Lung cancer is the leading cause of cancer deaths in the world with over 1,000,000 cases diagnosed every year.1 Human lung cancers are classified into 2 major types: small-cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC), the latter consisting of several types.2 Previously, squamous cell carcinoma was the predominant form of NSCLC, but in the last few decades it has been replaced by adenocarcinoma in the USA3, 4 and in other parts of the world. Furthermore, adenocarcinoma has always been the most common form of lung cancer in women, young subjects and never-smokers5, 6, 7, 8 In some parts of the world, especially the Far East, the incidence of adenocarcinoma is unusually high.9 These findings indicate major smoke exposure, histology, geography and gender-related differences in lung cancers, suggesting that these factors may be related to differences in tumor pathogenesis.
Recently aberrant methylation of CpG islands in the promoter region of tumor suppressor genes (TSGs) has become established as an important epigenetic mechanism for gene silencing.10 Inactivation of multiple TSGs by aberrant methylation is a fundamental process involved in the development of many malignant tumors, including lung cancers. Different tumor types, including NSCLC, have their own specific pattern of methylated genes.11, 12, 13 Assays such as the methylation-specific PCR (MSP) can detect aberrantly methylated DNA in a large excess of unmethylated DNA.12, 14 In our study, we used the MSP assay to analyze the methylation status of 7 genes in over 500 NSCLC samples from 4 different countries. The purpose of our study was to determine whether there was a relationship that existed between methylation and smoke exposure, histologic type, geography, gender and tumor stage in resected NSCLC.
MATERIAL AND METHODS
A total of 514 tumors and 84 corresponding nonmalignant lung tissues were obtained from patients with primary NSCLCs treated by curative intent surgical resection between 1993 and 2000. The tumor samples (Table I) were from the USA (MD Anderson Cancer Center, Houston, n = 106), Australia (Prince Charles Hospital, Chermside, Brisbane, n = 106), Japan (Kyushu University, Fukuoka and Okayama University, Okayama, n = 205) and from Taiwan (Veterans General Hospital, Taichung, n = 97). While we have previously reported the methylation profile of NSCLC cases from Australia,22 for our study we utilized a different set so that all cases could be analyzed at the same time using identical assay conditions. Institutional Review Board permission and informed consent were obtained at each collection site.
Mean age (range) in years–2AD, adenocarcinoma; SQ, squamous cell carcinoma–3Smoking data was not available for 4 cases with adenocarcinomas and 5 cases with squamous cell carcinomas for the Australian cases.
Demographic, clinico-pathologic and smoke exposure data were collected by trained physicians or research nurses at each site. Never smokers were defined as those with lifetime exposure of 100 cigarettes or less. Ever smokers were defined as current or former smokers with lifetime exposure of more than 100 cigarettes. Current smokers were defined as those who smoked within 12 months of tumor development. Former smokers were those who had quit smoking more than 12 months prior to tumor development. Smoking information was available for all countries; however, the duration of smoking cessation of former smokers was only available for 41 USA and 48 Australian cases. Adjacent nonmalignant lung tissue was available from 84 cases, 68 from ever smokers and 16 from never smokers.
DNA extraction and MSP assay
Genomic DNA was isolated from frozen tumor tissue by SDS/proteinase K (Life Technologies, Inc., Rockville, MD) digestion, phenol-chloroform extraction and ethanol precipitation.15
The methylation status of 7 genes that were reported to be frequently methylated in lung cancers was determined by MSP assay14 by using gene specific primers and conditions as described previously.12, 14, 16–21 Briefly, 1 μg of genomic DNA was modified by sodium bisulfite, which converts all unmethylated cytosines to uracils while methylated cytosines remain unchanged. PCR amplification was done with sodium bisulfite-treated DNA as template as described previously by using specific primer sequences for the methylated and unmethylated forms of each gene.14 Sensitivities of the assays, determined as previously described,13, 22 could detect 1 methylated allele in the presence of 1,000–10,000 unmethylated alleles. DNA from peripheral blood lymphocytes and buccal mucosa brushes, each from 10 healthy subjects, along with water blanks were used as negative controls for the methylated genes. DNA from lymphocytes of a healthy volunteer artificially methylated by treatment with Sss1 (New England BioLabs, Beverly, MA), and then subjected to bisulfite treatment, was used as a positive control for methylated alleles. PCR products were visualized on 2% agarose gels stained with ethidium bromide. Results were confirmed by repeat MSP assays after an independently performed bisulfite treatment.
Contingency table analyses were employed to examine the association between exposure and binary response variables. Mantel-Haenszel χ2 tests were used to make inference about associations controlling for effects of other exposure variables. For binary exposure variables, the Breslow-Day test for homogeneity of odds ratios across strata was examined. A multivariate generalized estimating equation (GEE) model23 with unstructured working correlation matrix was fit to examine the pairwise correlation between genes in aberrant methylation.
GEE models with logit link also were employed to examine methylation in nonmalignant compared to malignant tissues while accounting for within-person correlations among those for whom 2 tissue samples were available. Main effects of smoke exposure, gender and geographic region, as well as all orders of interaction among these risk factors, were included in the regression models. Inclusion of interactions results in a stratified analysis similar to the Mantel-Haenszel tests conducted for the uncorrelated responses.
To determine the overall rate of methylation in individual samples, we used the Methylation Index (MI).12 The MI is defined as a fraction that represents the number of genes methylated/the number of genes tested. Mean differences in MI by histologic types and cancer risk factors were examined by employing stratified analysis of variance (ANOVA). The stratified ANOVA was obtained as the main effect for the exposure of interest with main effects and all orders of interaction among other exposures included in the model. GEE with an identity link was used when comparing the MIs between malignant and nonmalignant tissues.
As shown in Table I, there were 355 males and 159 females, for a male to female ratio of 2.2. The number of ever smokers was 373 and the number of never smokers was 132 (smoker to never smoker ratio of 2.8). There were major gender differences in the ratio of smokers to never smokers. In males this ratio was 8.2, while in females it was 0.7 (Table I). However, these ratios varied significantly across countries. In Taiwan the ratio in males was 1.7, while in Australia the ratio in females was 11.5.
Geographic, gender and smoke exposure-related differences in NSCLC types
Of the 514 lung cancers examined, the vast majority (493 of 514; 96%) of the tumors from all 4 countries were either adenocarcinomas or squamous cell carcinomas (Fig. 1a). Thus, we limited the correlations of gene methylation with histologic type to these 2 dominant types. Gender and smoke exposure status significantly influenced the adenocarcinoma to squamous cell carcinoma (AD:SQ) ratio. In males the AD:SQ ratio was unity (1 to 1), while in females it was much higher (4.5 to 1) (Fig. 1b). The gender differences were particularly striking in never smokers, where the AD:SQ ratio in males was 2.4 to 1, while in females it was 14.3 to 1 (Fig. 1b). Few subjects with squamous cell carcinoma were never smokers. Therefore, analyses that compared histological types were restricted to those who had a history of smoking. Similarly, analyses that compared smoke exposure were restricted to subjects with adenocarcinomas. For the reasons described previously, correlations of methylation with smoke exposure, histologic type, geography and gender were restricted to subgroups as described below.
Rate of aberrant methylation in NSCLC
We analyzed 514 resected tumors and 84 corresponding nonmalignant lung tissues to determine the aberrant methylation of p16, RASSF1A, APC, RARβ, CDH13, MGMT and GSTP1 by MSP assay. Representative examples of the assays are illustrated in Figure 2 and the results are summarized in Figure 3 and discussed below. Analysis of all the cases indicated that 5 genes (p16, RASSF1A, APC, RARβ and CDH13) were frequently methylated (greater or equal to 25%), while 2 (MGMT and GSTP1) were less frequently methylated (less than 10 %).
The multivariate analysis of gene methylation indicated no strong correlations between gene pairs. Some gene pairs demonstrated modest correlations. Correlations that exceeded 0.15 were observed for CDH13 gene with 3 other genes: P16 (r = 0.22), RASSF1A (r = 0.19), and RARβ (r = 0.23). In addition, methylation of RASSF1A gene was correlated with methylation of APC gene (r = 0.30).
Methylation in nonmalignant lung tissues
For 84 resected cases where corresponding nonmalignant lung tissues were available, methylation of 6 of the studied 7 genes was present in relatively small numbers of cases: APC 12 cases (14%), RASSF1A 5 cases (6%), CDH13 5 cases (6%), p16 4 cases (5%), RARβ 2 cases (2%), and MGMT 1 case (1%). In all cases, the corresponding tumors had methylation of the same genes. These data, broken down by smoking status, are presented in Figure 4a and are discussed below in the section on smoking. With 1 exception, the differences in frequency of methylation and the MI were significantly higher in the corresponding tumor tissues (<0.001 by McNemar's test) except for MGMT, which had a low frequency of methylation in tumor tissues. GSTP1 was not methylated in any of the nonmalignant tissues examined. For further correlations of methylation and smoke exposure, histologic type, geographic region and gender, we restricted the analyses to the appropriate subgroups as described below.
Correlation between aberrant methylation and smoke exposure
Because adenocarcinoma was by far the dominant type in never smokers, we limited our analyses correlating methylation with smoke exposure to patients with this type (Fig. 4b). The rates of p16 and APC genes methylation and the MI were significantly higher in adenocarcinomas arising from ever smokers than from never smokers (Fig. 4b). Breslow-Day tests of homogeneity of the odds ratios indicated that no other risk factors affected the relationship between methylation and smoke exposure for any gene. We also compared the methylation profiles of never, former and current smokers in adenocarcinoma. Analysis was limited to 41 USA and 48 Australian cases from whom detailed information about time of smoking cessation were available. The methylation of APC gene (p = 0.023) and MI (p = 0.023) differed for contrasts between former and current smokers. Methylation rates did not differ between never smokers and former smokers (Fig. 4c).
Of the 84 nonmalignant tissues examined (see above), the smoking status of the subjects was identified for all cases (Fig 4a). With the exception of CDH13, the methylation frequencies of all genes and the MI were higher in the nonmalignant tissues of smoker subjects than in those of never smokers. Because of the small numbers analyzed, these differences did not reach statistical significance.
Methylation and pathologic correlations
Because there were few never smokers with squamous cell carcinoma, we restricted our analysis of the effect of histologic type to those who indicated they were ever smokers. The methylation rates of APC, RARβ, CDH13 and the MI were significantly higher in adenocarcinomas than in squamous cell carcinomas. Mantel-Haenszel odds ratios that control for effects of smoking status, gender and geography were consistent in direction with the methylation rates shown in Figure 5. The Breslow-Day test for homogeneity of the odds ratios was non-significant for all genes.
Geography-related differences in aberrant methylation
We examined the methylation rates of the 7 genes studied and the mean methylation indices by country of origin. Preliminary analyses indicated no significant differences for any of these parameters between the USA and Australia and between Japan and Taiwan (data not shown). For further comparisons, we combined the data from the USA and Australia (group A) and those from Japan and Taiwan (group B). Aberrant methylation of MGMT and GSTP1 genes were significantly more frequent in group A than in group B (Fig. 6) in the analysis stratified on gender and smoke exposure. Furthermore, the value of the mean MI of group A was significantly higher than that of group B.
Correlation of methylation with gender and tumor stage
There were no gender-related differences in the rates of methylation of the 7 genes or of the mean MIs, independent of smoke exposure or histologic type. There was no correlation between aberrant methylation of any of the 7 genes or the MI and tumor stage (data not shown).
There are many reports that aberrant promoter methylation plays an important role in the pathogenesis of malignant tumors including lung cancer.12, 17, 24 We examined the methylation status of 514 lung cancers from 4 countries: the USA and 3 Pacific rim countries, Australia, Japan and Taiwan. Adenocarcinoma was the most frequent form of NSCLC, especially in women and never smokers, and demonstrated modest geographic variations in the relative proportion of cases. As expected, the adenocarcinoma to squamous cell carcinoma ratio was much higher in women and in never smokers.
One of the most interesting aspects of our data is the finding that the rates and pattern of methylation of some genes varied with smoke exposure. The rates of methylation of p16 and APC genes and the MI were higher in ever smokers compared to never smokers. These findings should prompt further study of the dependent effect of smoke exposure to aberrant methylation of specific genes. Kim et al.25 have previously described in NSCLC a higher frequency of methylation of p16 in smokers. However, others have suggested that p16 is inactivated frequently in NSCLC tumors arising both in ever and never smokers, although the pattern of inactivation may vary.26 Promoter methylation can be modulated differentially, depending on carcinogen exposure.27 In addition, a disproportionally high number of mutations in p53 (and other genes) are found at specific methylated CpG dinucleotide sites28, 29 and appear to be related to carcinogen exposure. Thus, the distinctive pattern of aberrant methylation as well as the mutational spectrum of p53, RAS and other genes in lung cancers arising in smokers may be related to specific types and duration of carcinogen exposure. The pattern of allelic losses also supports the notion that lung cancers in never smokers arise through genetic alterations distinct from the common events observed in tumors from smokers.26
Our data that current smokers had a higher mean MI value and higher rate of APC methylation than former smokers indicate that aberrant methylation is not only related to smoke exposure but also decreases in frequency to that of never smokers after smoking cessation. Smoking cessation avoids some of the subsequent risk of lung cancer, and stopping before middle age avoids more than 90% of the risk attributable to tobacco.30 Smoking cessation is associated with changes in the histologic type of the subsequent cancers and with a decrease in the proliferative rate of the bronchial epithelium.31, 32 However, advanced dysplastic bronchial lesions and allelic losses persist in the bronchial epithelium of former smokers.33, 34 The decrease in the overall methylation rate (as determined by the Methylation Index) on smoking cessation is consistent with methylation being a reversible epigenetic change that does not alter the coding sequence of DNA.
Our study demonstrated important differences in the methylation patterns of adenocarcinomas and squamous cell carcinomas in ever smokers. We have previously described methylation differences between the 2 major types of NSCLC cases not stratified smoke exposure.12 While the overall trends in the 2 studies are similar, the significance of some of the differences are different in the current study which stratified for smoke exposure. The methylation rates of APC, RARβ and CDH13 genes, and the mean MI, were significantly higher in adenocarcinomas. Other molecular differences have been described between the 2 major forms of NSCLC, including allelic losses at chromosome 3p and other sites,35, 36, 37p53 mutations38 and RAS mutations.39
Based on our preliminary findings, we could combine the 4 countries into 2 groups (group A, USA and Australia and group B, Japan and Taiwan). Of the 7 genes tested, the methylation rates of 5 genes (p16, RASSF1A, APC, RARβ and CDH13) were greater or equal to 25% in both groups. The methylation rates of the other 2 genes tested, MGMT and GSTP1, while lower than 20%, were significantly greater in group A (15% for MGMT and GSTP1) than in the group B (1% for MGMT and GSTP1). The reasons for these differences are not known but may reflect geographic, cultural or racial differences in carcinogen exposure or molecular pathogenesis.
Some studies indicate that the odds ratios for developing lung cancer after smoking are consistently higher for women than for men,40 although more recent studies have not confirmed these findings.41 In addition, there are major gender-related differences in histologic patterns of lung cancer42 in bronchial epithelial changes associated with tobacco smoking,33 for lung CYP1A1 expression and DNA adduct levels among lung cancer patients,43 and for smoke-related expression of growth factors in the lung.44 However, we did not detect any sex-related differences, independent of smoke exposure, between smokers and never smokers for aberrant methylation of the 7 genes tested or for the MI.
As previously reported by us in a different subset of tumors,12 we found occasional methylation of 6 of the 7 genes tested in adjacent nonmalignant lung tissue. However, with 1 exception, the frequencies of methylation were significantly higher in tumor tissues. There were no significant differences between ever and never smokers, although a trend towards higher methylation frequencies in smokers was noted. In all cases the corresponding tumors had methylation of the same genes. While these data suggest that methylation changes precede the development of invasive cancer, we cannot exclude the possibility that occult tumor cells were present in the lung tissue samples.
Our study represents one of the largest molecular-based analysis of lung cancers performed to date. While there was considerable overlap in the overall patterns of methylation, we found significant differences in the methylation patterns of lung cancers related to smoke exposure, histologic type and geography. Our results will help us understand the impact of these factors on the pathogenesis of lung cancers.
We thank Dr. Keizo Sugimachi, Department of Surgery and Science, Graduate School of Medical Sciences, Kyushu University, Fukuoka, Japan for providing some of the NSCLC tumor samples.