Aberrant Promoter Methylation Profile and Association with Survival in Patients with Non–Small Cell Lung Cancer

Both in the United States and around the world, lung cancer is the leading cause of cancer death in men and women, with a 5-year survival rate of only 15%, a statistic that has changed very little over the past two decades (1). This dismal survival could be improved through earlier detection or through identification of prognostic markers, which could identify subsets of patients with worse prognosis who might benefit from a more aggressive treatment strategy.

It is now well established that epigenetic changes (i.e., heritable changes in gene expression without alterations in the primary DNA sequence of a gene; refs. 2, 3) play an important role in cancer development. The major epigenetic aberration in cancer development is inactivation of tumor suppressor genes (TSG) through hypermethylation of CpG islands in their promoter regions. The spectra and frequency of TSG inactivation at the CpG islands vary from cancer to cancer (3–5). The distinct profile and varying levels of TSG methylations among different cancers, coupled with a sensitive methylation-specific PCR (MSP) technique, have shown promise as a tool for the early detection and diagnosis of cancer and for determining prognosis (5–9).

A number of genes are heavily methylated in non–small cell lung cancer (NSCLC), (4, 10–13) which account for 80% of all lung cancers. As with other cancers, hypermethylation may affect NSCLC tumor behavior and therefore modulate clinical outcome. This, plus the fact that hypermethylation has been detected in the body fluids, including sputum, serum, and plasma, of NSCLC patients, suggests the makings of an important tool for early detection and prediction of outcome (14–16). However, although the clinical significance of the hypermethylation of TSGs in lung cancer has been examined in a number of studies, most of these studies have used MSP and focused on single genes in a small cohort of patients (11, 17–31). The results have been inconsistent, with each study showing a significant association between a single TSG methylation and survival (11, 17–31). In addition, most of these reported associations with NSCLC, with the exception of p16 and RAS association domain family 1A (RASSF1A), have not been reproduced in a second independent study. Thus, there needs to be more clarity regarding the associations between the methylation profile of NSCLC and clinical outcome.

In this study, we used quantitative MSP (QMSP) to determine the methylation frequencies of nine genes [p16, RASSF1A, E-cadherin (CDH1), tissue inhibitor of metalloproteinase-3 (TIMP3), adenomatous polyposis coli (APC), death-associated protein kinase (DAPK), fragile histidine triad (FHIT), O⁶-methylguanine-DNA-methyltransferase (MGMT), and glutathione S-transferase P1 (GSTP1)] in 155 NSCLC patients. These nine genes included all those whose hypermethylation has been previously shown to be associated with the survival of NSCLC patients (17–31). We then, individually and jointly, evaluated their prognostic value.

Study population. The study population consisted of 155 patients with newly diagnosed stage I to III NSCLC who were treated surgically and for whom frozen tumor tissues were archived. Patients were accrued between 1994 and 2001 at The University of Texas M.D. Anderson Cancer Center in Houston, TX. Demographic, epidemiologic, and clinical information, including survival information, were obtained from the computerized patient history database, from chart review, or from the tumor registry at M.D. Anderson. Ever smokers were defined as former and current smokers with a lifetime exposure of >100 cigarettes. Former smokers were defined as those who had quit smoking >12 months before tumor diagnosis. Study approval was obtained from the M.D. Anderson Institutional Review Board.

DNA extraction, bisulfite treatment, and QMSP. Frozen tissues were homogenized, and genomic DNA was extracted by the classic method, consisting of SDS/proteinase K digestion, phenol-chloroform extraction, and ethanol precipitation. After this, 2 μg of genomic DNA from each tumor specimen was denatured by NaOH and treated with sodium bisulfite (Sigma-Aldrich, St. Louis, MO). DNA was then purified with Wizard DNA purification resin (Promega, Madison, WI), treated again with NaOH, precipitated with ethanol, resuspended in 40 μL of distilled nuclease-free water, and stored at −80 °C until use. The modified DNA was used as a template for QMSP, which was done as described previously, with slight modification (32–34). Briefly, for each gene of interest, a pair of primers and a probe were designed specifically around regions with several potential CpG methylation sites. For the internal reference gene, β-actin (ACTB), the primers and probe were designed to avoid CpG nucleotides. All primers and probes were designed to amplify bisulfite-converted DNA sequences. However, the amplifications of genes of interest are proportional to the degree of CpG methylation, whereas the amplification of ACTB is independent of its methylation status. Parallel real-time PCRs were done for the gene of interest and for the ACTB reference gene. The methylation ratio of the particular gene was calculated as the ratio of the fluorescence emission intensity for these two reactions multiplied by 100 (target gene / ACTB × 100). The primer and probe sequences have been described elsewhere (33, 34). Quantitative real-time PCRs were done in 384-well plates in a 7900 HT Taqman sequence detector system (Applied Biosynthesis, Foster City, CA). Each amplification mix (15 μL) contained sample DNA (1 μL), 1× Taqman buffer A, deoxynucleotide triphosphates (200 mmol/L), MgCl₂ (5 mmol/L), AmpliTaq Gold (0.65 unit), each primer (600 nmol/L), and 200 nmol/L of probe. The thermal cycling conditions consisted of one cycle for 10 min at 95°C followed by 50 cycles for 15 s at 95°C and for 1 min at 60°C. All reactions were done in duplicate. The specificity and sensitivity of methylation reactions were controlled for in each analysis by including an unmethylated control DNA (human sperm DNA) and a CpGenome universal methylated DNA (Chemicon International, Temecula, CA). Each plate consisted of tumor samples and multiple negative and positive controls. Serial dilutions of bisulfite-treated CpGenome universal methylated DNA were used for constructing relative standard curves for each gene.

Statistical analysis. The frequency of methylation of each gene was determined by choosing a specific cutoff point of the ratio (target gene / ACTB × 100) for each gene to dichotomize the patients into methylated (yes) and unmethylated (no) groups. Most published studies have used cutoff points to dichotomize subjects into two categories: methylated and unmethylated. Dichotomization moderates the quantitative effect of gene loci with different levels of hypermethylation and amplification efficiency. In addition, by dichotomizing subjects into two groups, we are able to calculate a methylation index for each tumor and perform Kaplan-Meier survival analysis by methylation status. Because the methylation frequency and amplification efficiency vary significantly among different genes, the choice of cutoff points is somewhat arbitrary (23, 25, 33–37). We set our cutoff points to match expected methylation frequencies to allow a direct comparison of our results with previous results obtained from MSP. Although the methylation frequencies of some genes in NSCLC varied in literature, several well-done large studies yielded similar methylation frequencies for most genes (4, 10–13, 25, 26). The frequencies obtained from this study after dichotomizing using the following specific cutoff points were consistent with these major publications (4, 10–13, 25, 26). These cutoff points were 10 for RASSF1A; 2 for CDH1; 1 for p16, TIMP3, APC, and DAPK; and 0.01 for FHIT, MGMT, and GSTP1. We tried different cutoff points and found that the hazard ratios (HR) did not change much. We also tried a post-data mining tool (the classification and regression tree or CART analysis) to select a best cutoff point for each gene to predict survival. Again, the results were similar to what we reported in the article, although the HRs were a little higher for each gene. To determine the overall level of methylation in individual tumors, we used the methylation index (MI), as previously described (12). The MI for each sample was defined as the ratio of the number of methylated genes to the number of genes tested (nine in this study). The χ² test and Fisher's exact test were used for comparison of patient characteristics and distributions of methylated genes by vital status. Mean differences in MI by selected epidemiologic (gender and age) and clinical characteristics (histology and stage) were determined by Wilcoxon rank-sum test. HRs for risk of death were estimated from a multivariate Cox proportional hazards model, with adjustments for age, sex, ethnicity, smoking status, tumor grade, tumor stage, and histology. We also adjusted for pack-years, and the results were similar to those adjusted for smoking status only. We chose to present results adjusted for smoking status because there were some missing values for pack-years. Overall survival in relation to methylation status was evaluated by Kaplan-Meier survival curves and log-rank tests. The P denoting significance was <0.05. The STATA software (STATA Corp., College Station, TX) was used for all the statistical analyses. We used the bum function in Splus to estimate the false discovery rate (FDR). The Benjamini-Hochberg method was used to calculate FDR-adjusted Ps (38). We set the FDR at the level of 10% and calculated the FDR-adjusted Ps to assess if the resulting Ps remained statistically significant after multiple comparisons were taken into consideration.

Characteristics of patients. One hundred fifty-five patients with NSCLC were included in this study. The median survival time was 29.6 months, and the 5-year survival rate was 54.8%. Table 1 shows how selected patient characteristics were distributed between alive and dead patients. There was no significant difference in age by vital status (mean age = 63.9 ± 9.9 versus 66.4 ± 10.4 years, P = 0.173). As expected, higher stage at diagnosis was the most significant risk predictor for death (P = 0.001) followed by higher grade, but it did not reach statistical significance (P = 0.115). The death rate seemed to be higher in men than in women in this cohort (P = 0.077). Smoking status at enrollment was not significantly associated with survival (P = 0.613). The patients were predominantly Caucasians.

Distributions of individual methylated genes. The frequencies of the promoter hypermethylation of the nine genes are shown in Table 2. The overall frequencies of promoter methylation for each gene were as follows: 21.9% for p16, 27.1% for RASSF1A, 47.1% for TIMP3, 32.9% for CDH1, 67.1% for APC, 53.5% for DAPK, 31.6% for FHIT, 17.4% for MGMT, and 5.8% for GSTP1. We also did detailed stratified analyses to determine the distributions of methylation status according to selected demographic and clinical characteristics, including age, gender, histologic subtype, and stage of the tumor. This analysis revealed that p16 and FHIT methylation were borderline statistically significantly more frequent in men than in women (27.4% versus 15.5%, P = 0.075 and 38.1% versus 23.9%, P = 0.059, respectively). However, the frequencies of methylation in other genes were similar in men and women. When we dichotomized patients by their mean age (66 years), we found that five of the nine genes exhibited higher frequencies of methylation in older patients (≥66 years old) than in younger patients (<66 years old), as follows: p16, 28.0% versus 15.1% (P = 0.051); CDH1, 39.0% versus 26.0% (P = 0.086); TIMP3, 52.4% versus 41.1% (P = 0.158); FHIT, 40.2% versus 21.9% (P = 0.014); and GSTP1, 8.5% versus 2.7% (P = 0.174). There was also a noticeable difference in the methylation status of these genes by histologic subtypes. In particular, four genes exhibited significantly or borderline significantly higher frequencies of methylation in adenocarcinomas than in squamous cell carcinomas (SCC), as follows: TIMP3, 53.2% versus 38.5% (P = 0.099); RASSF1A, 37.7% versus 21.2% (P = 0.047); APC, 75.3% versus 55.8% (P = 0.020); and GSTP1, 10.4% versus 1.9% (P = 0.083). There was no obvious difference in methylation frequencies of any of these genes between early-stage (I + II) and late-stage (III) tumors; however, because there were only 36 stage III tumors, these observations need to be confirmed in a larger study.

Figure 1 illustrates the mean differences in the MI by age, gender, histology, stage, and smoking history. Specifically, the MI was significantly higher in tumors from older patients (0.37 ± 0.20) than younger patients (0.30 ± 0.22, P = 0.040) and in adenocarcinomas (0.38 ± 0.20) than in SCCs (0.30 ± 0.22, P = 0.027). There were no significant differences in MI between men and women (0.35 ± 0.22 versus 0.32 ± 0.20, P = 0.325) and between early-stage tumors (0.34 ± 0.21) and late-stage tumors (0.32 ± 0.24, P = 0.410). Because the patients were predominantly Caucasians (94.8%) and smokers (91.6%), we did not stratify patients by ethnicity and smoking status. However, when we dichotomized ever-smoking patients by their median pack-years (50 pack-years), MI was significantly higher in tumors from patients with >50 pack-years of smoking than in tumors from patients with ≤50 pack-years of smoking (0.39 ± 0.21 versus 0.29 ± 0.20, P = 0.042). Another notable observation was that p16 methylation was seen only in tumors from smokers (data not shown).

Associations between TSG methylations and survival.Table 3 illustrates the overall associations between individual TSG methylation and overall survival, as shown by the Cox proportional hazards model. After adjusting for age, gender, ethnicity, smoking status, tumor stage, grade, and histology, hypermethylation of p16 was associated with a significantly poorer survival [HR, 1.95; 95% confidence interval (95% CI), 1.12-3.39]. On the other hand, hypermethylation of the CDH1 and TIMP3 genes was associated with significantly better survival, with HRs of 0.51 (95% CI, 0.29-0.90) and 0.59 (95% CI, 0.36-0.97), respectively. There were no significant overall associations between methylation of the other TSGs and survival (Table 3). We used the Benjamini-Hochberg method (38) to calculate the FDR, and the estimated FDR for the overall analysis was 12%. After adjusting for the FDR at the 10% level, the associations between p16 or CDH1 methylation with survival remained statistically significant.

We then did stratified analyses by histology, age, and gender (Table 4). This analysis showed that the association between p16 methylation and survival was more evident in SCC (HR, 6.32; 95% CI, 2.12-18.8) than adenocarcinoma (HR, 1.24; 95% CI, 0.59-2.60) and in older (HR, 3.30; 95% CI, 1.54-7.09) than in younger patients (HR, 1.46; 95% CI, 0.50-4.26). The associations between CDH1 and TIMP3 methylations and survival were more evident in younger (HR, 0.14; 95% CI, 0.04-0.48 and HR, 0.27; 95% CI, 0.11-0.67, respectively) than in older patients (HR, 0.57; 95% CI, 0.23-1.38 and HR, 0.75; 95% CI, 0.37-1.52, respectively). In addition, RASSF1A methylation conferred a borderline significantly increased risk of death in patients with adenocarcinomas (HR, 1.91; 95% CI, 0.92-3.98). FHIT methylation conferred a significantly elevated risk of death in patients with SCCs (HR, 2.59; 95% CI, 1.07-6.27) and a borderline significantly elevated risk of death in men (HR, 1.99; 95% CI, 0.95-4.17). No other statistically significant associations were evident (data not shown).

We next did a Kaplan-Meier survival analysis to compare overall survival of patients stratified by gene methylation status. Survival was significantly shorter in patients with a methylated p16 gene (median = 21.7 months) than in patients without a methylated p16 gene (median = 62.5 months; P = 0.0001, log-rank test; Fig. 2A). In contrast, survival was significantly longer in patients with CDH1 methylation (median = 70.7 months) than in patients without CDH1 methylation (median = 36.8 months; P = 0.0038; Fig. 2B). Survival was only borderline significantly longer in patients with TIMP3 methylation (median = 67.7 months) than in patients without TIMP3 methylation (median = 37.4 months; P = 0.068; Fig. 2C).

Finally, based on results from analyses of individual genes, we selected three unfavorable events that were most strong predictors of poor survival: hypermethylation of p16 and low methylation of CDH1 and TIMP3. Joint analysis of these three events showed a significant trend toward poorer survival as the number of unfavorable events increased. That is, the median survival time for patients with 0, 1, 2, and 3 unfavorable events were >70.7, 47.0, 37.8, and 31.4 months, respectively (P = 0.0007, log-rank test for the four groups; Fig. 2D).

We have carefully reviewed the literature on the association between TSG methylation and survival in patients with NSCLC (Table 5). Most studies used MSP and investigated a single or a handful of genes. All studies reported either negative results or a single significant association between a specific TSG methylation and survival. With the exception of p16 and RASSF1A, the other significant results have not been reproduced in a second independent study. In this study, in which we applied a quantitative MSP method, we were able to replicate or extend several previously reported significant associations between TSG methylation and survival, most notably, p16 methylation with significantly poorer survival and CDH1 methylation with a significantly better survival. In addition, we suggested that TIMP3 methylation was associated with significantly better survival. Furthermore, joint analysis of p16, CDH1, and TIMP3 genes showed a significant trend toward poorer survival as the number of unfavorable epigenetic events increased. Three other significant findings were that (a) the MI was significantly higher in adenocarcinoma than in SCC, in older than younger patients, and in heavier than lighter smokers; (b) RASSF1A methylation was a worse prognosis factor in adenocarcinoma; and (c) FHIT methylation was associated with poor survival in SCC. Thus, there were significant histology-, age-, and smoking-related differences in the methylation status of tumors from NSCLC patients.

p16 promoter hypermethylation is a widespread epigenetic alteration and plays a significant role inactivating p16 in many tumor types (4). In particular, p16 methylation is an early event in lung carcinogenesis, occurring frequently in all stages of NSCLC, even in cancer-free individuals exposed to tobacco carcinogens (19, 39). In addition, three independent studies have shown that p16 methylation confers a significantly worse survival in NSCLC patients (19, 27, 28). Although our results were consistent with these reports, there was a discrepancy in the histologic subtype. That is, the large Japanese study showed a significant association of p16 methylation in adenocarcinoma (27); however, we found the association was more dramatic in SCC, which is the type most strongly associated with smoking. Previous data have shown that p16 methylation is strongly linked to smoking; indeed, even a dose effect of smoking on p16 methylation was reported in NSCLC (19, 40). In our study, all patients whose tumors showed p16 methylation were ever smokers. Our data, plus that of others, therefore suggest that p16 methylation affects tumor behavior and clinical outcome through an interaction with tobacco exposure.

E-cadherin protein (encoded by CDH1) is a transmembrane glycoprotein that mediates calcium-dependent cell-cell adhesion. Loss of CDH1 may therefore enhance tumor progression and invasion by multiple mechanisms, including reduced cell-cell adhesion. Indeed, somatic CDH1 inactivation by mutation or promoter methylation is frequent in many human cancers (4, 41). Furthermore, consistent with its biological function, hypermethylation of the CDH1 promoter has been shown to be a predictor of poor prognosis in a number of malignancies, including leukemia (42) and gastric (43), cervical (44), liver (45), and bladder (46) cancers. However, this finding has not been reported in lung cancer. Instead, a previous study, that of Zochbauer-Muller et al. (11) showed that CDH1 methylation was associated with a highly significant better survival time (P = 0.005, log-rank test; ref. 11), which we also observed but in a larger sample size. This seemingly “paradoxical” finding from two independent populations (Australian and the United States) is intriguing. Zochbauer-Muller et al. (11) suggested that it might be due to the heterogeneous and unstable CDH1 methylation that drives metastatic progression, as described by Graff et al. (47) in breast cancer.

Our data also suggested that hypermethylation of the TIMP3 gene may be associated with improved survival in patients with NSCLC. TIMP3 can suppress tumor growth, invasion, and metastasis by inhibiting matrix metalloproteinase activity. In keeping with this function of TIMP3, aberrant methylation of TIMP3 has been reported in primary cancers of the kidney, brain, colon, breast, and lung (48). As with the loss of CDH1, the inactivation of TIMP3 by promoter hypermethylation should therefore theoretically favor tumor progression. Yet, we observed the opposite, although we are not alone in this finding. In a recent bladder cancer study in which the associations between the methylation of a panel of nine genes and superficial bladder cancer recurrence were determined, TIMP3 methylation was also found to be associated with a significantly longer recurrence-free survival (49). Lung cancer and bladder cancer are two malignancies that are strongly associated with tobacco smoking. Therefore, it is conceivable that smoking modifies the effect of methylation of cell adhesion molecules, such as CDH1 and TIMP3, on tumor behavior. In addition, the methylation of cell adhesion genes might have a differential effect on early-stage and advanced tumors. For example, in patients with advanced tumors, the methylation and subsequent loss of cell-cell adhesion might favor tumor progression and metastasis. However, in patients with early-stage tumors, such as those in our study, which consisted mostly of patients with early-stage NSCLC, and in patients with superficial bladder cancer (49), a decrease in cell-cell adhesion may prevent the tumor from metastasizing.

We also found that FHIT methylation was associated with a significantly worse survival only in SCC patients. A previous study showing a significant effect of this gene on NSCLC did not stratify NSCLC according to histologic subtypes, presumably due to the limited number of SCC cases (n = 40; ref. 26). Kim et al. (50) recently showed that FHIT methylation was more frequent in older NSCLC patients and was associated with exposure to smoking (P = 0.001). Tomizawa et al. (51) showed that FHIT methylation was significantly associated with p16 methylation in smokers with SCC. Our observations with regard to p16 and FHIT were consistent with these reports: our Spearman correlation test showed a significant positive correlation between p16 methylation and FHIT methylation (r = 0.188, P = 0.019); methylation of both genes was more frequent in older and in male patients; methylation of both genes was associated with a significantly worse prognosis only in SCC. RASSF1A methylation, on the other hand, exhibited a borderline significant association with worse survival in adenocarcinoma in our study. In addition, we observed that RASSF1A methylation was significantly more frequent in adenocarcinoma than in SCC. There have been similar reports that RASSF1A methylation was associated with significantly worse survival in patients with adenocarcinoma (21), and that RASSF1A methylation was more frequent in adenocarcinoma than in SCC (51). We did not confirm or find other individual significant associations between APC, MGMT, DAPK, or GSTP1 methylation and survival.

In this study, we used QMSP to determine the degree of methylation, a technique that has several advantages: it is more sensitive and specific than MSP assays and also has a high-throughput capability. Numerous studies have validated this method, showing an excellent concordance between the results of QMSP and MSP assays (15, 25, 32, 33, 35, 52, 53). The accuracy of our assays is suggested by the following observations. First, the p16 methylation statuses determined by our QMSP method were in complete concordance with those obtained by another labor-intensive, but more accurate, method (pyrosequencing) in the same batch of samples (data not shown). Second, the MI was significantly higher in adenocarcinoma than SCC, consistent with the findings from the largest study published on this topic (12). Third, we found for the first time that the MI was significantly higher in older NSCLC patients and in heavier smokers, which conforms with the findings of other studies showing that age- and smoking-related methylation of individual genes are frequent in both tumor and normal tissues (12, 39, 54). Fourth, the methylation frequencies reported in this study were similar to those published elsewhere in the literature (4, 10–13, 25). Despite the strong biological plausibility and consistence with the literature for most of the individual associations noted and discussed above, some of these associations may be type I error (false positives) due to multiple testing. As an initial attempt to address the multiple comparison and false-positive issues, we estimated the FDR using the bum function in Splus and obtained a FDR of 12%. The significant associations between p16 and CDH1 methylation with survival remained statistically significant after FDR adjustment at the 10% α level. Stratified associations may be more prone to false positives due to smaller sample size in each stratum and therefore should be interpreted with caution. Larger studies are warranted to confirm these findings.

In summary, our study is one of the larger single-institution studies investigating the association between TSG methylation and survival in patients with NSCLC. In addition, the list of genes examined was by far the most inclusive. This study is also the first to show multiple significant associations between TSG methylation and survival in patients with NSCLC. However, contrary to the generally accepted notion that the methylation of multiple genes promotes tumorigenesis, our study suggests that the effect of promoter hypermethylation on prognosis is more complex. In other words, the simple summation of multiple methylated genes may not have prognostic value because the methylation of some genes confers a worse prognosis, whereas the methylation of other genes is associated with a better prognosis. Careful assessment of the effect of each methylated gene on survival is needed.

We thank Betty Notzon for editing the manuscript.