Smoking influences aberrant CpG hypermethylation of multiple genes in human prostate carcinoma

Authors


Abstract

BACKGROUND

Aberrant CpG methylation profiles of gene promoters and their correlation with advanced pathologic features have been well investigated in prostate carcinoma (PC). Several case–control and prospective studies have revealed a positive association between current smoking and PC. The authors hypothesized that smoking influences both progression and prognosis of PC through CpG hypermethylation of related genes.

METHODS

A total of 164 PC patients (52 current, 30 former, and 82 never smokers) and 69 benign prostatic hyperplasia (BPH) patients were examined by methylation-specific PCR (MSP) for 3 genes: adenomatous polyposis coli (APC), glutathione S-transferase pi (GSTP1), and multidrug resistance one (MDR1). The methylation status of representative samples was confirmed by bisulfite DNA sequencing analysis. The newly defined methylation score (M-score) of each sample is the sum of the corresponding log hazard ratio (HR) coefficients derived from multivariate logistic regression analysis for pathology (BPH vs. PC), and was related to clinical and pathologic outcome including smoking status.

RESULTS

The M-score was significantly higher in the current smokers than in never smokers (P = 0.008). Spearman rank correlation test demonstrated a significant correlation between pack-years smoked and M-score in PCs (P = 0.039). Significant correlation of the M-score methylation was observed with high pT category (P < 0.001), high Gleason sum (P < 0.001), high preoperative prostate-specific antigen (PSA) (P = 0.041), and advanced pathologic features. In addition, Gleason sum was significantly associated with PSA failure-free probability as a poor outcome (P = 0.020).

CONCLUSIONS

This is the first study to demonstrate significant correlation of the methylation status of multigenes with smoking status in PC. Smoking status may influence both progression and prognosis of PC through CpG hypermethylation of related genes. Cancer 2006. © 2005 American Cancer Society.

Prostate carcinoma (PC) is one of the most common malignancies among men.1 Smoking is strongly associated with cancers of the head and neck, esophagus, lung, and urinary bladder.2 Several potential mechanisms whereby smoking may increase risk of PC involve male hormones or cadmium.3–5 Aberrant methylation profiles of gene promoters and their correlation with advanced pathologic features have been well investigated in PC.6–8 In lung cancers, some investigators have previously reported that there were significant correlations between aberrant promoter methylation of genes and smoking status of patients.9–12 However, such studies in PC are lacking in the literature.

Several case–control studies have revealed a positive association between current smoking and PC.3, 13–18 Previous reports also have demonstrated an association between smoking and advanced PC.3, 17–19 Furthermore, the majority of prospective studies that used PC death as an outcome noted positive association between current smoking and PC.4 Recent studies from our laboratory have shown that methylation analysis of the APC, GSTP1, and MDR1 genes correlate with progression of prostate carcinoma.20, 21 When several genes are analyzed in the same samples, careful interpretation of the results is necessary because each gene or other clinical factors, including age, may influence one another.22, 23 We hypothesize that smoking influences both progression and prognosis of PC through CpG hypermethylation of related genes. To test this hypothesis, we used a methylation score (M-score) that is the sum of the log of the hazard ratios (HR) for each gene, analyzed by multivariate logistic regression analysis for pathology (benign prostatic hyperplasia [BPH] vs. PC). We also related the M-score to clinical and pathologic outcome, including smoking status. In addition, we assessed prostate specific antigen (PSA) failure-free probability against the clinicopathologic features of PC.

MATERIALS AND METHODS

Tissue Samples

A total of 164 newly diagnosed PC tissues from radical prostatectomies and 69 pathologically proven BPH samples from transurethral resection (TUR-P) were obtained from Shimane University Hospital (Izumo, Japan). The patients' clinicopathologic characteristics including their smoking status are summarized in Table 1. Each tumor was graded and staged according to the Gleason grading system and the TNM staging system.24, 25 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 before tumor development. None of these patients had received androgen deprivation therapy before radical prostatectomy. We used serum PSA levels after radical prostatectomy as a surrogate end-point, with a level equal to or above 0.2 ng/mL designated as PSA failure. Forty-five patients with PC were excluded from the PSA failure-free probability study because of adjuvant hormonal therapy immediately after radical prostatectomy. PSA failure-free probability was determined as the percentage of patients without PSA failure. Follow-up ranged from 0.7 to 91.4 months, with a median of 33.9 months.

Table 1. Clinicopathologic Characteristics of Prostate Carcinoma according to Smoking Status
 OverallSmokingP value
Current smokersFormer smokersNever smokers
  • PSA: prostate - specific antigen.

  • There are no significance differences in clinical backgrounds such as age, pT category, Gleason sum, and preoperative PSA among smoking statuses.

  • a

    Pack-years smoked is years smoked × cigarettes per day/20.

Total no.1645230820.154
Median age (range)69 (49–87)68 (49–80)71 (62–78)69 (51–80)0.154
Smoke exposure     
 Pack-yrs smokeda40 ± 2642 ± 2437 ± 30 
 Yrs smoked40 ± 1144 ± 834 ± 13 
 Starting age23 ± 823 ± 723 ± 10 
pT category     
 pT100000.590
 pT2108312156 
 pT35220923 
 pT44103 
Gleason sum     
 < 7822319400.495
 75320726 
 > 7299416 
Preoperative serum PSA     
 < 42246120.611
 ≥ 4, < 10772613380.611
 ≥ 1065221132 

Tissue Preparations

All samples were fixed in 10% buffered formalin (pH 7.0) and embedded in paraffin wax. For histologic evaluation, 5 μm-thick sections were used for hematoxylin and eosin (H & E) staining. All samples were microscopically dissected and analyzed for methylation.26 In BPH samples, the presence of high-grade prostate intraepithelial neoplasia (PIN) and carcinoma were ruled out by microscopic analysis.

Nucleic acid extraction

Genomic DNA from all prostate samples was extracted using a commercial kit (Qiagene, Valencia, CA), and precipitated with ethanol. The concentration of DNA was determined with a spectrophotometer, and its integrity was checked by gel electrophoresis.

Methylation Analysis

Genomic DNA from all prostate samples (100 ng) was subjected to sodium bisulfite modification using a CpGenome DNA Modification Kit (Intergen Co., Purchase, NY). Based on the functional promoter sequence of APC,27 methylation-specific polymerase chain reaction (MSP) and unmethylation-specific polymerase chain reaction (USP) primers were designed by using MethPrimer software (http://itsa.ucsf.edu/∼urolab/methprimer) developed in our laboratory.28 Primers used for MSP and USP analysis are as follows: universal primers: 5′-TAATTTTTTTGTTTGTTGGGGATT-3′ (sense), 5′-ACTACACCAATACAACCACATATC-3′ (antisense); MSP primers: 5′-TATTGCGGAGTGCGGGTC-3′ (sense), 5′-TCGACGAACTCCC-GACGA-3′ (antisense); USP primers: 5′-GTGTTTTATTGTGGAGTGTGGGTT-3′ (sense), 5′-CCAATCAACAAACTCCCAACAA-3′ (antisense). An initial polymerase chain reaction (PCR) product was amplified with universal primers, which have no CpG sites in either forward or reverse primers, followed by a second nested PCR with primers specific for MSP or USP. For semiquantitative analysis, a preliminary suitable number of PCR cycles for each MSP and USP were carried out to determine the linear range of the reaction. To ensure this, at least one initial PCR was performed using 32 cycles each for MSP and USP. Then, a suitable PCR cycle was chosen for each sample. The annealing temperature and PCR cycles used for MSP and USP primers was 64 °C and 32 cycles. The sequences of primers for GSTP1 and MDR1 and their PCR conditions have been described previously.20, 21 In each assay, absence of DNA template served as negative control. The obtained MSP and USP products were analyzed by electrophoresis in 3% agarose gels and stained with ethidium bromide. With ImageJ software (http://rsb.info.nih.gov/ij), relative methylation levels (%) were calculated by using the area under the curve corresponding to each band (MSP and USP).20, 21 For methylation analysis of APC, we used 5.3 % methylation as a cut-off value, which was the average percentage of methylation in 69 BPH samples. Using these criteria, MSP positivity for APC was defined as those PCs with a percentage of methylation of more than 5.3%, and negative methylation was less than 5.3%. The criteria for MSP positivity of GSTP1 and MDR1 were described previously.20, 21

Bisulfite DNA Sequencing Analysis

Bisulfite-modified DNA (1 μL) was amplified using a pair of universal primers in a total volume of 20 μL. Direct bisulfite DNA sequencing of the PCR products using either forward universal primer or reverse universal primer was performed according to the manufacturer's instructions (Applied BioSystems, Foster City, CA).

Statistical Analysis

To compare patients' background or methylation frequency for each gene, chi-square test was employed. Using a previously reported analytical technique,23 we calculated the M-score for each sample, defined as the sum of the corresponding log hazard ratio (HR) coefficients, which were derived from multivariate logistic regression analysis of each methylated gene in the BPH and PC samples (Table 2). For each clinicopathologic finding, the association with PSA failure-free probability was determined using Kaplan–Meier curves with a log-rank analysis. The relation between M-score and clinicopathologic findings, except smoking status, was analyzed by either the Mann–Whitney U test, Kruskal–Wallis test, or Spearman rank correlation test. The relation between M-score and smoking status was analyzed by the Bonferroni-adjusted Mann–Whiteney U test. For this comparison test among the three groups of smoking statuses, the nonadjusted statistical levels of significance of P < 0.05 corresponds to a Bonferroni-adjusted statistical significance of P < 0.0167.

Table 2. Multivariate Logistic Regression Analysis of Gene Methylations in the Series of PC and BPH
VariableCoefficientSEChi-squarePaHRb95% CIclog HRd
  • All data are adjusted by age. PC: prostate carcinoma; BPH: benign prostatic hyperplasia.

  • a

    APC, GSTP1, and MDR1 methylation was a significant dependent predictor of pathology.

  • b

    Individual gene hazard ratios for pathogenesis were different from one another.

  • c

    95% CI: 95% confidence interval.

  • d

    M-score was determined as the sum of the corresponding log hazard ratio for pathology.

APC2.0560.52515.338< 0.0017.8132.792–21.8592.056 (0.067)
GSTP11.7730.6447.5860.0065.8911.667–20.8141.773 (0.057)
MDR11.1510.5145.0090.0253.1621.154–8.6641.151 (0.037)

RESULTS

Methylation Status of the APC Promoter in Clinical Samples

Representative results of MSP and USP assays for APC in PCs and BPHs are shown in Figure 1A. MSP-positive bands were present in the majority of PCs, and less so in BPH samples. The result of the methylation study was also confirmed by bisulfite DNA sequencing. Figure 1B shows results of a typical bisulfite DNA sequencing in a PC sample. In sample ‘P’ (corresponding to Fig. 1A, lane ‘P’) with both MSP and USP bands, there was a “T” peak along with a “C” peak at the CpG sites, indicating partial-methylation (Fig. 1B, top). In sample ‘U’ (corresponding to Fig. 1A, lane ‘U’), where no MSP band was observed, the CpG sites were completely unmethylated (Fig. 1B, bottom).

Figure 1.

Methylation status of the APC promoter in clinical samples. (A) MSP and USP bands for 24 samples of BPH and PC are shown. Lanes ‘P’ and ‘U’ represent samples depicting partial methylation and unmethylion, respectively. (B) Bisulfite DNA sequencing of partially methylated (top), unmethylated (bottom) samples correspond to lanes ‘P’ and ‘U’, respectively. In partially methylated samples, there was a “T” peak along with a “C” peak at the CpG sites. In unmethylated samples, every CpG site was unmethylated. PM and UM correspond to partial methylation and unmethylation, respectively.

Evaluation of M-Score; Multigene Methylation Analysis with APC, GSTP1, and MDR1 to Distinguish PC from BPH

APC methylation analysis showed positive MSP bands in 104 of 164 (63.4%) PCs and 6 of 69 (8.7%) BPHs (Fig. 2). As we have reported previously, positive MSP bands for GSTP1 methylation analysis were found solely in 87 of 164 (53.0%) PCs and in 4 of 69 (5.8%) BPHs, whereas, for MDR1 methylation analysis, positive MSP bands were found in 89 of 164 (48.7%) PCs and in 8 of 69 (11.6%) BPHs.20, 21 A significant difference in the methylation status of each gene was found between the series of PCs and BPHs (Fig. 2). As shown in Table 2, multivariate logistic regression analysis revealed that APC, GSTP1, and MDR1 methylation was a significant dependent predictor of pathology (BPH vs. PC) (P < 0.001, P = 0.006, and P = 0.025, respectively). The individual gene hazard ratios for pathogenesis (PC vs. BPH) were different from one another. For instance, cases with APC methylation are 7.813 times more likely to have PC than those with negative methylation, whereas the HR for MDR1 is 3.162. The M-score determined by the sum of log HR was significantly higher in PC than in BPH (Fig. 2). All statistics were age adjusted to eliminate potential influence caused by age.

Figure 2.

Correlation of gene methylation frequency with pathology (BPH vs. PC) and methylation score (M-score). The methylation frequency in PC samples was significantly higher than that in BPH samples for APC, GSTP1 and MDR1 genes (64.1% vs. 8.7%, 54.0% vs. 5.8%, and 55.3% vs. 11.6%, respectively). Right side: M-score, determined as the sum of the corresponding log hazard ratio for pathology (BPH vs. PC) (Table 2), was also significantly higher in PC samples than in BPH samples.

Correlation between Aberrant Methylation and Smoking Status

Looking at PC samples by smoking status, positive MSP bands for APC, GSTP1, and MDR1 methylation analysis were found: in 48 (58.5%), 38 (46.3%), and 38 (46.3%) of never smokers,82 respectively; in 17 (56.7%), 13 (43.3%), and 17 (56.7%) of former smokers,30 respectively; and in 39 (75.0%), 36 (69.2%), and 34 (65.4%) of current smokers,52 respectively (Fig. 3). There was significant difference between smoking status in GSTP1 methylation status (P = 0.018), whereas no significant difference was found in APC or MDR1 methylation status (P = 0.109 and P = 0.094, respectively) (Fig. 3). When we employed the M-score that combined the methylation analysis for three genes, the difference in M-score between current and never smokers was statistically significant (P = 0.008) (Fig. 3, right). Conversely, the difference in M-score was not significant: between current and former smokers (P = 0.046) or between former and never smokers (P = 0.950) (Fig. 3, right). Spearman rank correlation test revealed significant correlation between pack-years smoked and M-score in PCs (P = 0.039). There were no relations between methylation status and other smoking variables, such as duration of smoking and age at starting smoking.

Figure 3.

Correlation of methylation frequency with smoking status and methylation score (M-score). The methylation frequency in current smokers was significantly higher than that in former or never smokers for GSTP1 (69.2%, 43.3%, and 46.3%, respectively). There was no significant difference in APC or MDR1 methylation frequency among smokers (75.0%, 56.7%, and 58.5%, respectively for APC; 65.4%, 56.7%, and 46.3%, respectively for MDR1). Right side: When methylation data was combined (M-score), there was a significant difference in M-score between current and never smokers (3.52 ± 0.27 vs. 2.56 ± 0.23, P = 0.008 by Bonferroni-adjusted test). However, there was only a trend to significance in M-score between current and former smokers (3.52 ± 0.27 vs. 2.59 ± 0.37, P = 0.046 by Bonferroni-adjusted test).

Correlation of M-Score with Clinicopathologic Findings

Among PCs, the M-score showed a significant stepwise increase with advancing pathologic stage (1.34 ± 0.26 in pT2a, 2.97 ± 0.24 in pT2b, 3.64 ± 0.29 in pT3a, 4.07 ± 0.56 in pT3b, and 4.98 ± 0 in pT4) (P < 0.001) (Fig. 4A). Similarly the M-score increased as the Gleason sum (GS) increased (2.20 ± 0.23 in GS < 7, 3.44 ± 0.26 in GS = 7, and 3.74 ± 0.31 in GS > 7) (P < 0.001) (Fig. 4B). For preoperative PSA levels, the M-score was higher in PSA > 10 ng/mL (3.27 ± 0.26) than in PSA ≤10 ng/mL (2.61 ± 0.20) (P = 0.041) (Fig. 4C). Moreover, the M-score was higher in advancing pathologic features as follows: in capsular invasion (Cap) (positive [3.65 ± 0.25] vs. negative [2.48 ± 0.20] [P < 0.001]), in seminal vesicle involvement (SV) (positive [4.71 ± 0.19] vs. negative [2.74 ± 0.17] [P = 0.002]), in pelvic lymph lymph node metastasis (pN) (positive [4.48 ± 0.37] vs. negative [2.55 ± 0.19] [P = 0.002]), in venous involvement (v) (positive [3.79 ± 0.24] vs. negative [2.44 ± 0.20] [P < 0.001]), in lymphatic vessel involvement (ly) (positive [3.70 ± 0.29] vs. negative [2.57 ± 0.19] [P = 0.002]), and in perineural invasion (PNI) (positive [3.29 ± 0.18] vs. negative [1.95 ± 0.29] [P < 0.001]) (Fig. 4C). We also observed that the M-score was age related in the total group of BPHs (P = 0.001) but not in the total group of PCs (P = 0.067).

Figure 4.

Correlation of M-score with clinicopathologic features. Among the overall PC patients: (A) Higher stage of disease (1.34 ± 0.26 in pT2a, 2.92 ± 0.24 in pT2b, 3.84 ± 0.26 in pT3a, 4.21 ± 0.62 in pT3b, and 4.98 ± 0 in pT4); (B) Higher Gleason sum (2.20 ± 0.23 in GS <7, 3.58 ± 0.25 in GS = 7, and 3.819 ± 0.29 in GS > 7; and (C) Higher preoperatve PSA levels (3.35 ± 0.25 in PSA > 10 ng/mL, 2.64 ± 0.20 in PSA ≤10 ng/mL) correlated with an increased methylation score (M-score). There was significant correlation between M-score and worse clinicopathologic findings. Cap: capsular invasion; SV: seminal vesicle involvement; pN: lymph lymph node invasion; ly: lymphatic vessel invasion; and PNI: perineural invasion. N and P correspond to negative and positive, respectively.

Prognostic Features

We analyzed PSA failure-free probability as disease-free survival. Of the clinicopathologic features considered, only Gleason sum was significantly associated with poor outcome in univariate analyses (P = 0.020) (Fig. 5).

Figure 5.

Kaplan–Meier PSA failure-free survival curves of PC patients after radical prostatectomy are grouped according to evaluated variables: Gleason sum, preoperative PSA level, and pT category. Follow-up ranged from 0.7 to 91.4 months, with a median of 33.9 months. Circles represent censored data points.

DISCUSSION

Our results indicate that the M-score was significantly higher in current smokers than in never smokers in PCs (Fig. 3, right). Spearman rank correlation test also revealed significant correlation between pack-years smoked and M-score in PCs (P = 0.039). Moreover, there was significant correlation of M-score with advanced pathologic features such as pT category and Gleason sum (Fig. 4). These results suggest that smoking status may influence tumor progression through CpG hypermethylation of related genes dose-dependently. Hickey et al. reviewed 23 prospective cohort studies, 5 nested case–control studies, 1 retrospective cohort study, and 36 case–control studies addressing smoking and PC.4 Although most of the prospective cohort studies and all of the nested case–control studies that used incident PC as the outcome found no association between current smoking and PC, 33% of 15 population-based case–control studies showed a significant association between smoking and PC risk. They concluded these conflicting results depend on the research design used and how well the study controlled for possible confounding factors. Furthermore, the majority of prospective studies (62% of 13) that used PC death as the outcome noted a positive association between current smoking and PC, supporting our findings where current smoking and advanced PC are associated through CpG hypermethylation of related genes. Therefore, the present study elucidates novel mechanisms whereby smoking may increase PC risk.

Interestingly in lung cancers, some investigators have previously reported that starting smoking during adolescence was the only significant parameter associated with RASSF1A gene methylation, and no association was observed with pack-years smoked.11, 12 In contrast, our result indicated that pack-years smoked was the only parameter associated with methylation status in PCs. Also, methylation status in PC was influenced by smoke exposure in a dose-dependent manner. These findings suggest that methylation induced by smoke exposure may depend on tumor type or gene and that it may be one of the reasons why the mean age of patients with lung carcinoma is younger than that of patients with PC.

We also analyzed PSA failure-free probability as disease-free survival. Of the clinicopathologic features considered, only Gleason sum was significantly associated with poor outcome (Fig. 5). The significant association of the M-score with Gleason sum (Fig. 4) suggests that smoking may influence PC prognosis through CpG hypermethylation of these genes. However, our median follow-up time was probably too short for thorough analysis. We also found there was a trend toward a decreased M-score among former smokers compared with current smokers (P = 0.046 by Bonferroni-adjusted test) (Fig. 3, right), suggesting that demethylation of these genes may be occurring in patients who had quit smoking for more than 12 months. Therefore, patients could decrease their risk for advanced PC by becoming nonsmokers.

Other investigators have used multigene methylation analysis in their studies of various cancers.7, 8, 10 However, their methods of multigene methylation analysis are the sum of the number of genes methylated. Studies have shown that when multigenes are analyzed in the same samples, interpretation of results should be carefully considered because each gene or other clinical factors, including age, may influence one another.22, 23 In the current study, we attempted to integrate the methylation status of multigenes by using a M-score that is the sum of the log HR analyzed by multivariate logistic regression analysis for pathology (BPH vs. PC). This analysis provides automatically adjusted statistical data,22 with each HR directly related to gene methylation in PC samples compared with BPH (methylation-negative) samples (Table 2). By adding the log HR of each gene in a multigene analysis, it is therefore possible to predict the risk of PC in individual patients. Similarly, Ray et al. employed multivariate Cox proportional hazards models for their multigene methylation analysis in medulloblastoma, and they used the sum of the log HR as a risk score for each patient.23 In our study, we found no significant correlation between smoking status and methylation frequency of the APC or MDR1 gene individually. However, by employing a combined analysis (M-score) of the methylation status of the three genes used in this study, differences in methylation status between categories of smokers became apparent (Fig. 3, right). Furthermore, among PCs, the M-score showed a significant stepwise increase with advancing PT category, increasing Gleason score, higher PSA levels, and advancing pathologic features (Fig. 4). Thus, when examining the methylation status of multigenes in PC, the M-score is a reliable, superior, analytical tool for diagnosis and outcome prediction.

In conclusion, this is the first study to demonstrate significant correlation of methylation status of multigenes with smoking status in PC. Smoking status may influence both progression and prognosis of PC through CpG hypermethylation of related genes.

Ancillary