Common Genetic Variation in TP53 Is Associated with Lung Cancer Risk and Prognosis in African Americans and Somatic Mutations in Lung Tumors

Lung cancer is primarily caused by tobacco smoking, but susceptibility is likely modified by common genetic variation. In response to many forms of cellular stress, including DNA damage, the p53 protein functions to induce cell cycle arrest, DNA repair, senescence, or apoptosis. We hypothesized that common TP53 haplotypes modulate pathways of lung carcinogenesis and lung cancer susceptibility or prognosis. To investigate our hypothesis, 14 polymorphisms in TP53, including haplotype tagging and coding single nucleotide polymorphisms, were genotyped in two studies from the greater Baltimore, Maryland area. One study is a case-control study and the second is a case-only study for which TP53 mutational spectra data are available. African Americans with Pro-T-A-G-G haplotypes of the combined TP53 polymorphisms TP53_01 (rs1042522), TP53_65 (rs9895829), TP53_66 (rs2909430), TP53_16 (rs1625895), and TP53_11 (rs12951053) had both an increased risk for lung cancer (odds ratio, 2.32; 95% confidence interval, 1.18-4.57) and a worsened lung cancer prognosis (hazards ratio, 2.38; 95% confidence interval, 1.38-4.10) compared with those with Arg-T-A-G-T haplotypes. No associations of TP53 polymorphisms with lung cancer were observed in Caucasians. In the case-only study, several polymorphisms in TP53 and TP53 haplotypes, overlapping regions of TP53 associated with risk and prognosis in African Americans, were associated with increased odds of somatic TP53 mutation in lung tumors in Caucasians. In conclusion, common genetic variation in TP53 could modulate lung cancer pathways, as suggested by the association with lung cancer in African Americans and somatic TP53 mutation frequency in lung tumors. (Cancer Epidemiol Biomarkers Prev 2007;16(2):214–22)

Lung cancer is the leading cause of cancer death worldwide (1). Although cigarette smoking is the predominant cause of lung cancer, not all smokers develop lung cancer (2), suggesting that particular individuals may be more susceptible to cigarette smoke. Familial aggregation studies have provided evidence for a genetic component to lung cancer risk (see refs. 3, 4 for review). Therefore, susceptibility to lung cancer may be due, in part, to interindividual genetic variation in the form of single nucleotide polymorphisms (SNP) or common allele variants.

Inactivation of the TP53 tumor suppressor gene is a frequent and early event in lung carcinogenesis (5-9). The p53 protein functions to induce growth arrest, DNA repair, senescence, and apoptosis in response to cellular stress, including DNA damage (10-12). Consistent with the p53 tumor suppressor functions, mutations in p53 are present in >90% of small cell lung cancers and >50% of non–small cell lung cancers (7, 13-16).

Based on the role of the p53 protein in preventing tumor formation, genetic variability in the TP53 gene may modulate lung cancer susceptibility. Many variants in TP53 were identified on National Center for Biotechnology Information SNP database (17-19).⁷

The biological activity associated with the most commonly studied polymorphism in TP53, Arg⁷²Pro, differs depending on the amino acid. Arginine at codon 72 of p53 more effectively induced p53-mediated apoptosis (20, 21), partially through targeting of p53 to the mitochondria (20). Meanwhile, the Pro⁷² p53 forms of p53, when compared with Arg⁷², more efficiently induced cell cycle arrest (21, 22) and DNA repair (23). Several epidemiology studies examined the association of TP53 Arg⁷²Pro polymorphisms with lung cancer, with inconsistent results (17-19).

Several studies observed that the Arg⁷²Pro polymorphism of TP53 is in linkage disequilibrium (LD) with other potential susceptibility alleles in TP53. Therefore, the Arg⁷²Pro polymorphism may incompletely mark a larger susceptibility haplotype. Previously, we showed that the Arg⁷²Pro polymorphism in TP53 was associated with an increased frequency of somatic mutations in lung tumors (24). Based on this observation, we expanded our study of the genetic variation in TP53. We hypothesized that variant TP53 haplotypes modulate pathways of lung carcinogenesis, therefore effecting lung cancer susceptibility and prognosis. To investigate this hypothesis, we examined 14 polymorphisms in TP53 in a case-control study and a case-only study of lung cancer, both of which were conducted in the greater Baltimore area, and studied the association of TP53 genotypes and haplotypes with lung cancer and somatic TP53 mutations in lung tumors.

Two study populations were used in this study, a case-control and case-only study, both of which were conducted in the greater Baltimore, Maryland area. Subjects were accrued from the same hospitals, but at different times and clinics. For the case-control population, lung cancer cases and controls were recruited from 1998 to 2003 as part of an ongoing study described previously (25, 26). Briefly, lung cancer patients were of Caucasian or African American descent, residing in Metropolitan Baltimore, the Maryland Eastern Shore, and recruited from seven hospitals in Baltimore. Hospital-based controls were frequency matched to cases by gender, ethnicity, age, smoking history, and hospital. Hospital-based controls were cancer-free patients recruited from the same hospitals as lung cancer cases and were recruited from internal medicine clinics, primary care, pulmonology, and cardiology clinics. Population controls were identified from Department of Motor Vehicles lists and matched to cases by age, gender, and ethnicity. Blood specimens were processed immediately after collection for isolation of blood components and stored at −70°C.

Lung cancer cases were recruited, in a separate study, from 1974 to 1999 to form a case-only study described previously (24). Inclusion in the study was based on the availability of paraffin-embedded tumor tissue for DNA sequencing for TP53 mutations. Exons 5 to 8 of TP53 were sequenced using the p53 GeneChip (Affymetrix, Santa Clara, CA), single-stranded conformation polymorphism and manual sequencing of DNA from paraffin-embedded tissues from surgical resections as described previously for this study (24).

Institutional Review Board approval was obtained from all participating institutions and the NIH. Informed consent was obtained from all participants.

Blood samples were available from 99% of case-control participants. DNA for genotyping was extracted using 300 μL of isolated buffy coat using the FlexiGene kit following the manufacturer's instructions (Qiagen, Valencia, CA). DNA was available for genotyping for 80% of the participants from the case-only study and DNA was extracted from noninvolved tissue (82%) or tumor tissue (5%; ref. 24), or buffy coat (13%) following manufacturer's instructions (Qiagen).

Haplotype tagging SNPs and coding SNPs were selected from a subset of validated TP53 SNPs from TP53 sequencing all exons and conserved regions 5 kb upstream and 3 kb downstream of the gene using the SNP500 population⁸

(27). Genotyping was done on the SNP500 population and a STATA module⁹ (28) was used to select haplotype tagging SNPs from all polymorphisms with a minor allele frequency >5% after phase was inferred using Phase 2.0 (29). Genotyping assays on the selected haplotype tagging SNPs plus coding SNPs (Arg⁷²Pro, TP53_01, rs1042522; Arg²¹³Arg, TP53_18, rs1800372; and Val²¹⁷Met, TP53_64) were done as described by the National Cancer Institute Core Genotyping Facility⁸ (Supplementary Table S1). Assays were designed for selected SNPs and were concordant with the sequencing results on the SNP500 population. At least 10% of the case-control and case-only samples from each genotyping assay was duplicated. Overall, the concordance between duplicates was 98% and each assay had >95% concordance.

Date and cause of death were obtained through the National Death Index (NDI)¹⁰

using the NDI Plus search, which provides cause of death codes. The NDI Retrieval Program is used to search the NDI file to determine whether a particular NDI death record qualifies as a possible record match with a particular user record. To qualify as a possible match, records must satisfy an algorithm based on at least one of seven matching criteria, including the Social Security number, exact month of birth, and first and last name, according to instructions from the NDI. A person scored with a ‘no match’ was presumed alive. Survival data are reported through December 31, 2004. Case survival was dichotomized as ‘alive’ or ‘dead’ based on survival status 5 years following diagnosis.

All survival analysis was done for lung cancer survival and all-cause mortality. Causes of death unrelated to lung cancer were censored from the study. Cause of death was evaluated from death certificate data obtained from NDI. Any mention of lung cancer, or another cancer death within 2 years of diagnosis, on a death certificate was treated as death from lung cancer.

Differences in the characteristics of lung cancer cases and controls were compared by χ² for categorical values or by Student's t tests for continuous measures as indicated. Departures from Hardy-Weinberg equilibrium for TP53 genotypes were evaluated by calculating the expected genotype frequencies based on observed allele frequencies and comparing expected frequencies with observed genotype frequencies using χ² tests. Never smokers were defined as those who smoked <100 cigarettes during their lifetime (case-control) or <6 months in duration (case-only). Former smokers were defined as those who reported quitting smoking ≥1 year before the date of diagnosis. Race was classified by self-report.

Unconditional logistic regression models were used to calculate adjusted odds ratios (OR) to assess the effect of TP53 genotypes on the odds of lung cancer in the case-control study or odds of TP53 mutation in lung tumors in the case-only study using PROC LOGISTIC in SAS (version 8.1; SAS Institute, Cary, NC) for those SNPs ≥5% in frequency. Models in the case-control study were adjusted for smoking status (never/former/current), age, and pack-years of smoking. Models in the case-only study were adjusted for age and pack-years of smoking. Due to the small number of never smokers in the case-only study, associations with TP53 mutation were not adjusted for smoking status. Results in the case-only study were similar when adjusted for former or current smoking. Participants with missing values for any of the variables in a regression model were omitted from the analysis. Tests for trend were conducted by calculating P values for the β coefficient in unconditional logistic regression models with TP53 genotype combinations coded as ordinal variables (P_trend).

Survival analysis was done for single TP53 SNPs in the case-control study using COX proportional hazard modeling (Proc PHREG) in SAS. Models were adjusted for smoking status, pack-years of smoking, age, and stage (II-IV versus I). Results were similar in crude and adjusted models. Tests for violation of COX proportional hazards assumption were done by estimating the P value for interaction of individual TP53 polymorphisms with time. All P values were nonsignificant (P > 0.05), except for TP53_12 in African Americans. None of the haplotype associations with survival reported included this polymorphism.

Haplotype analysis was done using individuals with ≥90% completeness of genotyping data for the polymorphisms selected and ≥5% frequency in our study population (African Americans or Caucasians). Haplotype blocks were determined in Haploview (30) using the block definitions from Gabriel et al. (31). Global permutation tests of association (1,000 permutations) of TP53 haplotypes with lung cancer or TP53 mutation were done using haplo.cc in haplo.stats module, implemented in R (32). Adjusted ORs for the association of individual TP53 haplotypes associated with lung cancer or TP53 mutation were estimated using haplo.glm in haplo.stats. Haplotype associations with lung cancer or all-cause survival, hazards ratios (HR), and 95% confidence intervals (95% CI) were determined using THESIAS (33), which did COX proportional hazards regression. All haplotype analysis assumed additive effects of haplotypes.

Fourteen polymorphisms in TP53 were genotyped in two study populations, an ongoing case-control study (25, 26) and a case-only study described previously (24, 34). No variation was observed for TP53_25 (Ex11 +567, G>A) or TP53_64 (Ex6 −24, G>A) in our case-control or case-only study (data not shown). The positions of the TP53 polymorphisms are shown in Fig. 1A. The frequencies of all polymorphisms were similar in the Caucasian lung cases in both studies (P > 0.05, for all). All polymorphisms were in Hardy-Weinberg equilibrium in African American and Caucasian controls, except for TP53_09 (IVS1 −112 G>A) in African American controls (P = 0.014). This violation is unlikely due to genotyping error because the concordance between duplicate samples was 98%, the violation was not observed in Caucasians, and samples were randomly distributed when genotyped. The violation may be partially driven by the rarity of this polymorphism. Given the violation, results using TP53_09 are not reported. The frequencies of the TP53_01 Arg⁷²Pro (35-42) and the TP53_16 (MspI, IVS6 +62 G>A; refs. 35, 39, 43) polymorphisms in our population were consistent with previous reports in African Americans and Caucasians (Supplementary Table S2). The frequency of all TP53 polymorphisms examined, except TP53_11 (IVS7 +92, T>G) and TP53_14 (1474 3′STP, C>T), were significantly different in African American and Caucasian controls (P < 0.001 to P < 0.01).

Hospital controls were younger than population controls and lung cancer cases (P = 0.005); lung cancer cases (P < 0.0001) and hospital controls (P = 0.026) had a lower proportion of African Americans than population controls; and lung cancer cases were more frequently current smokers (P < 0.0001) and smoked a higher number of pack-years (P < 0.0001; Table 1). Frequency matching was incomplete for some variables. The frequencies of all TP53 polymorphisms were similar in both control groups (data not shown) and because both control groups represent the study base from which cases were derived, all further analyses were done comparing lung cancer cases with total controls adjusted for age, smoking status, and pack-years of smoking.

The association of the individual TP53 polymorphisms with lung cancer in African Americans and Caucasians was examined (Table 2). Among African American participants, individuals with TP53 codon 72 Arg/Pro + Pro/Pro genotypes in comparison with those with TP53 Arg/Arg⁷² genotypes and individuals with TP53_11 T/G + G/G genotypes (IVS7 +92) versus TP53_11 T/T genotypes had a 2-fold increased odds of lung cancer. No other individual TP53 polymorphisms were associated with lung cancer in African Americans or Caucasians (Table 2).

Haplotype analysis was done by examining the block structure of TP53. Two linkage blocks were observed in Caucasians and one in African Americans (Fig. 1B). The LD structure of TP53 differed in African Americans and Caucasians. Block I encompassed slightly different SNPs for these ethnic/racial groups (Table 3). African American participants with Pro-T-A-G haplotypes (TP53_01, TP53_65, TP53_66, and TP53_16) had increased odds of lung cancer compared with participants with Arg-T-A-G haplotypes. In the analysis of single TP53 SNPs associated with lung cancer, TP53_01 and TP53_11 were both associated with lung cancer in African Americans (Table 2). Haplotype analysis in African Americans was also done by extending the block I definition to include TP53_11. African American participants with Pro-T-A-G-G haplotypes had almost a 2-fold increase in the odds of lung cancer compared with those with Arg-T-A-G-T haplotypes (Table 3). Using this more precise haplotype definition resulted in a stronger association with lung cancer in African Americans. No other haplotypes in the linkage blocks were associated with lung cancer risk in either African Americans or Caucasians.

The average amount of follow-up time for lung cancer patients in the case-control study was 26.1 months. Individual TP53 polymorphisms were studied for association with lung cancer prognosis (Table 4). TP53_11 in African American lung cancer cases was predictive of worsened lung cancer prognosis or worsened all-cause mortality. TP53_15 was predictive of improved lung cancer prognosis in Caucasians or all-cause mortality. None of the other individual polymorphisms were associated with survival. Results were similar in unadjusted models (data not shown) or in models examining all-cause mortality instead of specifically lung cancer death (Table 4).

Given the observed associations of individual polymorphisms, we examined the association of combined TP53 haplotypes with lung cancer survival using both the block definition and adding TP53_11 based on association with risk and survival (Table 5). The same haplotypes, Pro-T-A-G (TP53_01, TP53_65, TP53_66, and TP53_16) or Pro-T-A-G-G (TP53_01, TP53_65, TP53_66, TP53_16, and TP53_11), which were associated with lung cancer risk in African Americans were associated with worsened lung cancer prognosis in both crude (data not shown) and adjusted models.

To explore possible phenotypic effects of these polymorphisms, we examined the correlation between the TP53 polymorphisms with a phenotypic measure—the frequency of somatic TP53 mutations in lung tumors in a case-only study. Analysis of the case-only study was limited to Caucasians (n = 188) due to the small number of African Americans in the study. The characteristics of the Caucasian subset of the case-only study were similar to the population as a whole and results were similar when African American cases were included in the analysis (24, 34; data not shown). The average age was 65.3 ± 9.3 years. One third of the cases was female (n = 61). The population was mostly former (n = 90, 48%) or current (n = 92, 49%) smokers with an average number of pack-years of 60.0 ± 34.2.

Several of the individual germ-line TP53 polymorphisms were associated with increased odds of somatic TP53 mutations, including TP53_34, TP53_01, TP53_16, and TP53_14 (Supplementary Table S2). All of the ORs, assessing the association of TP53 polymorphisms with the presence of somatic mutations, were above 1. Haplotypes formed from block I SNPs (C-Pro-A-G versus G-Arg-A-G), block I SNPs plus TP53_11 (C-Pro-A-G-T versus G-Arg-A-G-T), and the combination of the TP53 SNPs associated with somatic TP53 mutations (TP53_34, TP53_01, TP53_16, and TP53_14; C-Pro-G-T versus G-Arg-G-C) were associated with somatic TP53 mutations in lung tumors (Table 6). Notably, the block I haplotypes associated with somatic TP53 mutations in lung tumors in Caucasians overlapped with the block I haplotypes associated with lung cancer in African Americans.

Susceptibility to lung cancer caused by smoking may be due, in part, to common genetic variation and other risk factors. In this study, we investigated the hypothesis that TP53 haplotypes modulate mechanisms of lung carcinogenesis and lung cancer susceptibility or prognosis by studying 14 polymorphisms in TP53 in a case-control and case-only study of lung cancer. African Americans with Pro-T-A-G-G haplotypes of the combined TP53 polymorphisms TP53_01 (rs1042522), TP53_65 (rs9895829), TP53_66 (rs2909430), TP53_16 (rs1625895), and TP53_11 (rs12951053) had an increased odds of lung cancer and worsened prognosis compared with those with Arg-T-A-G-T haplotypes. None of the individual TP53 haplotypes was associated with lung cancer in Caucasians. Several TP53 SNPs and haplotypes were associated with an increased frequency of somatic TP53 mutations in lung tumors.

African American participants with Pro-T-A-G-G (TP53_01, TP53_65, TP53_66, TP53_16, and TP53_11) haplotypes had an increased odds of lung cancer and worsened lung cancer survival when compared with those with Arg-T-A-G-T haplotypes. A few previous studies examined the association of TP53 haplotypes with lung cancer, focusing on three polymorphisms, TP53_166 (16-bp insertion, IVS3 +41), TP53_01 (Arg⁷²Pro), and TP53_16 (G>A, IVS6 +62; refs. 37, 39). In one study, Pro⁷² genotypes were only associated with lung cancer in combination with the 16-bp insertion in intron 3 (39). In another study, several haplotypes, or combinations of susceptibility alleles, were associated with lung cancer in Caucasians (37). Despite differences in the alleles examined in these studies and our report, given the observed linkage between the three polymorphisms (35, 37, 39, 44, 45) and the position of the haplotype on TP53 associated with lung cancer in our study, our results and previous studies suggest that this region of TP53 may have a role in modulating lung cancer susceptibility. No previous studies examined the association of TP53 haplotypes with lung cancer survival.

Our study is the first to describe the association of TP53 haplotypes with lung cancer in African Americans. There are several possible explanations for the observed association only in African Americans; however, the reason is currently unknown. The association only in African Americans could be attributed to the higher frequency of particular haplotypes in African Americans (Table 3). However, given the larger number of Caucasians in our study, the power to detect associations was similar among African Americans and Caucasians. Another possible explanation for the differential effects of the TP53 polymorphism on lung cancer could be related to differences in exposure because interaction of TP53_01 with cigarette smoking was observed in several previous studies (46-48), although an interaction of TP53_01 with smoking and lung cancer was not observed in this analysis. One possible exposure difference is menthol cigarette smoking. Menthol cigarette smoking is more prevalent in African Americans (69% versus 22%; ref. 49). Some studies suggested that smoking menthol cigarettes results in smoking larger puff volumes and higher nicotine exposures (50), and therefore, increased exposure to tobacco-specific carcinogens in African Americans. Finally, these results could be due to LD with another polymorphism, such as a rare TP53 polymorphism. Pro⁴⁷Ser (<5%) was observed only in African Americans and not observed in Caucasians (51, 52). The Ser⁴⁷ variant of p53 had a reduced ability to transactivate p53 target genes, p53AIP1 and PUMA, and induce apoptosis (52). The Pro⁴⁷Ser was shown to be in LD with the Arg⁷²Pro polymorphism of TP53 (52), and therefore, could be in LD with the haplotype associated with lung cancer in our study.

Several previous studies examined the association of the TP53 Arg⁷²Pro polymorphism with lung cancer (see refs. 17-19 for review). In a 2003 meta-analysis, overall, the authors observed no significant association of the TP53 Arg⁷²Pro polymorphism with lung cancer (19). However, in the combined analysis, the point estimate for the OR for individuals with Pro/Pro genotypes was elevated. Moreover, in studies with larger population sizes (N = 900-2,500) and some more recent studies not covered by the meta-analysis, individuals with Pro genotypes had an increased odds of lung cancer (38, 41, 46-48, 53). These observations are consistent with our observed association in African Americans. Few previous studies examined the association of TP53 polymorphisms with lung cancer in African Americans, and these studies were quite small (N = 25, 70, 121, 141; refs. 36, 37, 54, 55). In our study, only one haplotype (>1% frequency) containing Pro⁷² allele was associated with lung cancer and the Pro⁷² allele mapped to several additional common haplotypes, which were not associated with lung cancer (Table 2). Therefore, in our study, only examining the Pro⁷² allele imprecisely estimates the relationship between TP53 haplotypes and lung cancer, possibly partially explaining the observed inconsistency in its associations with lung cancer (19).

As an extension of our previous study of TP53_01 (Arg⁷²Pro) polymorphism with somatic TP53 mutations, several additional TP53 polymorphisms (TP53_34, TP53_16, and TP53_14) were associated with an increased frequency of somatic TP53 mutations in lung tumors. Haplotypes of these combined polymorphisms were also associated with somatic TP53 mutations. Several germ-line TP53 polymorphisms (including TP53_34, TP53_01, TP53_16, and TP53_14) and haplotypes were associated with an increased odds of somatic TP53 mutations in lung tumors. Our results suggest a combined influence of TP53 alleles and need to be confirmed in additional studies.

The region of TP53 between TP53_01 (Arg⁷²Pro in exon 4) to TP53_16 (IVS6 +62) contained TP53 germ-line polymorphisms and haplotypes associated both with lung cancer in African Americans and with somatic TP53 mutations in lung tumors from Caucasians (Fig. 1A). This genetic region was identified in two different ethnic populations and associated with different outcomes, suggesting that a locus, or loci, within this region of TP53 may be important in modulating lung cancer susceptibility. Consistent with this notion, several studies showed that the Arg⁷² versus Pro⁷² alters p53 function (17, 18, 20-23) and mutation in TP53 often occurs in exons 5 to 8 in lung tumors (56).

Our case-control study was designed using two separate control groups (hospital and population based). Due to the limitation of our population size, in this analysis, we combined the control groups and did comparisons between lung cancer cases and total controls. In contrast to the population controls, hospital controls had a higher proportion of smokers. Moreover, controls recruited from hospitals have a higher frequency of illness than population controls. Combining these heterogeneous populations (population plus hospital-based controls) could result in biased associations because the distribution of risk factors for lung cancer (such as smoking) differs in these two populations. We believe that this has not affected the results presented in this study because the frequency of TP53 polymorphisms was similar (not statistically different) in these two populations, and analyses were adjusted for smoking. Another source of heterogeneity between these two populations is the difference in race distribution in hospital and population controls. This potential source of bias was likely addressed in the stratification of all analysis by race. In addition, we feel that both study populations represent the study base from which the lung cancer cases were derived.

Our study has several limitations. Given that our case-control associations were only observed in one ethnic group, it is possible that the association may have been observed by chance. Although it is always formally possible that an association is driven by chance, given the biological plausibility of the association and consistency of the association with the observations in our case-only study, it is important to pursue these observations in follow-up studies. The TP53 gene, as was reported previously (57), had low LD across the entire gene, resulting in many rare haplotypes. Therefore, our study had limited power to examine smoking interaction with TP53 haplotypes. We observed an association of TP53 haplotypes with lung cancer and prognosis, only in African American participants, but we were unable to examine the association of TP53 polymorphisms with somatic TP53 mutations in lung tumors from African Americans due to the small number of African American cases in our case-only study. Further studies will need to investigate the association of TP53 haplotypes with lung cancer and with somatic TP53 mutations in lung tumors in African Americans.

In conclusion, our study examined the association of TP53 haplotypes with lung cancer in African Americans and Caucasians. We also investigated the association of TP53 polymorphisms and haplotypes with somatic TP53 mutations in lung tumors. Our results could suggest that TP53 haplotypes may modulate lung cancer pathways, but results need to be confirmed in further studies.

We thank Raymond Jones, John Cottrell, Audrey Salabes, Leoni Leondaridis, Glennwood Trivers, Andrew Borkowski, Dorota Butkiewicz, William P. Bennett, and Mark J. Krasna at University of Maryland and Baltimore Veterans Affairs and the Surgery and Pathology Departments at University of Maryland Hospital, BVA Medical Center, Sinai Hospital, Union Memorial Hospital, St. Agnes Hospital, Northwest Hospital Center, and Mercy Medical Center for their contributions to this study; Lauren Richey for her work extracting stage data for JHU lung cases; Wolfgang Resch for writing the gene and polymorphism position program; Dorothea Dudek-Creaven for editorial assistance; and Karen MacPherson for bibliographic assistance.