The objective of this study was to discover molecular biomarkers associated with the recurrence of esophageal squamous cell carcinoma (ESCC).
The objective of this study was to discover molecular biomarkers associated with the recurrence of esophageal squamous cell carcinoma (ESCC).
The authors retrospectively analyzed the hypermethylation status of 11 genes using methylation-specific polymerase chain reaction (PCR) and the expression of epidermal growth factor receptor (EGFR), O-6 methylguanine-DNA methyltransferase (MGMT), tumor protein 53 (p53), and transforming growth factor β (TGFβ) using immunohistochemistry in 329 formalin-fixed, paraffin-embedded ESCCs.
Recurrence was identified in 151 of 329 ESCCs (46%) at a median follow-up of 4.5 years. The recurrence was associated with hypermethylation of the genes cell adhesion molecule 1 (CADM1) (P = .003), deleted in colon carcinoma (DCC) (P = .04), or cyclin-dependent kinase inhibitor 2A (p14) (P = .02) in patients with stage I ESCC. Thirty-six of 37 Stage I ESCCs (97%) that had cohypermethylation of at least 2 of the 3 genes had hypermethylation of p14 plus either CADM1 or DCC or both CADM1 and DCC. The 5-year recurrence-free survival (RFS) rates were 93% in patients who had stage I disease without hypermethylation of the 3 genes and 56% in those who had cohypermethylation of p14 in combination with CADM1 and/or DCC. Patients who had stage I ESCC with cohypermethylation of p14 in combination with DCC and/or CADM1 had 7.13 times (95% confidence interval, 1.61-31.64 times; P = .009) poorer RFS compared with those who had no hypermethylation of the 3 genes after adjusting confounding factors. Hypermethylation of the other 8 genes and altered expression of 4 proteins were not associated with recurrence across pathologic stages.
The current results suggested that cohypermethylation of p14 in combination with DCC and/or CADM1 may be an independent prognostic factor for recurrence in patients with stage I ESCC. Cancer 2013. © 2013 American Cancer Society.
Esophageal cancer is 1 of the most common cancers and the leading cause of cancer-related death in the world, with 482,300 incident cases and 406,800 deaths estimated in 2008.1 Despite significant advances in the detection and treatment of esophageal cancer over the past 2 decades, the prognosis for patients with the disease is still poor, with long-term survival rates between 5% and 20%.2 The high rate of recurrence, which develops in approximately 50% of patients after curative surgical resection, is also partially responsible for the poor prognosis. Therefore, the discovery of accurate molecular biomarkers with a prognostic and predictive potential is clearly imperative for identifying patients at high risk for recurrence.
Hypermethylation of CpG islands at the promoter region of tumor suppressor genes is regarded as a prognostic target. Aberrant methylation of normally unmethylated CpG islands is 1 of the most common epigenetic modifications. It induces the transcriptional silencing of tumor suppressor genes and is widely recognized as a mechanism of gene inactivation in cancer.3 Promoter hypermethylation of functionally important cancer-related genes frequently occurs during tumor pathogenesis and also affects clinical outcomes. To discover recurrence-associated molecular biomarkers after patients undergo esophagectomy, we analyzed the promoter methylation of 11 genes in 329 esophageal squamous cell carcinomas (ESCCs). Of the 11 genes examined, 1 group is involved in the cell cycle (2 cyclin-dependent kinase inhibitor 2A genes [p14 and p16]; Ras association domain family member 1 [RASSF1A]; and checkpoint with forkhead and ringer finger domains, E3 ubiquitin protein ligase [CHFR]); and the other group is involved in cell adhesion (cell adhesion molecule 1 [CADM1] and deleted in colon carcinoma [DCC]), tissue invasion (death-associated protein kinase [DAPK] and tissue inhibitor of metalloprotease 3 [TIMP3]), retinoic acid receptor signaling (retinoic acid receptor β2 [RARβ2]), Wnt signaling (adenomatous polyposis coli [APC]), and purine metabolism (fragile histidine triad [FHIT]). In addition to hypermethylation of 11 genes, the expression status of 4 proteins (epidermal growth factor receptor [EGFR], transforming growth factor β [TGFβ], O-6 methylguanine-DNA methyltransferase [MGMT], and tumor protein 53 [p53]), which are involved in growth signaling, repair, or apoptosis, also was analyzed.
It is established that the 15 candidate molecular biomarkers analyzed in the current study are altered in esophageal cancer, but only a few have been studied in terms of prognostic significance. Because a tumor harbors alterations of multiple biomarkers at the gene or protein level and an individual tumor exhibits different frequencies of alteration within a particular profile of biomarkers, our objective was to identify a profile of molecular biomarkers associated with recurrence.
Three-hundred twenty-nine patients with ESCC who underwent surgery at the Samsung Medical Center in Seoul, Korea between May 1994 and November 2005 participated in this study. Information on clinicopathologic characteristics, including survival and recurrence as of March 31, 2012 was elicited from the medical records, and follow-up data were gathered by a trained interviewer. Pathologic stage was determined according to esophageal cancer staging criteria based on the American Joint Committee on Cancer TNM classification system.4 The inclusion and exclusion of patients and their postoperative follow-up for detection of recurrence was conducted as previously described.5
Genomic DNA was extracted from formalin-fixed paraffin-embedded tissues; and the methylation status of CpG islands at the promoter region of the APC, CADM1, CHFR, DAPK, DCC, FHIT, p14, p16, RARβ2, RASSF1A, and TIMP3 genes was determined by methylation-specific polymerase chain reaction (MSP) (Fig. 1A), as previously described.5 The primers used for MSP have been described by us and by and other groups.5-11
Tissue microarrays of ESCCs were prepared as previously described.12 Four-micrometer-thick sections were taken from tissue microarray blocks for immunohistochemistry, deparaffinized in xylene, rehydrated in a graded ethanol series, then subjected to microwave antigen retrieval. Endogenous peroxidase activity was blocked using 0.3% hydrogen peroxide, and nonspecific immunostaining was suppressed with 5% normal horse serum and 1% normal goat serum in phosphate-buffered saline. The sections were then incubated at 4°C overnight with the following primary antibodies: p53 (clone DO-7; Lab Vision, Fremont, Calif) at 1:30 dilution, MGMT (MT3.1; Lab Vision) at 1:50 dilution, TGFβ (TB21; GeneTex Inc., Irvine, Calif) at 1:500 dilution, EGFR (H11; Dako, Carpinteria, Calif) at 1:200 dilution, CADM1 (catalog no. ABT66; Millipore, Temecula, Calif) at 1:500 dilution, DCC (G97-449 clone; BD Pharmingen, San Diego, Calif) at 1:200 dilution, or p14 (NB200-355; Novus, Littleton, Colo) at 1:100 dilution. Detection of immunoreactivity by each antibody was performed using the EnVision DetectionSystem (Dako), and diaminobenzidine (Sigma Chemical Company, St. Louis, Mo) was used as a substrate. The sections were counterstained with Mayer hematoxylin, then dehydrated, and mounted. Primary antibody was replaced with phosphate-buffered saline as a negative control.
All available slides were interpreted by 2 independent authors (E.K. and D.-H.K.) who were blinded to the clinicopathologic variables of each patient to reduce interobserver variability. The expression of p53 was considered positive if immunoreactivity was observed in at least 5% of all nuclei, and the expression of EGFR was considered positive if membranous staining was detected in >5% of cancer cells. MGMT expression was considered negative if positive staining was observed in <5% of total cells. To evaluate TGFβ expression, the percentages of positive-stained cells and staining intensity were evaluated only in the cytoplasm. A composite score from 0 to 12 for TGFβ expression was obtained by adding the proportion score of positive-stained tumor cells (with a score of 0 for the absence of stained cells, a score of 1 for 0%-10% stained cells, a score of 2 for 10%-50% stained cells scored, a score of 3 for 50%-80% stained cells, and a score of 4 for >80% stained cells) and the intensity score (0, none; 1, weak; 2, moderate; 3, strong). TGFβ expression was considered negative in tumors that had a composite score <2. Representative photomicrographs of staining for EGFR, MGMT, p53, and TGFβ proteins are provided in Figure 1B. Immunostaining for CADM1, DCC, and p14 was performed in only 96 stage I ESCCs to analyze the association between expression and hypermethylation of the genes, and expression levels were counted as the percentage of positive staining without a cutoff level.
The association of recurrence with clinicopathologic features was evaluated using the Pearson chi-square test (or Fisher exact test) and the t test (or Wilcoxon rank-sum test) for categorical and continuous variables, respectively. Multivariate logistic regression analysis was conducted using an unconditional approach to estimate the relation between recurrence and covariates after controlling for potential confounding factors and to calculate odds ratios (ORs). Patients who died of other causes unrelated to lung cancer during follow-up, whose cancers did not recur before the end of the study, or who were lost to follow-up were considered censored in the analysis of recurrence-free survival (RFS). Comparisons of RFS between 2 groups were made using the Kaplan-Meier method and the log-rank test. A Cox proportional-hazards analysis was conducted to calculate hazard ratios of independent factors for RFS. All statistical analyses were 2-sided with a 5% Type I error rate.
The association between the disease recurrence and the patients' clinicopathologic features is summarized in Table 1. At a median follow-up of 54 months, 151 patients (46%) had developed disease recurrence. The mean age of patients was similar between those with and without recurrence (P = .55). Recurrence was more frequent among men (47%) than among women (30%), but the difference was not statistically significant (P = .14). Poorly or moderately differentiated carcinomas recurred more frequently than well differentiated tumors (P = .01). Three-field lymph node dissection was associated significantly with a high risk of recurrence compared with transhiatal esophagectomy and 2-field lymph node dissection (P = .02). Tumors located in the cervical or upper esophagus recurred more frequently than those in the middle or lower esophagus, and this difference was statistically significant (P = .05). Patients who had perineural invasion had a higher risk of recurrence than those without (86% vs 44%; P = .004). Recurrence also was associated with lymphatic invasion (P = .04) but not with vascular invasion (P = .09). In addition, a significant association was observed between the pathologic stage and recurrence (P < .0001). Recurrence also occurred with high prevalence among patients who received adjuvant therapy (chemotherapy, radiotherapy, or concurrent chemoradiotherapy; P = .01) or neoadjuvant therapy (P = .03) compared with those who did not receive adjuvant therapy.
|Recurrence: No. of Patients|
|No, n = 178||Yes, n = 151||P|
|Age: Mean±SD, y||62±8||61±8||.55|
|Location of tumor|
Correlations between recurrence and the hypermethylation status of 11 genes and the altered expression of 4 proteins were analyzed to identify biomarkers associated with recurrence after esophagectomy in patients with ESCC. CpG island hypermethylation was detected for APC in 71% of ESCCs, for CADM1 in 33%, for CHFR in 38%, for DAPK in 51%, for DCC in 31%, for FHIT in 32%, for p14 in 54%, for p16 in 43%, for RARβ2 in 22%, for RASSF1A in 31%, and for TIMP3 in 16% (Fig. 1C). The expression of p53 and E-cadherin was positive in 54% and 32%, respectively, of the ESCCs studied. TGFβ and MGMT were positive in 62% and 42% of ESCCs (Fig. 1D). We performed a stratified data analysis for each disease stage rather than computing a single summary risk ratio or RFS for all ESCCs, because the effect of DNA methylation on recurrence was not homogenous across pathologic stages.
For 96 patients with stage I ESCC, recurrence was associated with hypermethylation of CADM1 (P = .003), DCC (P = .04), or p14 (P = .02) (Fig. 2A). Multivariate logistic regression analysis was conducted to calculate the OR for recurrence according to methylated genes across pathologic stages after controlling for potential confounding factors. When CADM1, DCC, and p14 were hypermethylated in stage I ESCCs, but not in stage II through IV ESCCs, the adjusted ORs for recurrence were 3.18 (95% confidence interval [CI], 1.19-8.51), 1.57 (95% CI, 1.08-10.26), and 3.31 (95% CI, 1.10-9.91), respectively, after adjusting for age, sex, tumor differentiation, operation method, tumor location, lymphatic invasion, vascular and perineural invasion, neoadjuvant therapy, and postoperative adjuvant therapy (Fig. 2B). However, no relation was observed between recurrence and hypermethylation of the other 8 genes or expression of the 4 proteins across pathologic stages (data not shown).
Before analyzing the effect of CADM1, DCC, and p14 gene cohypermethylation on RFS in stage I ESCCs, we investigated the relation between hypermethylation and expression of the genes. Representative images of positive staining for CADM1, DCC, and p14 are provided in Figure 2C. The expression of individual genes was assigned to the percentage of tumor cells that stained positively in cytoplasm for CADM1 and DCC and in nuclei for p14. In most tumors that had hypermethylation of a gene, low expression of a protein was encoded by the gene (Fig. 2D). Possible interaction of the CADM1, DCC, and p14 genes was examined further for an association with recurrence or RFS in stage I ESCCs (Fig. 3). Twenty-nine (30%) of 96 stage I ESCCs had no methylation of any of the 3 genes. In total, 31% of 96 stage I ESCCs had 1 gene hypermethylated, 21% (20 of 96) had 2 genes hypermethylated, and 18% (17 of 96) had 3 genes hypermethylated (Fig. 3A). Thirty-six of 37 patients (97%) who had cohypermethylation of 2 or 3 genes had p14 hypermethylation (Fig. 3B). In addition, 36 patients who had p14 hypermethylation also had cohypermethylation of at least 1 of the other 2 genes (CADM1 and/or DCC) (Fig. 3C), suggesting that hypermethylation of p14 may play a critical role in ESCC in combination with CADM1 or DCC. In addition, hypermethylation of CADM1 (P < .001) and DCC (P < .001) occurred with significantly greater prevalence in patients who had hypermethylation of p14 than in those without (Fig. 3D). On the basis of these observations, it is likely that hypermethylation of p14 may contribute to recurrence in combination with hypermethylation of CADM1 or DCC in patients with stage I ESCC.
RFS was analyzed with regard to hypermethylation status of the CADM1, DCC, and p14 genes in stage I ESCCs. The 5-year RFS rates were 72% and 87% in ESCCs with and without hypermethylation of the p14 gene, respectively (Fig. 4A). This difference was statistically significant (P = .03). The 5-year RFS also was significantly different in patients with and without hypermethylation of the CADM1 gene (61% vs 82%, respectively; P = .02) (Fig. 4B), but RFS was not associated with hypermethylation of the DCC gene (P = .11) (Fig. 4C). RFS was analyzed further in relation to cohypermethylation of CADM1, DCC, and p14. One patient who had cohypermethylation of CADM1 and DCC but was without hypermethylation of p14 was excluded from further analysis. Patients with stage I disease were divided into 3 groups according to the number of methylated genes: 1) no hypermethylation of 3 genes, 2) hypermethylation of any 1 of 3 genes, and 3) cohypermethylation of p14 in combination with CADM1 or DCC. RFS was associated significantly with the presence of several hypermethylated genes (P = .009) (Fig. 4D): the 5-year RFS rates was 93% for patients with stage I ESCC who had no hypermethylation of the 3 genes and 56% for those who had cohypermethylation of p14 in combination with CADM1 or DCC.
Multivariate logistic regression analysis (Table 2) and Cox proportional-hazards analysis (Table 2) for patients with stage I ESCC were performed to control for potential confounding effects of variables, such as age, sex, tumor differentiation, operation method, tumor location, lymphatic invasion, vascular invasion, perineural invasion, neoadjuvant therapy, and postoperative adjuvant therapy and to calculate ORs and hazard ratios. Patients who had hypermethylation of any 1 of the CADM1, DCC, or p14 genes did not differ significantly in terms of recurrence or RFS compared with those who did not have hypermethylation of any of the 3 genes after controlling for confounding factors. However, we observed that the risk of recurrence in patients who had cohypermethylation of p14 gene in combination with CADM1 and/or DCC was 8.59 times greater (95% CI, 1.76-41.91 times greater; P = .008) than that of those without hypermethylation of any of the 3 genes (Table 2). In addition, we observed that patients with cohypermethylation had a 7.13 times poorer RFS (95% CI, 1.61-31.63; P = .009) than those without hypermethylation of any of the 3 genes (Table 2).
|No. of Methylated Genes||OR/HR||95% CI||P|
|Logistic regression analysisa|
|Cox proportional hazards analysis|
To identify recurrence-associated prognostic indicators after esophagectomy in patients with ESCC, we retrospectively analyzed the relation between recurrence and the promoter hypermethylation of 11 genes and the expression of 4 proteins. In this study, patients with stage I ESCC who had cohypermethylation of p14 in combination with CADM1 and/or DCC had a 7.2 times greater risk of recurrence compared with those who had no hypermethylation of any of the 3 genes. In addition, 36 of 37 patients (97%) with stage I ESCC who had cohypermethylation of CADM1, DCC, or p14 had hypermethylation of p14; and those 36 patients had hypermethylation either of CADM1 or DCC or of both CADM1 and DCC. These observations suggest that hypermethylation of p14, a key regulator of the p53 tumor suppressor pathway, is critical for recurrence, and additional hypermethylation of DCC or CADM1 may be required for recurrence to develop in patients with stage I ESCC. However, the mechanistic basis for the cooperative effect of p14 with DCC and/or CADM1 on the high risk of recurrence in stage I ESCCs is not clear.
The p14 gene is a tumor suppressor gene that is encoded by CDKN2A through alternative splicing on human chromosome 9p21. Several groups have reported the association of p14 with prognosis in esophageal cancer: CpG island methylator phenotype (CIMP), including p14, is associated significantly with poor 4-year survival in patients with ESCC,6 and hypermethylation of p14 is associated with a poor prognosis in patients who have adenocarcinoma of the esophagus.7 Reduced expression of p14 is also associated with a poor prognosis in patients with ESCC.8 In the current study, hypermethylation of p14 occurred in 54% of 329 ESCCs, consistent with the prevalence reported by other groups.6, 8 In fact, p14 is induced transcriptionally in response to hyperproliferative signals and interferes with tumor protein 53-E3 ubiquitin protein ligase homolog (p53-MDM2) complex formation and proteosome degradation, leading to apoptosis or cell cycle arrest through stabilization of p53 and activation of p53-dependent transcription. In addition, it is known that p14 inhibits cell growth independent of p53 by attenuating the transactivating activity of growth-promoting genes, such as E2F transcription factor 1 (E2F1) and myelocytomatosis viral oncogene homolog (c-myc). Accordingly, p14 may affect recurrence by inducing cell cycle arrest and promoting apoptosis.
The DCC gene is located at 18q21.1 and participates with other proteins in cell-cell and cell-matrix interactions, and hypermethylation of DCC has been detected in esophageal cancer.9 In the current study, hypermethylation of DCC was associated with a high risk of recurrence in combination with p14 in patients with stage I ESCC. Although stage I esophageal cancer spreads slightly deep, it is usually located in the 2 inside layers of the esophagus and does not extend to nearby tissues, lymph nodes, or other organs. Therefore, hypermethylation of DCC in stage I cancers may result in alterations in cell-to-cell adhesion, leading to the conversion of an epithelial cell to a fibroblastoid phenotype (epithelial-mesenchymal transition) and, eventually, to an increased risk of invasion. DCC is also known to induce cell death, and the loss of the proapoptotic activity of DCC is known to result in the development of highly aggressive intestinal adenocarcinoma in a predisposing APC mutant context.13 Those studies suggest that hypermethylation of DCC may contribute to recurrence through the failure of apoptotic activity in addition to the loss of cell adhesion.
To our knowledge, there is no direct evidence about the relation between p14 and DCC in human cancer. However, a mouse model of mammary carcinoma supports their correlation indirectly. DCC controlled apoptosis induction in p53-deficient mammary tumor cells in vitro, and additional loss of DCC in mammary tumor with somatic inactivation of p53 promoted the formation of metastasis without affecting primary tumor phenotype.14 Accordingly, it is possible that hypermethylation of p14 may result in destabilization of p53 by failing to control MDM2 and contributing to tumor metastasis in combination with hypermethylation of DCC. In the current study, we observed that hypermethylation of DCC occurred with greater prevalence in tumors that had hypermethylation of p14 than in those without (46% vs 19%, respectively; P < .001) (Fig. 3D). On the basis of these observations, it is likely that p14 may play a role with DCC in metastasis of ESCC. Further study is required to understand the molecular bases of the cooperative effect of DCC and p14 on recurrence in stage I ESCCs.
CADM1, also known as TSLC1 (tumor suppressor in lung cancer-1 [IGSF4]), is a single-transmembrane glycoprotein and was originally identified on chromosome 11q23.2 by functional complementation of A549 lung cancer cells through suppression of tumorigenicity in nude mice.15 CADM1 is involved in intracellular adhesion through homophilic and heterophilic trans interactions.16, 17 Ito et al10 also reported the association of TSLC1 with cell-cell adhesion in ESCC: TSLC1 protein accumulates in interdigitated structures at cell-cell membranes in TSLC1-transfected ESCC cells, revealing an aggregated morphology. CADM1 is also connected to actin cytoskeleton through the tumor suppressor DAL-1 (differentially expressed in adenocarcinoma of the lung-1 [EPB41L3]) at the cell-cell-attached site.17 In addition to cell adhesion, CADM1 is also involved in tumor metastasis: restoration of CADM1 expression strongly suppressed liver metastasis from the spleen in a lung adenocarcinoma cell line, A549, in athymic nude mice.18 The motility and invasive capacity of TSLC1-transfected ESCC cells is significantly suppressed in vitro, and TSLC1 expression is associated with the depth of invasion and status of metastasis in ESCC.10
Silencing of CADM1 by aberrant methylation has been reported in primary tumors from the esophagus,10 and the association of patient prognosis with CADM1 also was evaluated by a few groups: hypermethylation of CADM1 was associated with a short disease-free survival in patients with nonsmall cell lung cancer,11 and loss of CADM1 expression was associated significantly with a low survival rate in patients with ESCC.10 In the current study, hypermethylation of CADM1 was detected in 33% of ESCCs studied and was associated with poor RFS in combination with hypermethylation of p14 among patients with stage I ESCC. In addition, hypermethylation of CADM1 occurred with greater prevalence in patients with p14 hypermethylation than in those without (P < .001) (Fig. 3D). On the basis of previous reports and our current findings, we believe it is possible that CADM1 hypermethylation may contribute to poor RFS with p14 synergistically by leading residual cancer cells to invade or metastasize through the disruption of cell-cell adhesion after esophagectomy in patients with stage I ESCC.
The current study was severely limited by several factors. First, the small number of samples may be a possible risk for reaching a false-positive conclusion. Second, MSP is a qualitative method for analyzing DNA methylation. The current results need to be validated using a quantitative real-time method, such as MethyLight. Third, data concerning a preneoplastic lesion with genetically altered cells were not available in this study. Thus, some of the second-field tumors or second primary tumors may have been misclassified as locally recurrent tumors. Additional large-scale studies are needed to validate the clinical usefulness of the genes. In conclusion, the current study suggests that cohypermethylation of p14 in combination with CADM1 and/or DCC may be a valuable recurrence-related prognostic indicator in patients with stage I ESCC.
This work was supported by grants from the National R&D Program for Cancer Control, Ministry for Health and Welfare (#1120270), the Korea Healthcare technology R&D Project, Ministry of Health & Welfare (#A101148), and the National Research Foundation of Korea (NRF) grant funded by the Korea government (MEST) (No. 2011-0029138), Republic of Korea.
CONFLICT OF INTEREST DISCLOSURE
The authors made no disclosures.