Bladder cancer is associated with smoking, occupational exposures, and glutathione S-transferase (GST) M1 and N-acetyltransferase (NAT) 2 polymorphisms that may influence carcinogen metabolism, but somatic p53mutations are often CpG dinucleotide G:C-A:T transitions that can occur spontaneously. We conducted a case-control study to determine whether p53mutation characteristics might distinguish cases with environmental versus endogenous causes. p53exons 4–9 were amplified from 146 bladder tumors by PCR, screened by single-strand conformational polymorphism analysis, and sequenced. Thirty-one cases were p53-positive, and 112 were p53-negative (germ line or silent). G:C-A:T transitions were also subclassified as CpG or non-CpG. Cases and 215 clinic controls were interviewed. GSTM1, NAT1, and NAT2 polymorphisms were assayed from peripheral blood. Odds ratios (ORs) and 95% confidence intervals (CIs) were estimated using logistic and polytomous regression. Case-control ORs for smoking, occupations, and NAT1*10genotype were similar for p53-positive and p53-negative cases. Associations with GSTM1-null and NAT2-slow genotypes were somewhat stronger for p53-positive [OR, 3.3; CI, 1.4–7.8 (GSTM1 null); OR, 1.8; CI, 0.8–4.0 (NAT2 slow)] than p53-negative cases [OR, 1.5; CI:0.9–2.3 (GSTM1 null); OR, 0.9; CI, 0.6–1.4 (NAT2 slow)]. Smoking was strongly associated with CpG G:C-A:T (OR, 15.3; CI:3.6–65) versus other G:C-A:T (OR, 1.8; CI, 0.3–9.8). NAT2 slow genotypes were also associated with CpG G:C-A:T (OR, 6.2; CI:0.7–52), whereas GSTM1 null was associated with non-CpG G:C-A:T (OR, 7.8; CI, 0.9–65). Associations were not substantially different for case subtypes defined by p53mutation status alone. Estimates for p53 subtypes were imprecise but support in vitro evidence that some CpG G:C-A:T transitions may be caused by smoking and other environmental mutagens.
Missense mutations in the p53tumor suppressor gene (TP53) are the most common somatic mutations identified among cancers (1) . Cancer-associated mutations in the highly conserved DNA-binding domain may prevent or inhibit p53-mediated cell cycle arrest, DNA repair, programmed cell death, and other protective responses to cell stress and DNA damage (2, 3, 4) . Some p53mutations are associated with specific carcinogens, for example, CC to TT transitions are associated with UV light (5) and G to T transversions are associated with benzo(a)pyrene (6 , 7) . Others may occur spontaneously, most notably G:C to A:T transitions at CpG dinucleotides after 5-methyl cytosine deamination (8) . CpG G:C-A:T transitions are particularly common p53mutations in breast and colon cancers (9) , two cancers for which prominent environmental causes have not been identified. In contrast, CpG G:C-A:T transitions are relatively uncommon in cancers with strong environmental risk factors; namely, cancers of the lung, skin, and liver. Therefore, it has been proposed that p53G:C-A:T transitions at CpG are an attribute of cancers caused by endogenous cellular processes (9, 10, 11) .
CpG G:C-A:T transitions account for almost 25% of p53mutations in bladder cancers (9) , but this differs from the “endogenous” pattern associated with colon cancer, where almost half of all p53mutations are CpG G:C-A:T (10 , 11) . The spectrum of bladder cancer p53 mutations also differs from lung cancer, even though cigarette smoking is probably a contributing cause in over one-third of all bladder cancer cases (12) . Specifically, G-T transversions are relatively uncommon (about 8% of bladder cancer p53mutations compared with 27% of lung cancer mutations), whereas CpG G:C-A:T transitions are twice as common (22% for bladder versus 11% for lung cancer; Ref. 9 ). This indeterminate pattern of bladder cancer p53 mutations might be a consequence of etiologic heterogeneity; if so, characteristics of p53mutations might distinguish cases caused by smoking or other environmental risk factors from “endogenous” cases with carcinogenic mutations caused by spontaneous cellular processes (13, 14, 15, 16) . Genetic susceptibility factors for bladder cancer might also vary among p53case subtypes, if they increase risks by modifying bladder cell exposures to environmental mutagens. To evaluate these hypotheses, we compared case-control ORs 2 for bladder cancer case subgroups defined by acquired p53 mutations to determine whether p53 subtype-specific associations varied for smoking, occupational exposures, and GSTM1, NAT2, and NAT1polymorphisms that may influence bladder cell exposures to mutagenic byproducts of cigarette smoke.
MATERIALS AND METHODS
Patients with histologically confirmed transitional cell carcinomas of the urinary bladder (n = 245) were enrolled from Urology Clinics at the University of North Carolina Hospitals and Duke University Medical Center (17 , 18) . Controls were 215 patients from the same clinics without a history of cancer (other than nonmelanoma skin cancer), who were frequency-matched to cases based on ethnicity, sex, and 10-year age intervals. Institutional review boards at each participating institution approved all study protocols.
Demographic and exposure data were obtained from a structured questionnaire administered in person by a trained nurse-interviewer. Years of smoking, pack-years, and current smoking status were based on exposures up to 2 years before initial diagnosis (cases) or interview (controls). Participants provided information on all jobs held at least 2 years, on any work in specific jobs or industries of a priori interest, and on occupational or home use of specific products of a priori interest that were used on at least 5 occasions. These exposures were counted only if they began more than 2 years before diagnosis or interview. Exposure data were classified as missing for jobs or products with unknown start dates and years of exposure.
NAT1and NAT2polymorphisms were ascertained using RFLP-PCR analysis of DNA extracted from peripheral blood lymphocytes obtained at the time of interview, and results have been reported previously (18) . Four NAT1 alleles were assayed: NAT1*4 (wild type), NAT1*3 (C1095A), NAT1*10 (T1088A, C1095A), and a 9-bp deletion upstream of nucleotide 1088 (previously referred to as NAT1*11). Five NAT2alleles were assayed: NAT2*4 (wild type), NAT2*5D (C481T), NAT2*6B (G590A), NAT2*7A (G857A), and NAT2*14A (G191A; Ref. 19 ). Participants were classified as NAT1*10if they had at least one NAT1*10 allele, and as NAT2 slow if they lacked a NAT2*4 allele. GSTM1alleles were identified using a differential PCR assay, as described previously (17) . Participants were classified as GSTM1 null if both alleles were deleted.
p53 Mutation Status
Sections were cut from paraffin-embedded tumor blocks, reviewed by a pathologist, and manually microdissected if they included a substantial amount of nonneoplastic tissue. One or more sections were used to produce 1–3 μl of tumor lysate DNA from each sample. Samples were digested in 25–200 μl of lysis buffer containing 1% Triton X-100 and 0.4 μg/μl proteinase K in 1× Cetus PCR buffer [50 mm KCl, 10 mm Tris-HCl (pH 8.3), 0.001% gelatin, and 1.5 mm MgCl2]. The mixtures were incubated at 56°C for 24 h and 95°C for 10 min to inactivate the proteinase K. The final solution was centrifuged, and the supernatant was stored at 4°C.
Tumors were screened for mutations in exons 4–9, which encompass the DNA-binding domain of the encoded protein. Individual exons or exonic fragments were amplified in separate reactions. Due to its size, exon 4 was amplified as two overlapping segments. Exon 5 was amplified as a single 294-bp fragment or as two separate fragments in samples with lower quality DNA. Primer sequences for exons 4–8 were reported elsewhere (20) ; primers for exon 9 were 5′-GCCTCAGATTCACTTTTATCACC-3′ and 5′-CATTTTCACTGTTAGACTGGAAAC-3′. PCR was carried out in a 50-μl reaction mixture containing 1–3 μl of tumor lysate DNA, 300 nm of each primer, 1× Cetus buffer, 200 μm of each deoxyribonucleotide, and 1.25–2.5 units of AmpliTaq DNA polymerase (Perkin-Elmer, Norwalk, CT). The mixture was overlaid with mineral oil and amplified for 1 cycle at 94°C for 4 min, 55°C for 1 min, and 72°C for 1 min; followed by 33 cycles at 94°C for 1 min, 55°C for 1 min, and 72°C for 1 min; and a final extension cycle of 94°C for 1 min and 60°C for 10 min. PCR products were resolved by electrophoresis in nondenaturing 10% polyacrylamide mini-gels and visualized with ethidium bromide staining under UV light.
SSCP-PCR was performed using internal nested primers and an aliquot of each PCR product as template. Internal primers for exon 9 were 5′-ATCACCTTTCCTTGCCTCT-3′ and 5′-CATTTTGAGTGTTAGACTGG-3′; internal primers for exons 4–8 were reported elsewhere (20) . Reactions consisted of a total volume of 20 μl containing 1 μl of diluted PCR product; 300 nm of inner primers; 1× Cetus buffer; 100 μm each of dGTP, dTTP and dATP; 15 μm dCTP; 2.5 units of AmpliTaq polymerase; and 100 μCi/ml [α-32P]dCTP. Cycle conditions were as described for the first PCR. To improve the resolution of exon 4-2 bands, the SSCP-PCR products were digested with 5 units of AluI (Boehringer Mannheim) in 25 ml of 1× AluI buffer. All products were diluted 30–50-fold with 0.1% SDS/10 mm EDTA, mixed with an equal volume (3 μl) of stop buffer (95% formamide, 20 mm EDTA, 0.05% bromphenol blue, and 0.05% xylene cyanol FF), and denatured at 94°C for 5 min. Electrophoresis in 6% polyacrylamide nondenaturing gels was carried out at room temperature and at 4°C. Each gel included a negative control (wild-type p53) and a positive control for the appropriate exon. Samples with abnormal band migration at either temperature were sequenced.
The first PCR product was used as template to generate single-stranded DNA for sequencing. Asymmetric PCR was carried out as described above, but with inner primer concentrations of either 500 nm forward and 10 nm reverse primer or 10 nm forward and 500 nm reverse primer. Forward and reverse PCR products were purified through Centricon-100 concentrators (Millipore Corp., Bedford, MA) and sequenced using a Sequenase version 2.0 sequencing kit (Amersham Biosciences, Piscataway, NJ). If identical mutations were detected in both strands, a second aliquot of the original DNA lysate was amplified and sequenced. Only samples showing the same mutation on forward and reverse strands in two independent PCR reactions were considered true positives.
Analyses were run using Stata, Release 6.0 (21) . Ninety-nine cases were missing p53status because tumor tissue was not available (n = 88) or fewer than four p53exons were successfully amplified (n = 11). p53-negative cases had no mutations detected or silent mutations only. p53-positive cases had at least one protein altering mutation. For some analyses, p53-positive cases were subclassified by type of mutation, including deletions or insertions, G-T and other transversions, G:C to A:T transitions at CpG dinucleotides, and other transitions.
p53mutations in study cases were compared with p53mutational spectra for bladder tumors and other cancers from the IARC TP53Mutation Database (Release 7, September 2002; Ref. 9 ). IARC mutation data were for exons 4–9 only. Splice mutations, intron mutations, and mutations in cell lines or xenographs were not counted (9) .
Associations were estimated for potential risk factors when there were at least 15 exposed cases among those with known p53status. Unconditional logistic regression models were used to estimate ORs and 95% CIs comparing all bladder cancer cases with controls and p53-positive cases with p53-negative cases. Unconditional polytomous regression models were used to simultaneously estimate ORs for p53-positive cases compared with controls and ORs for p53-negative cases compared with controls (22) . All models included sex, race (white or other), age at diagnosis or interview (categorized or using upper and lower tail restricted quadratic splines; Ref. 23 ), and years of smoking (continuous, with a 2-year lag), unless otherwise specified.
Complete-data-only models (of data from 215 controls and 146 cases with known p53 status) were fit using standard maximum likelihood methods. A statistical missing-data technique known as the EM algorithm was used to derive maximum likelihood estimates based on all available data, including exposure data from 99 cases with unknown p53 status (24) . Complete-data-only estimates are unbiased but less precise than EM model estimates when case subtype data are missing completely at random. Complete-data-only estimates may be biased when missing case subtype data are associated with model covariates, but this bias will be reduced when models are fit using EM, as long as missing case subtype data are not also associated with the outcome. Details on the use of EM in case subgroup analyses and results of a simulation study comparing EM with complete-data-only methods are provided elsewhere (25) . For this study, EM model ORs were more precise but similar in magnitude to corresponding estimates from models based only on observations with complete p53 data. However, ORs comparing p53-positive and p53-negative cases with controls always fell on opposite sides of the corresponding OR for all study cases combined when the EM method was used, whereas ORs based on complete-data-only often did not. This suggests some bias in the complete data estimates because ORs for all study cases combined should correspond to a weighted average of subtype-specific associations (25) ; therefore, we reported EM model estimates unless otherwise noted.
Joint effects of smoking and gene polymorphisms were modeled using indicator variables for separate or joint exposure to pairs of dichotomous covariates [current smokers up to 2 years prior versus former smokers or never-smokers, GSTM1 null versus GSTM1 functional, any NAT1*10 allele versus none, NAT2 slow (no NAT2*4 allele) versus other NAT2genotypes]. The common referent group for each set of comparisons consisted of former smokers or never-smokers with the referent genotype. Joint ORs were compared with those predicted for average additive effects (26) .
Logistic regression models comparing controls and cases with specific types of G:C-A:T transitions were adjusted for age (<60 years, ≥60 years) and fit using standard maximum likelihood methods. G:C-A:T transitions were subclassified by site (CpG or non-CpG) or by the specific base change (C-T or G-A) to determine whether associations were specific for the location versus the type of mutation. Other mutations (insertions/deletions, transversions, or other transitions) were too uncommon to evaluate separately.
p53 mutation status was determined for 146 cases. Of these, 115 (79%) were classified as p53-negative (112 cases without detected mutations and 3 with silent mutations only). Thirty-one (21%) were classified as p53-positive, including one with two silent and three missense mutations, one with two missense mutations, and one with a missense mutation and a 17-bp deletion (Table 1) ⇓ . Four p53-negative cases had silent codon 36 polymorphisms (27) , and two p53-positive cases had silent codon 213 polymorphisms (28) .
Most mutations (63%) were G:C-A:T transitions, including 10 missense mutations at CpG dinucleotides, and 16 at other sites (9 missense, 2 nonsense, and 5 silent). There were five additional transitions (one T-C and four A-G), four G-T transversions (two missense and two silent), three other transversions, one 17-base deletion, and two single-base insertions. The spectrum of mutations was comparable to bladder tumors in the IARC TP53 mutation database (Release 7; Ref. 9 ; Fig. 1 ⇓ ).
Forty percent of the original 245 study cases were missing p53 data, including 88 that were not assayed and 11 with fewer than four exons amplified. These cases were similar to successfully assayed cases with regard to tumor grade but were more likely to have been diagnosed before age 50 years (19% versus 9%) or more than 2 years before interview (71% versus 47%; Table 2 ⇓ ).
Women were less likely to have p53-positive bladder cancer than men (p53-positive versus p53-negative: OR, 0.5; CI, 0.1–1.4; Table 2 ⇓ ). p53-positive cases were higher grade than p53-negative cases (grade IV versus grade I: OR, 6.5; CI, 2.0–22). p53 mutation status was not clearly related to age at diagnosis or years since first diagnosis.
Only 4 of 36 occupations and 6 of 11 exposures of a priori interest met our minimum sample size criteria for evaluation (15 successfully assayed cases in each exposure category). Most estimates were close to the null, and CIs surrounding case:control ORs for p53-positive and p53-negative cases overlapped (Table 3) ⇓ . There was weak evidence of a subtype-specific association for work as a gasoline station attendant (p53-positive: OR, 2.1, CI, 0.8–5.4; p53-negative: OR, 0.9, CI, 0.5–1.9) and exposure to soot (p53-positive: OR, 1.8, CI, 0.6–5.3; p53-negative: OR, 1.1, CI, 0.5–2.4).
Only three p53-positive cases never smoked (Table 4) ⇓ . Case-control ORs for p53-positive cases were comparable with corresponding ORs for p53-negative cases, particularly when current smokers were compared with ex-smokers and never-smokers combined (p53-negative OR, 5.3; p53-positive OR, 5.6). Case:control ORs for years and pack-years of smoking were also comparable for p53-positive and p53-negative cases.
GSTM1-null genotype was more strongly associated with p53-positive cases (OR, 3.3; CI, 1.4–7.8) than p53-negative cases (OR, 1.5; CI, 0.9–2.3; Table 5 ⇓ ). NAT2-slow genotype was associated with p53-positive cases only (p53-positive: OR, 1.8; CI, 0.8–4.0; p53-negative: OR, 0.9, CI, 0.6–1.4). Case:control ORs for no NAT1*10 versus any NAT1*10 allele were similar for p53-positive and p53-negative cases. Missing polymorphism data may have influenced results, but sensitivity analyses (29) showed that subtype-specific differences in associations with GSTM1and NAT2would persist under extreme missing data scenarios (for example, if all p53-positive cases with missing genotype had been NAT2-fast, and all missing controls had been NAT2-slow).
Joint and separate effect estimates for polymorphisms and current smoking were imprecise, and CIs surrounding subtype-specific estimates overlapped. For p53-positive bladder cancer, the estimated joint effect of smoking and GSTM1-null genotype was more than additive [observed joint OR of 15.4 (CI, 4.7–48) versus predicted joint OR of 6.1, based on OR = 4.6 (CI, 1.0–20) for smokers with active GSTM1 and OR = 2.5 (CI, 0.9–7.4) for nonsmokers who were GSTM1 null]. For p53-negative cases, joint effect estimates were close to additive [observed joint OR = 6.6 (CI, 3.3–13) versus predicted OR = 6.3, based on OR = 5.7 (CI, 2.8–12) for smokers with active GSTM1 and OR = 1.6 (CI, 0.9–2.6 for nonsmokers who were GSTM1 null)]. Other joint effect estimates (for NAT1*10 with smoking, NAT2-slow genotypes with smoking, and NAT1*10 with NAT2-slow genotypes) were comparable for p53-positive and p53-negative cases.
ORs for smoking and GSTM1, NAT1, and NAT2 polymorphisms were comparable for cases with any G-A versus any C-T transition, but ORs differed between cases with CpG G:C-A:T transitions (10 cases: 8 G-A and 2 C-T) and cases with non-CpG G:C-A:T (8 cases: 4 G-A and 6 C-T; Table 6 ⇓ ). Current smoking was strongly associated with CpG G:C-A:T (OR, 15.3; CI, 3.6–65), but not G:C-A:T transitions at other sites (OR, 1.8; CI, 0.3–9.8; Table 6 ⇓ ). NAT2-slow genotype was positively associated with CpG G:C-A:T (OR, 6.2; CI, 0.7–52) and inversely associated with other G:C-A:T transitions (OR, 0.5; CI, 0.1–2.3). Missing NAT2data may have influenced estimates because ORs for NAT2and cases with CpG G:C-A:T could have ranged from 1.8 to 8.5, depending on the genotype of two CpG-positive cases and nine controls with unknown NAT2. GSTM1-null genotype was more strongly associated with non-CpG G:C-A:T transitions (OR, 7.8; CI, 0.9–65) than with CpG G:C-A:T (OR, 1.9; CI, 0.4–8.0).
The association between smoking and bladder cancer did not differ substantially between cases with and without p53 mutations, consistent with the results of other studies (30, 31, 32) . However, current smoking was more strongly associated with cases that had CpG G:C-A:T transitions than cases with other types of p53 mutations. Associations between smoking and bladder cancers with CpG G:C-A:T transitions have not been specifically evaluated by others, but smokers were overrepresented among cases with CpG G:C-A:T in three (15 , 30 , 31) of five (14 , 33) published studies where smoking status was linked to specific mutations. In the largest of these, 7 of 8 CpG G:C-A:T cases were current smokers, while only 40 of 84 (48%) of all cases were current smokers (30) . Similarly, 7 of 10 CpG G:C-A:T cases were current smokers in our study, compared with only 117/244 (48%) of all cases combined. These findings contradict expectations, if the assumption that CpG G:C-A:T mutations are simply a byproduct of endogenous processes is correct (10) .
Several biological mechanisms could account for associations with bladder cancer subtypes defined by somatic p53 mutations, with the most obvious involving an exposure that is a direct cause of p53 DNA damage (34) . Smoking is associated with increased mutagenic BPDE and aromatic amine 4-ABP DNA adducts in urothelial cells (32 , 35, 36, 37, 38) . BPDE adducts are thought to be the primary cause of p53 G-T transversions in lung tumors, which occur more often than expected at CpG sites (39) . Cytosine residues in p53 CpG dinucleotides appear to be consistently methylated (8) , and guanines adjacent to 5-methylcytosines may be particularly susceptible to BPDE-mediated mutations because BPDE adducts formed at higher than expected rates at methylated CpG in experimental systems (40 , 41) . Repair of BPDE adducts also may be inhibited at methylated CpG (39) . BPDE guanine adducts have been strongly associated with G-T transversions in vitro (6 , 7 , 42) , but mutations that occur in vivo may be influenced by the local environment (43, 44, 45) , with G-A transitions occurring as the predominant mutation in some contexts (46) . Like BPDE, 4-ABP forms bulky adducts at guanine bases that may cause G-A transitions as well as transversions (44) . 4-ABP may also cause oxidative damage (47) , and 5-methylcytosine residues may be more susceptible to oxidative damage than unmethylated cytosines (48) . Taken as a whole, these findings suggest that smoking-associated BPDE, 4-ABP, or related carcinogens might cause or enhance the formation of G:C-A:T transitions at CpG sites (45) .
Associations between p53-positive cases and gene polymorphisms that might influence the activation and detoxification of BPDE and 4-ABP may clarify the role of these compounds in bladder cancer etiology (31) . Glutathione conjugation by GSTM1 may detoxify BPDE, and the inactive GSTM1-null genotype has been associated with increased BPDE adducts in lung tissue (49) and increased urine mutagenicity (50) . In our study, p53-positive cases were more likely to be GSTM1 null than p53-negative cases, in agreement with two previous case:case analyses (31 , 51) . We also found evidence of a greater than additive effect of GSTM1-null genotype and current smoking among p53-positive cases, but not p53-negative cases. However, the association with GSTM1 null was stronger for cases with G:C-A:T transitions at non-CpG versus CpG sites; therefore, the association between GSTM1 null and p53-positive bladder cancers could not have been driven by an excess of CpG mutations. We had inadequate numbers of transversion-positive cases to evaluate them as a distinct outcome, but three of six were GSTM1 null. In the study by Brockmoller et al. (31) , bladder cancer cases with p53 transversions were more likely to be GSTM1 null (6 of 7) than cases with p53 transitions (8 of 18).
4-ABP metabolism begins with N-hydroxylation by cytochrome P450 enzymes, which generates an active intermediate that may be N-acetylated in a detoxification reaction or O-acetylated to a more reactive product. NAT1 and NAT2 may catalyze both reactions, but hepatic NAT2 is thought to be primarily responsible for detoxification, whereas NAT1 activity in bladder cells may increase local O-acetylation. NAT2-slow (52) and NAT1*10 (53) genotypes have been associated with increased 4-ABP DNA adducts in bladder cells, although not all studies have confirmed these results (32 , 54) . We did not find an association between NAT1*10 genotype and p53 mutations, but we did find a positive association between NAT2-slow genotypes and p53-positive bladder cancer that was stronger for cases with CpG G:C-A:T than other case subtypes. In contrast, previous case:case analyses reported no association between NAT2and p53 (31) or a stronger association with p53-negative cases (51) . Brockmoller et al. (31) reported that the NAT2-slow genotype was more strongly associated with cases that had p53 transversion mutations than with other cases; in our study, three of six transversion-positive cases were NAT2 slow.
Eight of ten CpG G:C-A:T transitions among our cases were G-A, which suggests the possibility of strand bias in CpG G:C-A:T mutations. Strand bias is associated with preferential repair of damage on the transcribed DNA strand during TCR, a process triggered by bulky DNA adducts, including BPDE and 4-ABP guanine adducts. Strand bias would not be expected as a consequence of spontaneous deamination of methylated cytosine residues to thymine because this would normally be corrected with equal efficiency on both DNA strands (10) via base excision repair (55) . Therefore, evidence of strand bias supports the hypothesis that some CpG transitions might be caused by preferential adduction of guanine residues in CpG dinucleotides. Evidence of G-A strand bias in CpG G:C-A:T transitions has not been specifically noted by others (15 , 30, 31, 32, 33) , but G-A account for 59% of CpG G:C-A:T among primary bladder cancers in the IARC TP53 mutation database (Release 7; Ref. 9 ). Several other cancers in the database also appear to have more G-A than C-T transitions at CpG sites, including breast cancer (62% G-A), ovarian cancer (59% G-A), hematopoietic cancers (69% G-A), and colon and colorectal cancers (59% G-A). A notable exception is lung cancer, where only 47% of CpG G:C-A:T transitions are G-A. The extent of this apparent strand bias in CpG G:C-A:T is far less pronounced than strand bias seen in G:C-T:A transversions (about 89% G-T in lung cancer), and IARC TP53data must be interpreted with some caution because they are derived from peer-reviewed reports that cannot be independently verified (9) . Nonetheless, these data suggest that at least some CpG G:C-A:T transitions may by a consequence of lesions that induce TCR, such as BPDE and 4-ABP adducts (56) , and some forms of oxidative damage (57 , 58) .
NNK and other tobacco-associated nitrosamines are carcinogens that also might be preferentially associated with G:C-A:T at CpG sites. Nitrosamine-mediated guanine alkylation (for example, to O6-methylguanine) typically results in G:C-A:T mutations (59 , 60) , and there is experimental evidence that direct repair of alkylated guanines by O6-methylguanine DNA methyltransferase enzymes is inhibited at methylated CpG (61) . We would not expect NNK-mediated p53 mutations to be associated with the NAT1, NAT2, and GSTM1 polymorphisms we evaluated because nitrosamine activation occurs via CYP450-mediated α-hydroxylation, and detoxification is primarily via glucuronide conjugation (62) . Strand bias also would not be expected because non-bulky adducts caused by NNK and other alkylating agents do not trigger TCR (56) .
Our study included more p53-positive bladder cancer cases than previous studies of bladder cancer and environmental exposures (14 , 15 , 30, 31, 32, 33 , 51 , 63 , 64) , but numbers of cases with specific types of p53 mutations were small, and associations may have occurred by chance. For example, random error might explain discordant associations for GSTM1-null and NAT2-slow genotypes with G:C-A:T at non-CpG versus CpG sites. On the other hand, the spectrum of p53 mutations among our cases was similar to bladder tumors in the IARC TP53 database (Fig. 1 ⇓ ; 9 ), and we confirmed previous findings that acquired p53 mutations are associated with higher grade bladder cancers (13, 14, 15 , 30, 31, 32, 33 , 51 , 65, 66, 67) . False positive results due to PCR contamination or Taq polymerase errors were unlikely, given the variability of the mutations that were detected and the stringency of our laboratory protocol, which required verification of all abnormal p53 sequences in an independently amplified sample of DNA. Some mutations may not have been detected in exons that were not successfully amplified, but samples missing data for three or more exons were excluded from analyses (n = 11). False negative results might have occurred if mutant sequences were substantially diluted by wild-type DNA from normal cells, but nonneoplastic cells were removed from tissue sections before DNA extraction when possible, and SSCP electrophoresis was run at both room temperature and 4°C to increase sensitivity. Our use of clinic-based controls could have biased estimates if bladder cancer risk factors were associated with conditions leading to clinic participation (68) , but most study controls were being treated for impotence and benign prostatic hyperplasia, conditions not known to have risk factors in common with bladder cancer. Our study included incident and prevalent bladder cancer cases, with about one-third having been diagnosed >5 years before interview. This may have distorted relations with factors associated with survival, but years since diagnosis was not related to p53 status among assayed cases.
In conclusion, our results did not indicate a pronounced difference in risk factors for case subtypes defined by p53 mutation status alone, although several occupations and occupational exposures, as well as NAT2-slow and GSTM1-null genotypes, were more common among p53-positive cases. However, we did find evidence that risk factors varied when cases were classified by type of p53 mutation: smoking and NAT2-slow genotypes were associated with G:C-A:T transitions at CpG dinucleotides, whereas GSTM1-null genotype was associated with G:C-A:T transitions at other sites. Although common smoking-related adducts are associated with G-T transversions in experimental systems, they can also cause G-A transitions and other mutations (43 , 44) , and there is increasing evidence that CpG bases may be more susceptible than other sites to attack by environmental mutagens (8 , 45 , 48 , 69 , 70) . In addition, DNA repair may be inhibited at CpG, including TCR of bulky adducts (39) and direct repair of O6-methylguanine adducts by O6-methylguanine DNA methyltransferase (61) . Finally, tissue-specific selective pressures related to the specific functional effects of individual mutations might have a substantial influence on p53 mutational spectra associated with environmental exposures among different cancers because the effects of specific p53 mutations vary, with some causing only partial loss of function, whereas others may cause functional gains (71 , 72) . Our findings require confirmation in a larger study but are compatible with hypotheses suggesting that G:C-A:T transitions at CpG may be caused by smoking and other environmental exposures, as well as by endogenous processes (41 , 43 , 45 , 70) . If this is true, a larger proportion of carcinogenic p53 mutations in bladder cancers (and possibly other “endogenous” cancers) might be caused by modifiable environmental exposures than has previously been assumed (11) .
We thank all who contributed to the case-control study of bladder cancer on which this work was based, including Elizabeth Stephens and Trisha Castranio for technical assistance with genotyping, David Paulson and Cary Robertson (Urologic Surgery Division, Duke University Medical Center), and James Mohler (Urologic Oncolgy, University of North Carolina and Roswell Park Cancer Institute).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
↵1 To whom requests for reprints should be addressed, at Senior Investigator, Epidemiology Branch and Head, Molecular and Genetic Epidemiology Group, Laboratory of Molecular Carcinogenesis, National Institute of Environmental Health Sciences, NIH, MD A3-05, Room A362, 111 Alexander Drive, P. O. Box 12233, Research Triangle Park, NC 27709. Phone: (919) 541-4631; Fax: (919) 541-2511; E-mail:
↵2 The abbreviations used are: OR, odds ratio; CI, confidence interval; GST, glutathione S-transferase; NAT, N-acetyltransferase; SSCP, single-strand conformational polymorphism; EM, Expectation Maximization; BPDE, benzo(a)pyrene diol epoxide; 4-ABP, N-(deoxyguanosin-8-yl)-4-aminobiphenyl; TCR, transcription-coupled repair; NNK, 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone.
- Received June 23, 2003.
- Revision received August 22, 2003.
- Accepted August 26, 2003.
- ©2003 American Association for Cancer Research.