Clinical Correlates of Promoter Hypermethylation of Four Target Genes in Head and Neck Cancer: A Cooperative Group Correlative Study

Head and neck squamous cell carcinoma (HNC) arises in the mucosa of the upper aerodigestive tract as the result of an accumulation of molecular alterations that affect cellular behavior resulting in a malignant phenotype. These alterations may involve the DNA coding sequence directly (genetic) or indirectly (epigenetic) and result in variation in expression and/or function of genes with tumor suppressive or oncogenic activity. Primary sequencing of the cancer genome of HNC has identified numerous target genes (1, 2), however, the scope and number of identified gene mutations are insufficiently robust to account for the entire cancer phenotype. Epigenetic alterations, specifically promoter hypermethylation, are well documented to contribute to tumorigenesis through the abrogation of expression of tumor suppressor genes (3). Many candidate genes have been found to display promoter methylation in HNC (4). These genetic and epigenetic alterations interact in complex and integrated fashion, affecting key cellular functional pathways that control cell cycle and proliferation, response to genotoxic stress and apoptosis, metastasis and invasion, and terminal differentiation. It is reasonable to hypothesize that analysis of molecular alteration in cohorts of cancer patients may provide valuable insight into their individual and cumulative effect on tumor aggressiveness and prognosis.

Aberrant DNA methylation of a particular gene or a set of selected genes may be used as a cancer biomarker for clinical assessment. Similar to genetic mutations, promoter-specific DNA hypermethylation results in silencing tumor suppressor genes, mediating tumorigenesis, invasiveness, and metastasis (5, 6). As a stable, clonally propagated event, aberrant DNA methylation has significant tumor-type specificity and variability between patients, which may indicate clinical usefulness (7). Single or panels of DNA methylation markers have been proposed as suitable prognostic biomarkers for cancer progression and outcomes. Methylated PITX2 promoter has been associated with a poor outcome in breast and prostate cancer (8, 9). The promoter hypermethylation of CDKN2A, CDH13, RASSF1A, and APC were associated with early recurrence in stage I non–small cell lung carcinoma (10). Other prognostic hyperemethylated genes in HNC have been reported: DCC, hMLH1, MGMT, HMT, ATM, GSTP1, MINT, etc. (11). DNA methylation biomarkers show promise for clinical implementation, however, the identification of reliable prognostic biomarkers await confirmation from large, well-defined sample cohorts (12).

The immense complexity of both clinical and biological factors that serve as potential prognostic factors in any cohort of HNC dictates that the clinical data used to evaluate the effect of individual molecular events be very robust. Large, well-defined cohorts followed for adequate periods of time with rigorous quality control for data collection are necessary to provide the statistical power needed to assess the relative contribution of various molecular factors and their independence with respect to clinical parameters that are more readily and economically derived. Eastern Cooperative Group (ECOG) 4393/Radiation Therapy Oncology Group (RTOG) 9614 is a cooperative group, multiinstitutional study involving the collection of tumor, and resection margin tissue samples along with clinical and treatment outcome data from a large cohort of patients with HNC. Molecular analysis of tissue samples for genetic and epigenetic alterations was done at a centralized location (Johns Hopkins Head and Neck Tumor Laboratory, Baltimore, MD) whereas clinical and outcome data were collected by the cooperative groups and statistical analysis conducted by the ECOG. With 480 evaluable cases treated with primary surgery with curative intent and followed for more than 10 years after surgery, this cohort provides one of the best resources for exploring possible associations between molecular alterations and clinical tumor behavior available in the world. We have been examining the potential utility of candidate tumor promoter hypermethylation events in this cohort for molecular detection of rare cancer cells in the margin samples (the primary goal of the protocol) to improve the sensitivity and specificity of molecular margin analysis. As a result, the presence of methylation of 4 genes (p16, DCC, KIF1A, and EDNRB) in tumor samples has been determined providing opportunity to test these events individually and in combination for associations with survival. This strategy has already been done to examine the prognostic significance of TP53 gene mutation in the cohort. The value of the ECOG 4393/RTOG 9614 cohort was showed by our ability to display a prognostic association between those TP53 mutations, which disrupted the DNA binding capacity of the p53 protein and decreased overall survival (13). The current report expands on and refines those earlier findings by adding the information about the spectrum of tumor-specific methylation of the 4 target genes to that of TP53 mutation.

The study cohort was from a prospective, multicenter study involving 18 member institutions of ECOG and RTOG between 1996 and 2002 (study no. E4393/R9614). The first objective of the study protocol was to determine the clinical utility of molecular detection of residual cancer cells in tumor margins and the second independent objective was to seek association between molecular alterations in HNC and survival. The protocol was approved by the cooperative groups and the investigational review board of each participating institution. Written informed consent was obtained from each patient. The study enrolled 560 eligible patients with HNC who underwent primary surgical extirpation of tumor with curative intent. The cooperative group data managers collected and reviewed demographic, clinical, and pathologic information for each patient perioperatively and follow-up patient information at scheduled intervals (at every 6 months for the first 3 years and annually thereafter).

Tumor and margin samples were collected during operation and rapidly frozen in −80°C. The samples were shipped on dry ice by overnight courier to the Johns Hopkins Head and Neck Cancer Research Laboratory, and kept frozen in liquid nitrogen until DNA extraction. Frozen tumor specimens were microdissected in series of 5-μm sections and processed with hematoxylin and eosin (H&E) staining for light microscopic examination to ensure the presence of HNC. Tumor samples with at least 70% cancer cells were candidates for molecular studies. Frozen tissues with less than 70% tumor cells were microdissected to enrich the tumor-cell content. Additional 12-μm sections were then cut and digested with DNA elution buffer containing 1% sodium dodecyl sulfate and 50 μg/mL proteinase K (Boehringer) at 48°C for 2 to 3 days. Initial, middle, and last sections cut for DNA extraction were also stained with H&E to ensure the presence of >70% HNC. Genomic DNA was extracted with phenol–chloroform extraction and ethanol precipitation, eluted in low-salt 10 Mmol/L Trix-HCl and 2.5 mmol/L EDTA (LoTE) buffer.

DNA (2 μg) extracted from tissue samples was subjected to bisulfite treatment, which modifies CpG islands, using the EpiTect Bisulfite Kit (Qiagen) according to the manufacturer's instructions (14, 15). The bisulfite-treated DNA was resuspended in 50 μL of elution butter and stored at −80°C.

The bisulfite-treated DNA was used as a template for fluorescence-based real-time PCR and was amplified using quantitative methylation-specific PCR (Q-MSP) as described previously (16). Genes selected for this study came from the studies previously conducted by the authors chosen for utility (specificity and sensitivity) for detection of HNC in saliva: p16, DCC, KIF1A, and EDNRB (17–20). Primers and probes were designed to amplify the bisulfate-modified DNA for 4 gene promoters and internal reference gene (ACTB). The sequences of primers and probes are available in previous publication (21). For all Q-MSP reactions, 3 μL of bisulfite-treated DNA were added to a final volume of 20 μL consisting of each 600 nmol/L primer and 200 μmol/L probe, 0.75 U platinum Taq polymerase (Invitrogen), 200 nmol/L ROX reference dye (Invitrogen), 200 μmol/L dNTP each, 16.6 mmol/L ammonium sulfate, 67 mmol/L Trizma (Sigma), 6.7 mmol/L magnesium chloride, 10 mmol/L mercaptoethanol, and 0.1% dimethylsulfoxide. All reactions were carried out in triplicate to ensure consistent results in 384-well plates using a 7900 Sequence Detector System (Perkin-Elmer Applied Biosystems). Thermal cycling included a first denaturation step at 95°C for 10 minutes, followed by 50 cycles of 95°C for 15 seconds and 60°C for 1 minute. As a standard reference for Q-MSP, leukocyte DNA from a healthy individual was completely methylated using excess SssI methyltransferase (New England Biolab Inc.) and methylated DNA was serially diluted 45 to 0.00045 ng to construct a calibration curve for each plate (22). Each plate contained wells with only DNA- and RNA-free water and the reaction mix for negative control to ensure no contamination. Each reaction was done in triplicate and the average of the triplicate was considered for analysis. The methylation level of each gene in each DNA sample was normalized to ACTB and calculated as an equation of (gene of interest/ACTB) 100×. The normalized value is set to 100 if the ratio (gene of raterest/ACTB) is greater than 1.

TP53 mutation data have been previously published and was derived with the use of the GeneChip p53 assay (Affymetrix) and the Surveyor DNA endonuclease–denaturing high-performance liquid chromatography (DHPLC) for high-throughput detection of mutations in exons 2 to 11 in head and neck primary tumors (13). The TP53 mutations were grouped as “disruptive” and “nondisruptive,” as previously defined. Disruptive mutations included nonconservative mutations occurring within key DNA binding domains (L2–L3 regions) or stop codons in any region, and all other mutations were defined as nondisruptive mutations.

The presence or absence of human papilloma virus (HPV) DNA has been determined using in situ hybridization for a subset of the ECOG cohort and previously published (23). Additional samples have been screened using quantitative PCR. To quantify the viral load of HPV-16, genomic DNA from patients was used for real-time PCR. Specific primers and probes sets are reported in a previous publication (24). PCR for housekeeping gene β-actin was done in duplicate and parallel to normalize the input DNA. The final primer and probe concentrations were 0.3 and 0.1 mmol/L, respectively, in a total volume of 10 mL. Each reaction was run for 50 cycles. All experiments were done at least twice in duplicate. Four water negative controls were included in each run. Samples in which 4 results were not concordant were repeated twice more in duplicate as they were usually because of failed PCR in one of the initial reactions. By using serial dilutions, standard curves were developed for the HPV-16 viral load using CaSki (American Type Culture Collection, Manassas, VA) cell line genomic DNA, as it has been previously characterized to harbor 600 copies of HPV-16 DNA per genome equivalent (6.6 pg of DNA/genome). Standard curves were developed for HPV-16 E6 and E7, using serial dilutions of genomic DNA extracted from CaSki cells with following concentrations: 25, 2.5, 0.25, 0.025, 0.0025, and 0.00025 ng of genomic DNA. A standard curve was developed as well for the β-actin housekeeping gene, which has 2 copies per genome equivalent, using the same serial dilutions of the CaSki genomic DNA. For all samples, duplicate control β-actin amplifications of 10 ng of total DNA were positive. Only tumors arising in the oropharynx were included in PCR HPV testing. HPV status was available for 69 oropharynx samples (23 negative, 46 positive) and one sample from the oral cavity (negative). Cases of oropharyngeal tumors for which HPV status could not be determined were coded as unknown. HPV status was imputed to be negative for samples with missing HPV status from sites other than the oropharynx.

Laboratory and clinical data were submitted to the ECOG central office and analyzed at the ECOG statistical center, including patient demographics, staging, risk factor exposure, pathology, treatment, and outcome. Descriptive statistics were used to summarize the methylation levels by baseline characteristics of the patients. Wilcoxon or Kruskal–Wallis rank-sum test was used to test for differences in methylation levels among baseline categories. Overall survival was defined as the time between study entry (surgery) and death from all causes or last follow-up. Patients who were alive at their last follow-up were censored. The Kaplan–Meier method (25) was used to estimate survival curves. Multivariate Cox proportional hazards models (26) were used to assess the significance of variable on overall survival. Multivariate models considered TP53 status, marker methylation levels, HPV status, tumor site and stage, nodal stage, smoking history, average alcohol use, treatment type, interaction between TP53 status and HPV status, and interaction between TP53 status and marker methylation levels. Methylation levels were included as continuous variables, and all other variables except for age were categorical. P-values for HRs were calculated using the likelihood-ratio test, and values of less than 0.05 were considered to indicate statistical significance. Fisher exact test (categorical variable with 2 categories), χ² test (categorical variable with 3 or more categories) and Wilcoxon test (continuous variable) were used to test for differences in characteristics between included and excluded patients. Linear models were used to assess the association between methylation levels and smoking history while controlling for age, gender, site, and HPV status.

The ECOG 4393/RTOG 9614 population includes 560 subjects with HNC. Sixteen subjects did not meet entry criteria, DNA was not available for 176 others, and 15 did not have p53 mutation status, leaving 353 tumor samples/cases that form the basis of this analysis. A comparison of clinical characteristics between included and excluded patients is shown in Table 1. The quantity of methylation for each of the 4 target genes compared to β-actin is shown in Fig. 1. The relative quantity of methylation by QMSP was considered both as a continuous and as dichotomous (high and low) values. Tumors from 48 cases displayed below-median methylation in all 4 target genes, whereas 50 cases displayed above-median levels of methylation of all 4 genes. Associations of promoter methylation with clinical features including tumor site, T and N stage, gender, treatment modality, and use of alcohol and tobacco as well as with the molecular factor, status of p53 mutation, and HPV status were examined (Table 2). There was some variability of methylation by tumor site with higher methylation of p16 and KIF1A in oral cavity tumors whereas p16 methylation was higher in lower T stage tumors. Of note, methylation of p16, DCC, EDNRB, and KIF1A varies across the amount of tobacco used (Fig. 2). Methylation level for these markers is negatively associated with tobacco use when explored in a multivariate linear model (Table 3).

The effect of each factor on overall survival was analyzed. Advanced nodal status was associated with a significantly increased risk of death (HR = 3.2; P < 0.001; Table 4). Disruptive p53 mutation conveyed a 1.6-fold increased risk of death compared to wild-type (HR = 1.632; P = 0.069). There was no difference in survival seen for methylation of KIF1a, or EDNRB taken as individual events alone (Fig. 3). Although methylation of DCC appeared to be significantly associated with decreased survival in univariate analysis (P = 0.038; Fig. 3), this effect was not independent of other molecular factors in the multivariate model (Table 4). However, methylation of p16 was associated with decreased survival for the overall cohort in multivariate analysis (HR = 1.008; P = 0.045). When methylation of these genes is combined with p53 mutational status, the presence of p16 methylation conveyed a significant incremental improvement in survival for tumors with a disruptive p53 mutation (P = 0.019). The log HR for death decreased for subjects with tumors with disruptive TP53 mutation by 0.017 for each unit of increase in p16 methylation level. Taking p16 as a dichotomous value, the 26 cases with disruptive p53 and positive p16 methylation had survival similar to that of cases with nondisruptive mutation with any p16 status, whereas those with disruptive p53 mutation and 0 p16 methylation fared more poorly (Fig. 4A and B). This protective effect was independent of smoking status in multivariate analysis. It also seems to be free of interaction with HPV status. The presence of HPV DNA was not associated with risk of death or methylation status in this cohort (Tables 3 and 4). The observed effect of p16 methylation on survival in our cohort is independent of HPV status in our multivariate model. The protective effect was observed predominantly in cases arising outside the oropharynx, which do not contain HPV. Of the 14 oropharynx cases that have disruptive TP53, only 3 have p16 methylation levels greater than 0. Because of a paucity of tumors with both HPV DNA and disruptive p53 mutation, however, it cannot be rigorously showed that p16 methylation remains protective in the presence of both disruptive p53 mutation and oncogenic HPV.

Tumor-specific hypermethylation of the promoter region of genes is a well-recognized mechanism of abrogation of gene function that may contribute to the malignant transformation and phenotype of cancer. When hypermethylation of the promoter is present, the gene is not transcribed, reducing or eliminating the function of tumor suppressor genes such as p16. As a result, methylation of genes involved in key pathways that contribute to tumorigenesis may affect tumor behavior and prognosis. The potential for prognostic implication of methylation of individual genes has been explored for a variety of human cancers including HNC. The power and reliability of studies exploring prognostic effect of tumor markers is dependent on sample size, length of follow-up, and integrity of clinical outcome data. We have looked for association between clinical outcome and promoter hypermethylation of 4 tumor-specific target genes (p16, DCC, EDNRB, and KIF1A) in a large, prospectively collected cohort of HNSCC from the multiinstitutional cooperative group study ECOG 4393/RTOG 9614. This cohort is among the largest collection of HNC tumors ever evaluated for correlation between molecular alterations and clinical outcome. In this study, the effect of methylation of each candidate gene was considered individually as well as in combination with the others and with mutation status of TP53 and the presence or absence of HPV DNA. Mutation status of TP53 in this cohort has previously been shown to have a significant correlation with overall survival. In particular, patients with tumors with disruptive mutation of TP53 have a greater than 2-fold rate of death compared those who have HNC with either wild-type or nondisruptive TP53 mutation (13).

It is increasingly apparent that the functional disruptions underlying tumorigenesis and the cancer phenotype are complex, affecting an interwoven system of key pathways that control functions such as apoptosis, cell growth regulation, signal transduction, and differentiation. There are numerous gene products involved in each of these pathways, rendering multiple candidate treatment targets and detection/prognostic markers. Our results reflect the complexity and interdependence of the many genes involved in 2 ways. The methylation status of p16 was found to affect the outcome of patients particularly the subsets defined by the concomitant status of the TP53 gene. In particular, prognosis of the group of patients with tumors containing disruptive TP53 mutation, known to have a poor clinical outcome, was further refined by simultaneously considering the methylation status of p16.

Most published reports of the prognostic association of p16 methylation in cancer indicate a poorer outcome for tumors with this epigenetic event, in accordance with our findings for the overall cohort. A significant survival disadvantage was observed for p16 hypermethylated colorectal cancers (n = 84) of which 28.6% were affected. Five-year survival probability was 25% for methylated tumors and 72% for nonmethylated tumors (27). A larger study of 212 colorectal carcinoma cases showed an association between high p16 methylation and large tumor size, more frequent recurrence and shorter cancer-related survival (HR = 3.38; P < 0.001; ref. 28). These results were not confirmed by a more recent cohort study and literature review indicating that promoter methylation of CDKN2A was not associated with colorectal-specific mortality (29). In a study of 44 Stage 1A NSCLC cases, multivariate analysis showed reduced relapse free survival associated with p16 methylation P = 0.016 (30). However Wif-1 methylation was more strongly associated with overall survival in the NSCLC cohort, showing the complexity of analysis of multiple parameters in a small study population. In another report, p16 methylation was again associated with poor prognosis of NSCLC in a study of 119 cases, showing a reduced 5-year survival rate for resectable cancers, particularly those with early stage (31). Methylation of p16 has also been implicated as a poor prognostic marker for ovarian carcinoma as indicated in a study of 249 cases. Among the 100 cases with p16 methylation, there was a significantly higher risk of disease progression (32).

Methylation of p16 has also been associated with poor outcome in HNC in several published reports. Among 52 oral squamous cell carcinoma cases, hypermethylated p16 was associated with higher rates of lymph node metastasis and distant metastasis and with shortened disease-free survival. However, these results were only seen in certain subsets of patients (older or younger) in this small study and were not confirmed with multivariate analysis (33). In another report of 78 cases with oral epithelial dysplasia, the rate of progression to oral cancer in lesions with p16 methylation was significantly higher than when p16 was not methylated. This observed effect was stronger for older patients and moderate dysplasia cases, again showing the limitation of the observation in that it applied only to very small subsets (34).

In our cohort of HNC patients treated with surgery with or without postoperative radiation, p16 methylation status alone was found to be associated with a significant difference in survival when analyzing the entire group. However, when considering subsets of patients, the group with the worst prognosis based on p53 mutation status (disruptive p53 mutation), a high level of p16 methylation was associated with a better outcome. Individuals with disruptive p53 mutation who also had high p16 methylation had a rate of survival similar to those with nondisruptive mutation (Fig. 4B). The cohort utilized for this study was of a size and follow-up duration to offer sufficient power and reliability. Compared to published results of smaller cohorts, our ability to conduct multivariate analysis of subsets of the cohort is enhanced. The mechanism by which p16 methylation portends an improved prognosis for tumors with TP53 disruptive mutation, a group with a poor prognosis, has not been investigated. However, it is well documented that most HNC have some disruption of cell-cycle control at the G₁ checkpoint, including deletion or mutation of p16, presence of HPV16 E7, or rarely Rb mutation (35). The relationship with smoking and lower p16 methylation levels suggests that exposure to carcinogens likely results in a higher frequency of mutation and deletion of p16 and less likelihood of endogenous transformation through p16 methylation. The observed “protective” effect may reflect a better outcome from lower carcinogen exposure and a less damaged cancer genome. It may also be that p16 methylation represents partial abrogation of cell-cycle control, sufficient for tumorigenesis but with less virulent tumor behavior as a result.

The prognostic implication of methylation of other individual genes has been reported in several studies of limited size. Methylation of DCC was correlated with aggressive phenotype for HNC in a study of 96 tumors with increased rate of bone invasion, invasive growth pattern, and reduced survival (P = 0.050; ref. 36). Intriguingly, in this study, methylation of p14ARF was associated with a good prognosis (HR = 0.30 for methylated tumors; P = 0.021), perhaps implicating a mechanism similar to the protective effect of p16 methylation in our cohort. Our group recently examined promoter hypermethylation of a larger panel of genes in DNA collected in salivary rinses from head and neck cancer patients as a biomarker for outcome (20). Overall survival in a group of 61 HNC patients was significantly shorter if pretreatment saliva was found to contain tumor-specific hypermethylated DNA of any of the target genes (DAPK; DCC; MINT-31; TIMP-3, p16, MGMT, CCNA1) P = 0.015. Hypermethylated DNA in saliva was found to independently predict local recurrence in a multivariate analysis (P = 0.010). However, in a subsequent study of 97 patients, only methylation of TIMP-3 in pretreatment salivary rinses remained predictive of local recurrence (15). The inconsistency of these results may be because of the relatively small sample size as well as the indirect approach of sampling tumor DNA that has been released into saliva of cancer patients.

In contrast to the result for p16 methylation, there was no association in the study cohort between survival and methylation of KIF1A, EDNRB, or DCC alone. The presence of oncogenic HPV DNA in HNC has been reported to be a good prognostic indicator in numerous series (37–39). However, HPV is a factor only in the oropharynx because the presence of HPV DNA in tumors from other sites is vanishingly rare. HPV status did not seem to influence the clinical effect of promoter hypermethylation in this series.

No potential conflicts of interest were disclosed.

Conception and design: J.-L. Roh, D. Sidransky, A.A. Forastiere, W.M. Koch

Development of methodology: J.-L. Roh, W.M. Koch

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.-L. Roh, D. Sidransky, W.M. Koch

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X.V. Wang, J. Manola, D. Sidransky

Writing, review, and/or revision of the manuscript: J.-L. Roh, X.V. Wang, J. Manola, D. Sidransky, A.A. Forastiere, W.M. Koch

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Manola, A.A. Forastiere

Study supervision: W.M. Koch

This work is supported by grants from National Institute of Health and the National Institute of Dental and Craniofacial Research (R01 DE013152-11).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.