risk of laryngeal cancer for the GSTP1 any valine genotype (P = 0.04). Twelve case–control studies were identified for the association between the CYP1A1 genotype and head and neck cancers. After the exclusion of one study suspected to be an outlier (28), the summary OR for carriage of the Val462 allele was 1.35 (95% CI, 0.95–1.82; Table 4). Differences between ORs for the CYP1A1 genotype in the stratified analysis were not identified.

Results of the pooled analysis of GSTM1, GSTT1, GSTP1, CYP1A1, and the risk of head and neck cancer are shown in Table 5. The race/ethnicity distribution for the Lazarus study was 64.3% Caucasian and 35.7% African American among cases, and 61.8% Caucasian and 38.2% African American among controls. For the Romkes study, the cases were 97.4% Caucasian and 2.6% African American, whereas the controls were 78.9% Caucasian, 15.5% African American, 4.2% Hispanic, and 1.4% Asian. Including 11 studies, the OR adjusted for study center was 1.32 (95% CI, 1.07–1.62) for the GSTM1 null genotype. When further adjusted for age, sex, and race, the OR was 1.18 (95% CI, 0.97–1.44; not presented). The risk of head and neck cancer was also elevated by the GSTT1 null genotype (OR adjusted for study center, age, sex, and race was 1.41; 95% CI, 1.00–1.57; not presented), but not by the GSTP1 and CYP1A1 missense substitutions, according to the pooled analysis. Differences among ORs in the stratified analysis were not observed in the pooled analysis.

Gene–gene interactions were assessed for all combinations of GSTM1, GSTT1, GSTP1, and CYP1A1 on the risk of head and neck cancer in the pooled analysis, but strong interactions were not identified. We also analyzed the data for possible gene–environment interactions between each genotype and smoking, but again interactions were not obvious. However, from the subset of studies that had genotype data on all three GST loci (906 cases and 543 controls; Refs. 26, 30, 31, 32, 37, 38, 42), we observed an increasing risk of head and neck cancer with inheritance of modest risk genotypes at increasing numbers of the GST loci that we studied. The null genotypes for GSTM1 and GSTT1 and carrying the Val105 allele of GSTP1 were considered likely to confer modestly increased risk. Taking the subjects with the genotype of GSTM1 present, GSTT1 present, and Ile/Ile for GSTP1 as the reference, the OR was 1.13 (95% CI, 0.83–1.53) for subjects who inherited a modest risk genotype at one GST locus, 1.19 (95% CI, 0.87–1.63) for subjects who inherited modest risk genotypes at two GST loci, and 1.69 (95% CI, 0.99–2.88) for subjects who carried likely modest risk genotypes at all three GST loci when adjusted for study center (test for trend, P = 0.08). When further adjusted for age, sex, and race, the OR was 1.16 (95% CI, 0.83–1.63) for one modest risk GST genotype, 1.23 (95% CI, 0.86–1.75) for two modest risk GST genotypes, and 2.06 (95% CI, 1.11- 3.81) for carrying three modest risk GST genotypes (test for trend, P = 0.04). When stratified by smoking, the results were not statistically significant.

The results from the meta-analysis supported the hypothesis that specific genotypes at the GSTM1, GSTT1, and CYP1A1 loci modestly increase the risk of head and neck cancer. Potential sources of heterogeneity included sample size and geographic region. The pooled analysis confirmed the association of head and neck cancer with GSTM1 and GSTT1, but the associations with GSTP1 or CYP1A1 missense substitutions were not clear. The pooled analysis was based on a subset of published studies from the meta-analysis that tended to report no associations or weak associations. Although pooling of the data provided increased statistical power to detect gene–environment interactions, we did not observe any strong interactions. One possible explanation for the lack of interaction may be that these gene–environment interactions are heterogeneous by ethnicity, in which case pooling data across different ethnicities may have diluted the interaction. A relationship was suggested, however, between genotypes at multiple GST loci and head and neck cancer risk.

Case–control studies with small sample size (<100 cases or 100 controls) may be reporting inflated ORs. These results suggest caution in the interpretation of small case–control studies. The summary ORs for GSTM1 and GSTT1 may also differ by geographic region. The prevalences of these genotypes in controls varied widely among and within regions. In the Indian population, the prevalence of the GSTM1 and GSTT1 null genotypes seemed to be particularly low. It will be of interest to further explore whether these genotypes are more relevant in specific ethnic groups, with respect to the risk of head and neck cancer.

Because we were unable to control for matching factors in the meta-analysis, we may have bias in our study-specific effect estimates. However, matching did not seem to be a source of heterogeneity among the studies, and individually matched studies did not have ORs that were biased toward the null, as might be expected, when compared with unmatched studies. Therefore, not controlling for matching factors may not be a strong limitation.

The modest association we observed between the risk of head and neck cancer and the CYP1A1 Val462 allele (OR, 1.35; 95% CI, 0.95–1.82) could reflect a possible association with the MspI variant allele because the CYP1A1 Val462 allele has been reported to be in strong linkage disequilibrium with the CYP1A1 MspI variant allele in Japanese (56) and Finnish populations (57). The CYP1A1 Ile462Val studies in our meta-analysis that had also examined the CYP1A1 MspI sequence variation in most cases showed similar association results for either marker (18, 34, 40, 46). One study that had presented data for a combination of these sequence variants (40) did not show any associations for either sequence variant alone or together, possibly because of the limited number of subjects who carried the sequence variants. Further studies with adequate sample size that examine combinations of CYP1A1 sequence variants will be helpful in clarifying their role on the risk of head and neck cancer.

Carcinogen metabolism is complex, involving the interaction of numerous carcinogens and enzymes. The GSTs have a variety of substrates, including environmental carcinogens, pesticides, drugs, and endogenous molecules of lipid peroxidation as well as inducing agents, some of which double as substrates, including polycyclic aromatic hydrocarbons, phenolic antioxidants, isothiocyanates, and reactive oxygen species (58). The metabolic action of GST enzymes may differ by cancer site; the highest concentrations of GSTP1 have been observed in oral and pharyngeal tissues, and the highest concentrations of GSTM1 have been observed in laryngeal tissue, relative to the other GSTs (2). GST enzyme expression may also differ according to the general controls of gene expression, such as the rates of transcription, translation, and degradation as well as possible posttranslational modifications.

Individually, sequence variants in carcinogen-metabolizing genes may be modest to moderate risk factors, explaining the inconsistent results seen in epidemiological studies. This meta-analysis supports the hypothesis that genotypes at the GSTM1, GSTT1, and CYP1A1 loci are modest risk factors for head and neck cancer. On the other hand, combinations of genotypes that each confer a small relative risk may add up to a relative risk large enough to be observed in epidemiological studies. Our pooled analysis supported the idea that inheritance of multiple modest risk GST genotypes may confer a greater risk of head and neck cancer. Future epidemiological studies focusing on complex genotypes within the same gene family or other related gene families may be helpful in identifying individuals at high risk for head and neck cancers and in elucidating gene–gene interactions.

W. D. F. thanks Nancy Hamel for assistance in compiling data for this publication.