Tobacco Documents Online

1997, 8, pp. 544-555 A case-control study of cytochrome P450 1A1, glutathione S-transferase M1, cigarette smoking and lung cancer susceptibility (Massachusetts, United States) Montserrat Garcla-Closas, Karl T. Kelsey, John K. Wiencke, Xiping Xu, John C. Wain, and David C. Christiani (Received 14 August 1996; accepted in revised form 22January 1997) Cytochrome P450 1A1 (CYP1A1) and glutathione S-transferase M1 (GSTM1) genetic polymorphisms are involved in the activation and detoxiflcation of chemical carcinogens found in tobacco smoke; thus they may influence host susceptibility to lung cancer. In this study at Massachusetts General Hospital (Boston, MA, USA) of 416 cases and 446 controls (mostly White) we evaluated the assochtion between the CYP1A1 MspI and GSTM1 polymorphisms and lung cancer risk, and their interaction with cigarette smoke. The CY~PIA1 MspI heterozygous genotype was present in 18 percent of cases and 16 percent of controls, and one percent of cases and controls were CYP1A1 MspI homozygous variant. The GSTM1 null genotype was detected in 54 .percent of cases and 52 percent of controls. After adjusting for age, gender, pack-years of smoking, and years since quitting smoking, while ndther the CYPIA1 MspI heterozygous genotype alone nor the GSTM1 null ge/~otype alone were associated with a significant increase in lung cancer risk, having both genetic traits was associated with a twofold increase in risk (95 percent confidence interval [CI] =. 1.0-3.4). Our data did not provide enough evidence for a substantial modification of the effect of pack-years on lung cancer risk by the CYPIA1 MspI and GSTM1 genotypes. However, limitations of our study preclude a conclusion about this potential interaction. Cancer Causes and Control 1997, 8, 544-553 Key words: Cytochrome P450, genotype, glutathione transferases, lung neoplasm, smoking, United States. Introduction Most carcinogens in tobacco smoke require metabolic activation by phase I enzymes in order to manifest their carcinogenic effects. The activated intermediates resulting from these reactions then can be detoxified through reactions catalyzed by phase II enzymes.I"~ Therefore, genetically determined differences in metabolizing Authors are with the Division of Cancer Epidemiology and Genetics, National Cancer Institute, Be~esda, MD, USA (Dr Garcia-Closas); Department of Environmental Health, Occupational Health Program, Harvard School of Public Health, Boston, MA (Drs Kelsey, Xu, Christiani); Department of Cancer Biology, Harvard School of Public Health, Boston, MA (Dr Kelsey); Laboratory for Molecular Epidemiology, Department of Epidemiology and Biostatistics, University of California at San Francisco, CA, USA (Dr Wiencke); Thoracic Surgery Unig Massachusetts General Hospital Harvard Medical School Boston, MA (Dr Wain); Pulmonary and Critical Care Unit, Massachusetts General Hospital Harvard Medical School Boston, MA (Dr Christiani). Work was performed at the Departments of Epidemiology and Environmental Health, Harvard School of Public Health, Boston, MA. Address correspondence to Dr ChristianL Harvard School of Public Health, 665 Huntington Ave., Boston, MA 02115, USA. This study was supported by Grant ES/CA P01 06409 and ESO0002 from the US National Institute of Environmental Health Sciences and the National Cancer Institute. Dr Garcia-Closas was supported by a scholarship from the Real Colegio Complutense, Cambridge, MA. 0 544 Cancer Causes and Control. Vol 8. 1997 ~) 1997 Rapid Science Publishers

enzymes involved in these reactions could influence host susceptibility to lung cancer, as well as modify the rela- tionship between cigarette smoking and lung cancer risk.~ Cytochrome P450 IA1 and glutathione S-transferase M1 (GSTM1) are, respectively, a phase I and phase II enzyme involved in the metabolic activation and detoxi- fication of polyaromadc hydrocarbons (PAH) and other combustion products found in cigarette smoke.~ These two enzymes have been found to be polymorphic in human populations. There are two described restriction fragment-length polymorphisms (RFLP) of the gene coding for P450 1Ah the Msplr *and the Ile/Val~ poly- morphisms. While the relationship between these two closely linked polymorphisms and the enzyme activity remains to be established,~'~s the homozygous variant genotype of both polymorphisms has been associated with an increased risk of lung cancer in Japanese popu- lations.'n In contrast, this association has not been confirmed in studies conducted in Caucasian (White) popu- lations.~*s The polymorphism in-the GSTM1 gene is due to a complete deletion of the gene which results in a lack of enzyme activity.'~'1. Several epiderniologic studies*'2. have evaluated the association between lung cancer and GSTM1 null genotype (homozygous deletion) obtaining contradic- tory results. Overall these studies indicate that subjects with the GSTM1 null genotype may have a 10 to 60 percent average increase in lung cancer risk. This increase in risk has been stronger and found more consistentlyinJapanese than in Caucasian populations. Several studies have evaluated the potential modifica- tion of the association between these polymorphisms and !ung cancer by cigarette smoke.'I''~'.4a~'~ However, results from these studies have been contradictory and the question of whether these genes have stronger asso- ciations with lung cancer at high or low levels o~ exposure, or whether the associations are the same across levels of exposure remains controversial. The small number of subjects included in most of these studies, the use of different methods of assessment and categorization of cumulative dose, and differences in ethnicity are possible explanations for the apparendy contradictory results. Some studies also have evaluated the potential inter- action between the CYP1A1 and GSTM1 genotypes. ~apanese studies have reported that subjects with the combined GSTM1 null genotype and CYP1A1 MspI or Ile/Val polymorphisms are at remarkably high risk of lung cancer, especially at low cumulative doses of cigarette smoke)° Two subsequent Scandinavian studies~a* sug- gested an interaction between the GSTM1 null genotype and the CYP1A1 MspI variant allele or the aryl hydro- carbon hydroxylase (AHH) inducible phenotype, however these studies were too small to be conclusive. We recently reported'z a posidve association between the CYP1A1 MspI allele and lung cancer risk, which was GSTMI Folyrrtorphisra and lung cancer only evident after adjusting for cumulative smoking dose. Specifically, in a case-control study of 207 lung cancer cases and 283 controls from a North American (mostly White) population, the combined homozygous/hetero- zygous MspI variant of the CYP1A1 gene was associated with ~t twofold increase in lung cancer risk (95 percent confidence interval [CI] =* 12.-3.7). We have now expanded our study to include 416 cases and 446 controls, providing us with one of the largest studies conducted to date. In this expanded population we have evaluated the association between the CYP1A1 MsivI and GSTM1 polymorphisms and lung cancer risk, as well as the modification of the effect of cigarette consumption by these polymorphisms. Materials and methods The present study is a case-control study of incident cases of lung cancer which was conducted at the Massachusetts General Hospital (MGH) in Boston, MA, United States. Eligible cases included all patients with newly diagnosed primary lung cancer (stages I and II) presenting for thoracic surgery at the MGH between December 1992 and August 1996, and who had a confirmed diagnosis after histologic examination of post-surgery tumor samples. Lung cancer padents were referred by physicians from various services of the MGH as well as from physicians outside MGH, and the vast majority (- 90 percent) lived in the greater Boston area. During the study period, we identified 572 eligible patients, most of whom were White. Among these, 42 refused to participate, 34 were missed, 26 were unable to participate, and 11 agreed to participate but did not complete the questionnaires. This left 459 lung cancer cases with a participation rate of 80 percent. Controls (n = 458) were selected among friends or spouses of lung cancer patients (n = 167), and among friends or spouses of cardiac surgery patients (n = 83) or other thoracic surgery patients (n = 208), all from MGH. A formal record of the participation rate among controls was not kept during the most of the study period. However, based on an informal record of potential candidates identified by the hospital nurse, our best estimate of the participa- tion rate among controls is about 90 percent. Data collection ~06~ 68 ~2 EO Stage and histologic type of lung cancer was assessed by histologic examination of post-surgery tumor samples. Information on demographic characteristics (gender; age; level of education; race; and family history of cancer in parents, siblings, or children) was collected through ques- tionnaires administered to cases and controls by trained interviewers at the hospital. Smoking and other occupa- tional/environmental exposure history was collected through a modified standardized American Thoracic C~ncer Cause~ and Control Vol 8. 1997 545

Af. G=rci~-Cloms et =l Society respiratory questionnaire.2" Thi~ questionnaire included information on: current smoking stares; age when started smoking; whether they temporarily quit smoking for six months or more during the period they smoked; age when quit smoking; and the average amount of ciga= rettes smoked per day, period, and frequency of exposure to other environmental or occupational substances such as asbestos and solvents. Levels of food intake were assessed through a semiquantitative food=frequency ques- tionnalre with 126 food items.~ Details on the assessment of dietary intake in the population under study have been previously described.~° Questions about current habits were referred to time of diagnosis for cases and time of interview for controls. Blood samples were obtained in the hospital from cases and controls and sent to the molecular biology laboratory for genotype determination. DNA was obtained from heparinized whole blood by use of Chelex® solution as described by Walsh et,;/. ~; Genotyping for CYPIAI MspI and GSTM1 wa~ completed using PCR-based methods previously published,v°~3 Data analysis Odds ratios (OR) and CIs were used to quandfy the association between lung cancer risk and genotype. Generalized additive models were used to assess non- linear associations of continuous covariates on the logit of the probability of disease?* The following potential confounders were considered in the analysis: gender (male, female); race (White and others); level of education (less than high school, high school graduate, 1-4 years of college, and college graduate); smoking status (never, current or quit within the last year, former); current age; age at onset of smoking; years since quit smoking; average duration of smoking; average cigarettes per day; pack- years smoked; occupational exposure to asbestos (yes, no); family history of lung cancer among parents or siblings (yes, no); and dietary intake of antioxidant vitamins (vitamin C, vitamin E, alpha-carotene and beta- carotene). Among these potential confounders, only age, gender, dose and duration of cigarette smoking, and years since quit smoking were included in our final model (see Results section). Generalized addidve models indicated the presence of a non-linear association on the logit of the probability of disease for age and pack-years, and a linear association for years since quit smoking. The functional form for age and pack=years was approximated by second=degree polynomials. Therefore, the adjusted genotype ORs were obralned from parametric models which included age and pack-years as second=degree polynomials, a linear term for years since quit smoking, and an indicator term for gender. The genotype ORs were practically unchanged by alternative ways of modeling dose and duration of 546 Cancer C~uses a~d Control. Vol 8. 1997 cigarette smoking, e.g., including dose and duration as two separate terms, or including pack-years only. Statis- tical analysis were performed with the statistical software STATA version 5.0 and S-Plus version 3.3 for Windows. ~$U~ The study population included 459 cases of lung cancer and 458 controls. Five controls and 37 cases were excluded because we were unable to obtain a blood sample for genotype analysis. Among the remaining subjects, six cases and seven controls were excluded because of incom- plete smoking data, leaving 416 cases and 446 controls available for the main analysis. The distribution of characteristics of cases and controls with complete genotype and smoking information (n = 917) is displayed in Table 1. Approximately, 97 percent of cases and 98 percent of controls were Whites. Respec- tively, cases and controls had a percentage of females of 46 percent and 53 percent, and a median age of 67 years (33 to 89 range) and 64 (27 to 84 range) years. Of the cases, current smokers accounted for 41 percent, former smokers (defined by having quit smoking at least one year before em'ollment) for 54 percent, and never-smokers for five percent. Of the controls, current smokers accounted for 17 percent, former smokers for 52 percent, and never- smokers for 31 percent. The most common histologic type was adenocarcinoma (54 percent),, followed by squamous cell carcinoma (28 percent), large cell carcinoma (five percent), small cell carcinoma (four percent), and others including mixed cell types (eight percent). Association between CYP1A1 MspI and GSTM1 genotypes and lung cancer The distribution of the CYP1A1 Msp[ and GSTM1 geno- types was very similar for cases and controls (Table 1). The CYP1A1 MspI homozygous variant genotype was present only in four controls and four cases. Since past studies have shown different magnitude of associations for homozygous variant and heterozygous subjects9';° these eight subjects were excluded from further analyses.. A positive association between the CYP1A1 MspI polymorphism and lung cancer risk has been reported previously in a subset of the population under study,a~ This association was only present after adjusting for ciga- rette smoking. In the current population, the CYP1A1 Ms?I heterozygous genotype was associated with a 50 percent increase in risk of lung cancer (CI = 1.0=2.3) (Table 2). As in our pervious report, pack-years was the main variable responsible for the differences between the crude and adjusted estimates. This reflected a negative association between the average cigarettes smoked per day and the CYP1A1 Ms?I heterozygous genotype among the control group (23 ± 15 and 17 ± 11 [mean ±

GSTM1 ?olymo~hisra and lung cancer Table 1. Distribution of selected variables among lung cancer patients and controls, Massachusetts General Hospital Characteristics Cases Controls (n = 416) (n = 446) Age in years, median (range) Gender, n (%) females Race, n (%) White Educationa n (%) College graduate Smoking status, n (%) Never-smokers Former smokers Current smokers Duration of smoking in years, mean (SD) Former smokers Current smokers Average cigarettes per day, mean (SD) Former smokers Current smokers Years since quit smoking, mean (SD) Passive exposure to cigarette smoke at home or workb n (%) positive Occupational exposure to asbestos,¢ n (%) positive CYPIA1 Msp/ n (%) homozygous wild-type heterozygous homozygous variant n (%) n (%) GSTM1 n (%) homozygous wild-type or heterozygous n (%) homozygous deleted Family history of lung cancer,d n (%) positive Histologic typee Adenocaminoma, n (%) Squamous cell caminoma, n (%) Small cell caminoma, n (%) Large cell caminoma, n (%) Other tumors,f n (%) 67 (83-89) 64 (27-84) 193 (46.4) 237 (53.1) 403 (96.9) 436 (97.8) 86 (21.3) 96 (21.8) 21 (5.0) 139 (31.2) 226 (54.3) 233 (52.2) 169 (40.6) 74 (16.6) 36 (12) 24 (12) 44 (11) 38 (11) 30 (16) 22 (15) 29 (15) 21 (11) 13 (10) 20 (12) 18 (94.7) 127 (92.7) 60 (14.6) 43 (9.6) 337 (81.0) 369 (82.7) 75 (18.0) 73 (16.3) 4 (1.0) 4 (1.0) 190 (45.7) 214 (48.0) 226 (54.3) 232 (52.0) 81 (22.8) 58 (15.1) 225 (54.3) 116 (28.1) 18 (4.3) 23 (5.6) 32 (7.7) 0 Information was missing for 13 cases and 5 controls. Among never-smokers only. Information was missing for 2 cases and 2 controls. Information was missing for 5 cases. Information was missing for 61 cases and 61 controls. Information was missing for 2 cases. Includes 14 cases with mixed tumors and 18 cases with more than one tumor. SD = standard deviation. standard deviation] cigarettes per day for ml/ml and ml/m2 respectively; Wilcoxon Sum Rank Test: P : 0.01). In contrast to the MslH genotype, our data did not show evidence for an association between the GSTM1 null genotype and lung cancer risk (Table 2). When different histologic types of lung cancer were evaluated separatelB no relevant differences in risk were observed for either of the genotypes (data not shown). Models restricted to Whites or including dummy variables for different levels of education did not show important changes in the estimated genotype ORs. We also evaluated the confounding effects of family history of lung cancer, dietary intake of antioxidant vitamins and occupational exposure to asbestos among the subset of subjects with complete information on these variables (n = 684). Since we did not find evidence for important confounding by either of these variables and given that several subjects had missing information, we did not in- clude them in the finat adjusted model. Interaction between CYP1A1 MspI, GSTM1, and cigarette smoke Table 3 shows results from a stratified analysis performed to evaluate the modification of the asso6iation between Cancer Causes and Control. Vol 8. 1997 547

M. Garc~-Clos~s ct ~1 Table 2. Association between lung cancer risk, CYPIA1 Mspl and GSTMI polymorphisms Cases Controls Crude Adjustedb OR (Cl)a OR (CI)" CYPIA1 Mspl c ml/ml 337 369 1.0 -- 1.0 -- ml/m2 75 73 1.1 (0.8-1.6) 1.5 (1.0-2.3) GSTM1 d Present 190 214 1.0 -- 1.0 ~ Null 226 232 1.1 (0.8-1.4) 1.0 (0.7-1.4) OR = odds ratio; (CI) = 95 percent confidence interval. Odds ratios adiusted for age, gender, pack-years, years since quit smoking, and the other genotype. ml/ml = homozygous wild-type; ml/m2 = heterozygous. Homozygous variant subjects (m2/m2) were excluded from this analysis (n = 8). GSTM1 null genotype refers to homozygous deletion. Table 3. Association between lifetime cumulative cigarette dose and lung cancer risk stratified by the CYPIA1 Mspl and GSTM1 genotypes= Pack-years of smoking Caaes Controls Adjustedb OR (Cl)a All subjects <20 46 119 1.0 ~ 21-40 76 102 1.2 (0.7-1.9) 41-60 109 50 3.0 (1.7-5.3) • 60 161 36 5.8 (3.1-10.6) CYPIA1 Mspl genotypec'd rnl/ml < 20 37 91 1.0 -- 21-40 60 90 1.0 (0.5-1.7) 41-60 90 44 2.5 (1.4-4.7) > 60 135 31 5.4 (2.8-10.4) ml/m2 < 20 8 27 0.8 (0.3-1.9) 21-40 19 12 2.1 (0.9-5.0) 41-60 19 4 5.2 (1.5-17.5) • 60 23 5 5.1 (1.7-15.6) GSTM1 genotypee'f Present ~ 20 19 56 1.0 -- 21-40 31 49 1.1 (0.5-2.3) 41-60 56 23 3.9 (1.7-8.6) • 60 73 11 9.2 (3.7-22.8) Null ~ 20 27 63 1.3 (0.6-2.7) 21-40 48 53 1.6 (0.8-3.2) 41-60 53 27 3.1 (1.4-6.8) • 60 88 25 5.2 (2.5-12.2) Analysis limited to cigarette smokers (395 cases and 307 controls). Odds ratios (OR) and 95% confidence intervals (CI) adjusted for the other genotype, age, gender, pack-years, and years since quit smoking. ml/ml = homozygous wild-type; ml/m2 heterozygous. Homozygous vadant subjects (m2/m2) were excluded from this analysis (n = 8). Likelihood ratio test for homogeneity: X,2(3)= 3.8, P= 0.3. GSTMI null genotype refers to homozygous deletion. Likelihood ratio test for homogeneity: X,2(3) = 4.1, P= 0.3. 548 Cancer Causes and Control. Vol'8. 1997

GSTMI polyrnorphism and lung cancer cumulative smoking dose and lung cancer risk by the CYP1A1 Mspl and the GSTM1 polymorphisms. This analysis was restricted to current and former smokers. The genotype ORs among never-smokers were very similar to the ORs for all subjects reported in Table 3 (OR = 1.8, CI = 0.7-5.0 for CYP1A1 MspI hetcrozygous genotype, and OR = 1.1, CI -- 0.4-2.6 for GSTM1 null genotTpc). Stratification by CYPIAI MspI suggested a stronger effect of smoking among subjects with the Msp/" heterozygous genotype than among MspI wild-type sub- jects, however, CIs were too wide.to enable any inference (likelihood ratio test [LRT] for homogeneity: Z~(3) = 3.8, P = 0.3). Stratification by GSTM1 genotype suggested a weaker effect of pack-years for subjects with the null genotype than for subjects with GSTM1 present. However, as in the case of the CYP1A1 genotype the test for homogeneity indicated that this difference is likely to be explained by chance (LRT for homogeneity: Z~0) = 4.1, P = 0.3). In order to improve the power to study these potential interactions and to avoid the dependence of results on arbitrary cut-off points, we also modeled pack-years as a continuous variable. Generalized additive models indicated that the relationship between the logit of the probability of disease and pack-years for the different genotype gro.ups was well-approximated by second- degree polynomials. Therefore, we present results from parametric models which included second-degree polynomials to describe the effect of pack-years by the different genotypes. Figures 1 and 2 represent the relationship between pack-years and the logit of the probability of disease after stratification by the CYP1A1 MspI and GSTM1 geno- types. Figure 1 shows an about 50 percent increase in risk of lung cancer for subjects with the MspI heterozygous genotype, as seen in Table 2, which does not seem to change across levels of cumulative cigarette dose (LRT for interaction: Deviance =-0.02, P = 1.0). Thus, the association between cumulative smoking dose and lung cancer does not appear to be modified by the CYPIAI MspI genotype. According to Figure 2, GSTM1 null subjects seem to have an increased risk of lung cancer at low to moderate doses of smoking. However, as the number of pack-years increases, the lung cancer risk increases at a lower rate among subjects with the GSTM1 null genotype than among subjects with GSTM1 present; thus, GSTM1 null subjects appear to have a decreased risk of lung cancer at high doses of smoking. This pattern of interaction agrees with the pattern observed in the stratified analysis (Table 4), however according to the test for interaction we cannot reject the null hypothesis that the effect of smoking is the same across GSTM1 genotypes (LRT for interaction: Deviance = -4.1, P = 0.1). Interaction between the GSTM1 and CYP1AI Msp[ polyrao~hisras When compared with subjects with at least one GSTM1 allele and with the CYPIA1 MspI homozygous wild-type genotype, neither the CYPIA1 MspI heterozygous geno- type nor the GSTM1 null genotype was associated with an increase in lung cancer risk; however, having both traits was associated with an approximately twofold increase in risk (CI = 1.0-3.4) (Table 4). Differences between the crude and adjusted estimates shown in Table 4 were due Figure 1. Fitted values for the relationship between lung cancer risk and pack-years stratified by the CYPIAI Msp/polymor- phism, from a logistic regression model which included pack-years as a second-degree polynomial and terms for age, gender, years since quitting smoking, and GSTM1 genotype. Figure 2. Fitted values for the relationship between lung cancer risk and pack-years strs.tified by the GSTM1 polymorphism, from a logistic regression model which included pack-years as a second-degree polynomial and terms for age, gender, years since quitting smoking, and CYPIA1 Mspl genotype. 3.0 34.3 Cancer C~uses and Control. Vol 8. 1997 549

M. Garcfa-Closas et al Table 4. Association between CYPIA1 Mspl and GSTM1 polymorphisms and lung cancer risk Cases Controls Crude Ad}ustedb OR (Cl)a OR (Cl)" CYPIA1 Mspl ml/ml¢ GSTM1 present 174 166 1.0 ~ 1.0 w GSTM1 null 195 182 1.1 (0.8-1.4) 0.9 (0.6-1.3) CYPIA1 Mspl ml/m2c GSTM1 present 38 32 0.9 (0.6-1.6) 1.0 (0.5-1.9) GSTM1 null 35 43 1.4 (0.8-2.3) 1.9 (1.0-3.4) Total 442 412 a OR = odds ratio; (CI) = 95 percent confidence interval. b Odds ratios adjusted for age, gender, pack-years, and years since quit smoking. c ml/ml = homozygous wild-type; ml/m2 = heterozygous. Homozygous variant subjects (m2/m2) were excluded from this analysis (n = 8). Likelihood ratio test for homogeneity: 7,2(1) = 1.7. P= 0.1. to the inclusion of pack-years in the adjusted modal. This reflected the association between CYPIA1 MspI geno- type and cigarettes smoked per day reported in the previous section. No important differences in the interaction between the GSTM1 and CYP1A1 MspI genotypes were observed across different histologic types. The adjusted OR for the interaction term between the two genotypes was 2.1 (CI = 0.9-4-.0, P = 0.1) for all tumors combined, 2.3 (CI = 0.8-6.0, P = 0.1) for adenocarcinoma and 2.2 (CI = 0.5- 10.0, P--0.3) for squamous cell carcinoma. Thus, although there is an indication of a positive multiplicative interaction between GSTMI and CYPIA1 MspI poly- morphisms, our data do not provide enough evidence to reject the null hypothesis of no multiplicative interaction at the five percent level of significance. Since only the combination of GSTMI null and CYP1A1 MspI heterozygous genotype seems to be associated with lung cancer risk, we compared the risk of subjects with this genotype combination with the risk of subjects with any other genotype combination, namely GSTM1 presenff CYP1A1 MspI homozygous wild-type, GSTM1 present/ C YP1A1 Msp I heterozygous or GSTM1 nuI1/CYP1A1 MspI homozygous. The adjusted OR for this comparison was 2.0 (CI = 1.1-3.5) for all tumors combined, 1.8 (CI = 0.9-3.4) for adenocarcinoma, and 1.8 (CI = 0.7-4..5) for squamous cell carcinoma. Our data suggested a negative interaction between pack-years and the GSTM1 genotype. Therefore we attempted to explore a potential three-way interaction between the two genotypes and pack-years. Given the limited power of our study to assess a three-way inter- action, we constructed a reduced model to explore the modification of the OR for the combined genotype by two levels of cumulative smoking dose using 40 pack- years as the cut-off point (data not shown). This reduced ~50 Cancer Causes and Control Vol. 8, 1997 model provided no evidence of a substantial modification of the pack-years effect by the combined genotype (LRT for homogeneity: Z2(1) = 0.1, P = 0.8). Discussion In this North American, mostly White, population we did not find evidence for an overall association between the GSTM1 null genotype and lung cancer risk. However, our data suggested that the GSTM1 deletion may contribute to lung cancer risk in combination with the CYP1A1 MspI heterozygous variant genotype. The frequency of the GSTMI null genotype among White controls (227/436 = 52 percent) was very similar to the only previously published study in North American Whites (244/465 = 52 percent)~° and it was within the range of frequencies found in other White populations: 53 percent (175/329) in a Swedish population,~53 percent (82/155) in a German population,~s 44 percent (62/142) in a Finnish population,~ 42 percent (94/225) in a English population,17 and 46 percent (67/147) in a northwestern Mediterranean population.:I These variations might reflect differences in ethnicity within Caucasian popula- tions.~' However, they also could reflect random variations due to the small samples in many of these studies. As in our previous report,v the CYP1A1 MspI heterozygous genotype was associated with an increased risk of lung cancer, which was evident only after adjusting for pack-years (OR = 1.5, CI = 1.0-2.3). This association did not appear to be modified by the level of cigarette smoking. The crude OR for the GSTM1 null genotype (OR = 1.1, CI = 0.8-1.4) was similar to the estimate obtained in a recent metanalysfs of past studies in Whites (OR--1.2, CI = 1.0-1.4)'9 and to the crude estimate observed by London et alz° among North American

Whites (OR = 1.0, CI = 0.7-1.3, calculated from Table 2 of the paper). We did not find important differences be- tween adenocarcinoma and squamous cell carcinoma. This finding agrees with a recent recta-analysis*' and the London et al study.'= Overall, the London et al study and the present study indicate that the GSTM1 null genotype is not associated with a substantive increase in lung cancer risk in North American White populations. Several epidemiologic studies which have evaluated the interaction between GSTM1 and cumulative cigarette smoking have obtained contradictory results. Out of the six studies which evaluated this interaction, three"~( found a positive association between GSTM1 and lung cancer only among smokers with high cumulative cigarette dose, two"'~° found this association only among smokers of low cumulative dose, and one study~ found no evidence for differences in two categories of cumulative smoking dose. A stronger association between GSTM1 null geno- type and lung cancer for heavier than lighter smokers is also supported by a study~7 which found an association between high sister chromatic exchange (SCE) levels and the GSTM1 deficiency among subjects with high cotinine levels but not among subjects with lower cotinine levels. Our data suggested a weaker effect of pack-years for GSTM1 null subjects than for subjects with the GSTM1 present, which implied a positive association between GSTM1 null geno .twpe and lung cancer for low to mod- erate doses of cigarette smoke, and a negative association for high doses of smoking. However, given the biological implausibility of a cross-over effect, chance seems a reasonable explanation for the suggested pattern of inter- action. When we stratified subjects according to both the CYP1A1 MspI and GSTM1 genotypes, while neither the GSTMI null genotype alone nor the CYPIA1 MspI heterozygous genotype ~lone were associated with a significant increase in lung cancer risk, having both genetic traits was associated with an approximately twofold increase in risk (CI = 1.0-3.4). This multiplicative inter- action was similar for adenocarcinoma and squamous cell carcinoma cases. A similar interaction was reported in a Swedish study of all tumor types-'s but not in two Japanese studies of squamous cell and small cell carcinomasnos which found a positive multiplicative interaction with the CYP1A1 MspI homozygous variant genotype but not with the heterozygous genotype. Several potential biases might have influenced our results. First, our control population includes friends or spouses referred by cases, by cardiac patients, or by thoracic patients. A necessary, condition for the validity ,>f our estimates of the genotype association is that, con- ditional on the variables controlled in the analysis, (i) the genotype distribution of the three controls series is the same, and (ii) represents the distribution in the source GSTM1 polymorphisrn and lung cancer population. To evaluate whether (i) is satisfied, we tested for differences in the genotype frequencies of the three control groups conditional on the variables included in our final model, finding no significant differences (g2(2) = 4.4, P = 0.1 for GSTM1 genotype; and g2(2) = 4.9, P = 0.1 for CYP1A1 Mslvl genotype). Given that we found no significant differences in genotype frequencies and that the frequencies in the collapsed control group were similar to the estimates observed in past studies of White popu- lations, it is seems acceptable to assume that the three control groups represent a common population. This still leaves us with the possibility that the three control groups, while originating from the same popula- tion, might not represent the source population of the cases. Selecting controls among friends or spouses of lung cancer patients or other surgical patients might introduce. bias by two different mechanisms: (i) the distribution of' exposures among'gregarious subjects,'~'/. e., subjects who tend to be named by more than one other person as controls, might be different than among 'non-gregarious subjects' and, therefore, it might not represent the distri- bution in the source populati.'on which includes both types of subjects;~ (ii) exposures among friends and spouses of patients might be similar to the exposures among the patients themselves, and thus it might not represent the exposure distribution in the source population. These two factors are likely to be relevant for exposures such as age, gender, smoking, or dietary habits. However, within cate- gories of these variables and among the same racial or ethnic group, metabolic polymorphisms are unlikely to be associated with being a gregarious individual or with having a friend or spouse with lung cancer. Therefore, the genotype-disease association is unlikely to be affected substantially by this type of bias. On the other hand, the reported estimates for the effect of pack-years on lung cancer are likely to be underestimated. Second, our case-population includes only surgical lung cancer patients (stages I and II), and more advanced tumors not eligible for surgery are not included. Thus, differential selection of cases with respect to the genotype could occur if the genotype is related to the stage at which the lung cancer is detected. Although to our knowledge this has not been evaluated for GSTM1 in past studies, two Japanese studies suggested a higher frequency of the CYP1A1 Msplvariant allele for poorly differentiated than for well-differentiated adenocarcinomas~ and for lung cancer patients with metastasis than for patients without metastasis.'* These differences could have introduced underestimation of the genotype ORs because poorly differentiated tumors and tumors with metastasis are more likely to become inoperable. Third, ethnicity, which was not measured in this study, could be a confounder of the ORs for the genotype- disease association. However, in the context of our study, Cancer Causes and Control. Vol 8. 1997 551

this type of confounding is unlikely to be important since in this almost exclusively White population, ethnicity is unlikely to be strongly associated either with the geno- type, as indicated above, or with lung cancer risk. On the basis of this discussion, we conclude that among Whites, the combination of the CYP1A1 MspI heterozy- gous genotype and the GSTM1 null genotype is likely to be associated with an increased risk of lung cancer. This indicates that host susceptibility to lung cancer may depend on the metabolic balance between the P450 1A1 and the GSTM1 enzyme activities. Our data did not provide with enough evidence for a modification of the effect of pack-years on lung cancer risk by the CYP1A1 MspI and GSTM1 genotypes, however this result is tentative in view of the limited statistical power to assess comprehensibly a potential three-way interaction be- tween CYP1A1 MspI, GSTM1, and lung cancer risk. Acknowledgements- The authors thank the follow- ing members of the lung cancer susceptibility project: Ms Linda Lineback for patient recruitment; Ms Lucy Ann Principe-Hasan and Mr Nick Weidemann for data entry and management; Ms Marcia Chertok for project coordination; Ms Marlys Rogers and Ms Lucille Poth- ier for computer programming; Ms Zheng-fa Zuo for PCR analysis; Ms Debby Strahs for sample handling and preparation; and Drs H. Grillo, Douglas Mathie- sen, and H. Kazemi of Massachusetts General Hospital. We also thank Drs David Hunter, Richard Monson, James Robins, and Dimitrios Trichopoulos for their critical review of the manuscript. Re{erences 1. Amdur ~0, Doull ~, I~aassen CD. To~c~'olog.y. ~[ew ~or~ ~ (USA): McG~v-H~, Inc., 1991. 2. Ke~erer B, H~s ~, TM~ka G, et aL The h~ glu- ta~one S-tr~sfer~e supergene fa~y, i~ pol~o~sm, ~d its effect on susceptibili~ to lung cancer. Environ Health P~p 1992; 98: 87-94. 3. ~ub~ FF, H~mons GJ. Role of c~oc~ome P450 h me~bolism of che~ c~chogem. ~: Guengefich F~ ed. Mammal~n Cyrochromes P~50. Volume 2. Boca ~ton FL (USA): CRC Pros, 1987: 81-130. 4. Peterson DD, McKi~ey CE, Ikeya K, e~ aL Hum~ ~IA1 gene: Cose~egafion of &e e~yme inducibiliw pheno~ ~d a ~LP. AmJHum Genet 1991; 48: 720-5. 5. Haym~ S, Wa~abe J, Nak~ ~ Kawaj~ ~ Genetic liege of lung c~cer msociat~ MspI polymo~hisms ~& ~ho acid replacement in the hem¢ bind~g re,on of the h~an c~oc~ome P450 1A1 gene.J Biochem 1991; 110: ~7-11. 6. Cosma G, Cro~ F, T~o~ E, To~olo ~ G~e S. Relation- s~p be~een geno~pe ~d hnc~oa of the human C~IA1 gene. J Toxi~l Environ Health 1993; 40: 309-19. 7. Cro~ F,T~oliE, Trachm~J,e~aL Funcfion~signific~ce of ~ferent hum~ C~IA1 geno~es. Cardnogenes8 1994; 15: 2961-3. 552 C~cer ~us~ ~d Con~ok Vol 8. 1997 8. Nakachi K, Imai K, Hayashi S, Watanabe J, Kawajiri K. Genetic susceptibility to squamous cell carcinoma of the lung in relation to cigarette smoking dose. CancerRes 1991; 51: 5177-80. 9. Kawajiri N, Nakachi K, Imai K, Yoshii A, Shinoda N, Watanabe J. Identification of genetically high risk individu- als to lung cancer by DNA polymorphisms of the cytochrome P4501AI gene. FEBS 1990; 263: 131-3. 10. Hayashi S, Watanabe J, Kawajiri K. High susceptibility to lung cancer analyzed in terms of combined genotypes of P450IA1 and Mu-class glutathione S-transferase genes.Jpn J Cancer Res 1992; 83: 866-70. 11. Nakachi IC Imai K, Hayashi S, Kawajiri K. Polymorphisms of the CYP1A1 and the gtutathione S-transferase genes associated with susceptibility to lung cancer in relation to cigarette dose in a Japanese population. Cancer Res 1993; 53: 2994-9. 12. Drakoulis N, Cascorbi I, Brokmotler J, Gross CR, Roots L Polymorphisms in the human CYP1A1 gene as suscep- tibility factors for lung cancer:, exon-7 mutation (4889 A to G), and a T to C mutation in the 3'-flanking region. Clin Inraes~ig 1994; 72: 240-8. 13. Hirvonen A, Husgafvel-Pursiainen K, Karjalainen A, Antila S, Vainio H. Point-mutational MspI and Ile-Val polymorphisms closely linked in the CYP1A1 gene: lack' of association with susceptibility to lung cancer in a Finish' study population. Cancer Et~idemiol Biomark Pre~ 1992; 1: 485-9. 14. Shields PG, Caporaso NE, Falk RT, et aL Lung cancer, race, and a CYPIA1 genetic polymorphism. Cancer El~idemiol Biomark Pre~ 1993; 2: 481-5. 15. Sugimura H, Hamada GS, Suzuki I, et aL CYP1A1 and' CYP2E1 polymorphism and lung cancer, case-control study in Rio de Janeiro, Brazil. Pharmacogenetics 1995; 5: $145-8. 16. Seideglu'd ~, Vorachek WR, Pero RW, Pearson WR. Heredi- tary differences in the expression of the human glutathione transferase active on trans-stilbene oxide are due to a gene deletion. Proc Natl Acad Sci 1988; 85: 7593-7. 17. Zhong S, Howie AF, Ketterer B, et aL Glutathione S-tran- ferase mu locus: use of genotyping and phenotyping assays to assess association with lung cancer susceptibility. Car- cinogenesis 1991; 12: 1533-7. 18. Brockm611er J, Gross D, Kerb R, Dakoulis N, Roots I. Correlation between trans-stilbene oxide-glutathione activity and the deletion mutation in the glutathione S- transferase class Mu gene detected by polymerase chain reaction. Biochem Pharmacol 1992; 43: 650-4. 19. McWilliams JE, Sanderson BJS, Harris EL, Richert-Boe ICE, Hetmer WE). Glutathione S-transferase M1 (GSTM1) deficiency and lung cancer risk. Cancer Epidemiol Biomark Prey 1995; 5: 589-94. 20. London SJ, Daly AK, Cooper J, Navidi WC, Carpenter CL, Idle JR. Polymorphism of glutathione S-transferase M1 and lung cancer risk among African-Americans and Caucasians in Los Angeles County, Callfornia. JNC2" 1995; 87: 1246-53. 21. To-Figueras ~, Gen~ M, G6mez-Catal~n J, et al. Gluthathione S-transferase M1 and ¢odon 72 P53 polymor- phism in a Northwestern Mediterranean population and their relation to lung cancer susceptibility. Cancer El~idemiol Biomark Prey 1996; 5: 337-43. 22. Nazar-Stewart V, Mutulsky AG, Eaton DL, et al. The glutathione 5-transferase mu polymorphisms as a marker

for susceptibility to lung cancer. Cancer Res 1993; 53(Suppl): 2313-8. 23. Hirvonen A, Husgafvel-Pursiainen K, Anttila S, Vainio H. The GSTM1 null genotype as a potential risk modifier for squamous cell carcinoma of the lung. Carcinogenesis 1993; 14: 1479-81. 24. Kihara M, Khiara M, Noda K. Lung cancer risk of GSTM1 null genotype is dependent on the extent of tobacco smoke exposure. Carcinogenesis 1994; 15: 415-8. 25. Alexandrie AIL Sundberg MI, Seidegh-d ]', Tornling G, Rannug A. Genetic susceptibility to lung cancer with special emphasis on CYP1A1 and GSTMI: a study on host factors in relation to age at onset, gender and histological cancer types. Carcinogenesis 1994; 15: 1785-90. 26. Anttila S, Hirvonen A, Husgafvel-Pursiainen K, Kar- jalainen A, Nurminen T, Vainio H. Combined effect of CYP1A1 inducibility and GSTM1 polymorphism on histological type of lung cancer. Carcinogenesis 1994; 15: 1133-5. 27. Xu X, Kelsey K, Wiencke ]', Wain J, Christiani D. Cyto- chrome P450 CYP1A1 MspI polymorphism and lung cancer susceptibility. Cancer Epidemiol Biomark Prey 1996; 5: 687-92. 28. Ferris BG. Epidemiology standardization project. Am Rev Respir Dis 1978; 118(Part 2): 1-120. 29. Willett WC, Sampson L, Stampfer MJ, et aL Reproducibil- ity and validity, of a semiquantitative food frequency questionnaire. AmJ Epidemiol 1985; 122: 51-65. 30. Garcfa-C1osas M, Kelsey KT, Wiencke JK, Wain JC, Chris- tiani DC. Dietary intake as a modifier of the association between GSTM1 and CYP1A1 MspI polymorphisms and lung cancer susceptibility. Doctoral Thesis, Chapter 3. Harvard School of Public Health, Boston, MA, 1996. 31. Walsh PS, Metzger DA, Higuchi R. Chdex 100 as medium for simple extraction of DNA for PCR-based typing from forensic material. Biotechniques 1991; 10: 506-13. 32. Chen H, Sandier DP, Taylor JA, et aL Increased risk for myelodysplastic syndromes in individuals with glutathione GSTM1 polymorphism and lung cancer transferase theta 1 (GSTT1) gene defect. Lancet 1996; 347: 295-6. 33. Cheng T]', Christiani DC, Xu X, Wain JC, Wiencke JK, Kelsey KT. Glutathione S-transferase la genotype, diet, and smoking as determinants of sister chromatid exchange frequency in lymphocytes. Cancer Epidemiol Biomarkers /~re'o 1995; 4: 53542. 34. Hastie TJ, Tibshirani RJ. Generalized Additive Models. London, UK: Chapman & Hall, 1990: 35. Broekmtller J, Kerb R, Drakoulis N, Nitz M, Roots I. Genotype and phenotype of glutathione S=transferase class la isoenzymes and in lung cancer patients and controls. Cancer Res 1993; 53: 1004-11. 36. Lin HJ, Han C-Y, Bernstein DA, Hsiao W, Lin BK, Hardy S. Ethnic distribution of the glutathione transferase Mu 1-1 (GSTM1) null genotype in 1473 individuals and application to bladder cancer susceptibility. Carcinogenesis 1994; 15: 1077-81. 37. Poppel GV, Vogel NCL Balderen PJV, Kok FJ. Increase cytogenetic damage in smokers deficient in glutathione S- transferase isoenzyme mu. Carcinogenesis 1992; 13: 303-5. 38. Kihara M, Kihara M, Noda K. Risk of smoking for squamous and small cell carcinomas of the lung modulated by combinations of CYP1A1 and GSTM1 gene polymor- phisms in a Japanese population. Carcinogenesis 1995; 16: 2331-6. 39. Robins J, Pike M. The validity of case-control studies with nonrandom selection of controls. Epideraiology 1990; 1: 273-84. 40. Nakachi K, Hayashi S-i, Kawaiiri K, Imai K. Association of cigarette smoking and CYP1A1 polymorphisms with adenocareinoma of the lung by grades of differentiation. Carcinogenesis 1995; 16: 2209-13. 4~. Okada T, Kawashima K, Fukushi S, Minakuchi T, Nishimura S. Association between a cytochrome P450 CYP1A1 genotype and incidence of lung cancer. Pharrna- cogenetics 1994; 4: 333-40. Cancer Causes and Control. Vol 8. 1997 553