Abstract
Endothelin receptor type B (EDNRB) and kinesin family member 1A(KIF1A) are candidate tumor suppressor genes that are inactivated in cancers. In this study we evaluated promoter hypermethylation of EDNRB and KIF1A and their potential use for risk classification in prospectively collected salivary rinses from patients with premalignant/malignant oral cavity lesions. Quantitative methylation-specific PCR(Q-MSP) was performed analyzing methylation status of EDNRB and KIF1A in salivary rinses of 191 patients. We proceeded to determine the association of methylation status with histologic diagnosis and estimate classification accuracy. On univariate analysis, diagnosis of dysplasia/cancer was associated with age and KIF1A or EDNRB methylation. Methylation of EDNRB highly correlated with that of KIF1A(p<0.0001). On multivariable modeling, histologic diagnosis independently associated with EDNRB(p=0.0003) or KIF1A(p=0.027) methylation). A subset of patients analyzed (n=161) without prior biopsy proven malignancy received clinical risk classification based on examination. On univariate analysis, EDNRB and risk classification were associated with diagnosis of dysplasia/cancer, and remained significant on multivariate analysis (EDNRB:p=0.047, risk classification:p=0.008). Clinical risk classification identified dysplasia/cancer with a sensitivity of 71% and specificity of 58%. The sensitivity of clinical risk classification combined with EDNRB methylation improved to 75%.
EDNRB methylation in salivary rinses was independently associated with histologic diagnosis of premalignancy and malignancy and may have potential in classifying patients at risk for oral premalignant and malignant lesions in settings without access to a skilled dental practitioner. This may also potentially identify patients with premalignant and malignant lesions that do not meet criteria for high clinical risk based on skilled dental examination.
INTRODUCTION
Head and neck squamous cell carcinoma (HNSCC) accounts for greater than 37,000 new cases in the United States each year. The incidence of oral cavity cancer was approximated as 35,310 new cases per year and 7,590 estimated deaths.1 Over the years, improvements have been made in the diagnosis, management, and targeted therapies for these cancers. However, despite these advances a considerable number of patients continue to present with advanced stage disease. Advanced staged cancers have traditionally been associated with higher rates of mortality and decreased locoregional control rates. Intuitively, early detection of oral cancers would lead to improved quality of life and survival for these patients. The cost-effectiveness for screening in oral cavity cancers in high-risk patients has been previously demonstrated.2
The application of salivary rinses from high-risk patients has been explored as a potential for molecular screening in HNSCC. 3–7 The inactivation of tumor suppressor genes caused by epigenetic changes such as promoter region CpG island hypermethylation has been well established in the literature.8–10 The use of real time quantitative methylation-specific PCR (Q-MSP) provides a high throughput mechanism for detecting promoter hypermethylation in patient samples. The ability to quantify the methylation through Q-MSP allows for the potential of identifying high-risk patients with premalignant lesions. This has been previously demonstrated in salivary rinse samples obtained in lung cancer patients 11, oral cavity cancer patients 12 and oropharynx/hypopharynx cancer patients.3
We have previously published results of salivary rinse screening using promoter hypermethylation based markers in patients with previously diagnosed HNSCC.12 To date, however, the effectiveness of this strategy which includes salivary rinse samples collected in a prospective cohort at who are at risk for oral cancer and oral premalignancy has not yet been evaluated. In this study, we evaluated the utility of detection of methylation of two gene promoters, KIF1A and EDNRB, and the association of methylation of these promoter regions with the presence of oral cancer and premalignancy.
METHODS
Tissue Samples
Salivary rinse samples from 191 patients were prospectively obtained from the dental and oral medicine clinics at New York University College of Dentistry (NYUCD) (140 patients, 73.30%), University of Puerto Rico (39 patients, 20.42%), St. Vincent’s Cancer Center (1 patient, 0.52%), and Moffitt Cancer Center (11 patients, 5.76%). Institutional review board approval was obtained prior to the collection of the samples. A written informed consent was obtained from each subject. Enrollment included collection of demographic information and risk factor history (tobacco and alcohol). Inclusion criteria for enrollment were as follows: 1) English and/or Spanish speaking, 2) Age>18years, 3) the presence of a candidate oral epithelial lesion, and 4) the absence of a medical condition that would preclude a scalpel biopsy. Those lesions with an obvious etiology such as trauma, apthous ulceration, infection, or lichen planus were excluded. All patients were enrolled and assigned lesion risk classifications of low or high risk; this included an additional group of patients with histopathologically confirmed oral cancer who were also enrolled. Low risk groups were defined as having leukoplakia without the presence of erythroplakia, ulceration, erosion, or submucosal extension/induration. High risk patients exhibited one or more of following: leukoplakia with ulceration, erosion, or submucosal/extension; erythroplakia; erythroleukoplakia; or ulceration. If a subject presented with more than one lesion, each lesion was separately classified and the overall risk classification was designated based on the worst lesion.
Salivary rinses were obtained on all subjects as previously described.12 In brief, salivary rinses were then obtained by rinsing and gargling with 25cc of normal saline solution for 15 seconds. Three strokes were performed using cotton tipped applicators to collect exfoliated cells from the buccal mucosa, alveolar ridge, lateral tongue, floor of mouth, and pharyngeal inlet. The cellular materials from the applicators were also agitated and released into the salivary rinse specimen. This was then centrifuged to obtain a cell pellet after discarding the supernatant. Pellets were then immediately frozen and stored at −80°C. All low and high risk lesions underwent an incisional scalpel biopsy and a histologic diagnosis was obtained which categorized the patients into one of six histologic diagnoses. Each specimen was examined by two calibrated oral pathologists blinded to the results of any other clinical data. Each specimen was examined at multiple levels and, in addition to a standard description and diagnosis, each pathologist recorded a possible primary histopathology outcome classification: benign (with or without atypia), dysplasia (mild, moderate and severe grades), and invasive carcinoma (squamous cell carcinoma). Differences between oral pathologist classifications were resolved by joint review resulting in a consensus classification.
The tissue samples and clinical data were obtained from NYUCD in a collaborative effort by the Department of Otolaryngology-Head and Neck Surgery at Johns Hopkins Medical Institutions, Baltimore. These patients constituted a subset of patients analyzed as part of a study previously conducted under a U54 mechanism at the NYUCD (U54DE014257, Dr. David Sirois). The following procedures, including DNA extraction, bisultfite treatment, and Q-MSP, were all performed by two individuals blinded to the clinical data pertaining to the clinical risk classification and the histologic diagnoses.
DNA extraction
DNA obtained from the salivary rinse samples was extracted by the tissue bank by digestion with 50μg/ml of proteinase K (Boehringer) in the presence of 1% SDS at 48°C overnight followed by phenol/chloroform extraction and ethanol precipitation.
Bisulfite treatment
The DNA obtained from the salivary rinse samples were subjected to bisulfite treatment as has been previously described.13 Briefly, the EpiTect® Bisulfite Kit was used to for the conversion of 2μg of genomic DNA. The included Qiagen protocol was followed. After thermal denaturation and sodium bisulfite DNA conversion, the DNA was applied to an EpiTect spin plate. Optimized buffers and a vacuum manifold were used to wash and remove all traces of sodium bisulfate. The DNA was eluted. The eluted DNA was then ready for use for quantitative methylation-specific PCR.
Quantitative methylation-specific PCR
The bisulfite treated DNA was used as a template for flourescence based real-time QMSP as described previously.14 The EDNRB and KIF1A genes had been previously detected on a prior screen of salivary rinses in HNSCC patients.15 We had previously optimized the primer and probe sequences for Q-MSP. Briefly, primers and probes were designed specifically to amplify the bisulfite-converted DNA for the βACTIN, EDNRB, and KIF1A genes. βACTIN forward primer, 5′-TGGTGATGGAGG-AGGTTTAGTAAGT-3′, βACTIN reverse primer, 5′-AACCAATAAAACCTACTCCTCCCT-TAA-3′, and βACTIN TaqMan probe, 5′-ACCACCACCCAACACACAATAACAAACACA-3′. EDNRB forward primer, 5′-GGGAGTTGTAGTTTAGTTAGTTAGGGAGTAG-3′, EDNRB reverse primer, 5′-CCCGCGATTAAACTCGAAAA-3′, and EDNRB TaqMan probe, 5′-TTTTTATTCGTCGGGAGGAG-3′. KIF1A forward primer, 5′-GCGCGATAAATTAGTTGG-CGATT-3′, KIF1A reverse primer, 5′-CTCGACGACTACTCTACGCTAT-3′ and KIF1A TaqMan probe, 5′-CCTCCCGAAACGCTAATTAACTACGCG-3′. The ratios between the values of the EDNRB gene the reference gene βACTIN was obtained by TaqMan analysis and used as a measure for representing the relative quantity of methylation in a particular sample (value for gene of interest/value for βACTIN gene × 100). Flourogenic PCRs were carried out in a reaction volume of 10 μl 300nmol/L of each primer; 100nmol/L of probe; .375 unite of platinum Taq polymerase (Invitrogen); 100μmol/L of each dATP, dCTP, dGTP, and dTTP; 100nmol/L of ROX Reference Dye (Invitrogen); 8.4mmol/L ammonium sulfate; 33.5mmol/L Trizma (Sigma); 3.35 mmol/L magnesium chloride; 5mmol/L mercaptoethanol; and 0.05% DMSO. Each real time Q-MSP reaction consisted of 1.5μl of treated DNA solution. Amplifications were carried out in 384-well plates in a 7900 Sequence Detector System (Perkin-Elmer Applied Biosystems). Thermal cycling was initiated with a first denaturation step at 95°C for 2min followed by 50 cycles of 95°C for 15 s and 60°C for 1 min. Each reaction was done in triplicate; the average of the triplicate was considered for analysis. The triplicate reactions also provided evidence of reproducibility of the individual reactions. Standardization was obtained by collecting leukocytes from a healthy individual that were subsequently methylated in vitro with excess Sss1 methyltransferase (New England Biolabs) to generate completely methylated DNA. This DNA was then Bisulfite treated as described above. Serial dilutions of this DNA were used for constructing the calibration curves on each plate. A separate sample of leukocytes from a healthy individual was obtained and only Bisulfite treatment was performed on the samples. These samples were used as a negative control for the reactions. There were also several control wells in each plate that contained only the reaction mix and water to ensure there was no contamination. Results for Q-MSP were analyzed considering the quantity of methylation (normalized by βACTIN) as well as considering the quantity of methylation as a binary event, in which any quantity of methylation in a sample would be considered positive for methylation.
Statistical Analysis
Proportions of EDNRB or KIF1A gene methylation were compared between patient salivary rinse samples. The initial analyses were performed with a cohort including 191 patients, where the subcategories of histologic outcome included benign, premalignant and malignant. Two pre-specified candidate genes (e.g., EDNRB and KIF1A) were evaluated. Gene hypermethylation was dichotomized at zero (i.e., no methylation vs. any methylation). Predictors associated with head and neck cancers were evaluated as well, including age, gender, race, smoking status, and alcohol consumption. Age was analyzed as a continuous variable, whereas all the other variables were considered as categorical variables. Univariate and multivariable proportional odds modeling were constructed sequentially to first explore the association of the variables of interest with the histologic outcome. Variables of significance based on the univariate models (p<.20) along with those deemed to be biologically/clinically important were retained for further analysis. Simultaneous effects expressed by these variables were studied using the multivariable proportional odds model. Odds ratios were reported with 95% confidence intervals, which indicated the strength of the association and its uncertainty. Throughout the analyses, proportional odds assumption of common slope for all of the cumulative logits was checked by Score test.
A secondary analysis explored the association of methylation that is independent of other predictors of histology, where patients with known cancer prior to coming to the clinic were excluded (n=30). This subset of 161 patients, i.e. the patient cohort not including the known cancer patients, was categorized as benign vs. dysplasia/cancer exploring the association of clinical risk classification to histopathology. Univariate and multivariable logistic regression analyses were performed using the same biologically/clinically important covariates as described above.
Receiver operating characteristic (ROC) analyses were conducted to estimate classification accuracy, sensitivity (true-positive rate) and specificity (false-positive rate) of the predictor along with 95% confidence intervals. AUC, an index of predictive power, was also provided. A logistic prediction model using the clinical risk classification combined with methylation of EDNRB was developed and internally validated. Statistical analyses were performed using SAS (v 9.2, SAS Institute, Cary, NC) and STATA software (v 8.2, College Station, Texas, and all statistical tests were two-sided with p < 0.05 considered statistically significant.
RESULTS
Population Characteristics
In analyzing the population characteristics of patients with oral cavity lesions (n=191) we found that the majority of patients were males and Caucasian. There were 68.1%, 67.4%, and 74.3% males in benign, dysplasia, and cancer categories respectively. We noted also noted that there were 68.1%, 74.4%, and 68.6% Caucasians in benign, dysplasia, and cancer categories respectively. Median age was 54 years and ranged from 18 to 90 years in the entire cohort. The baseline characteristics of the groups were similar (Table 1). Furthermore, when assessing the exposure to clinically relevant risk factors such as tobacco and alcohol consumption, again we noted that the groups were inherently well matched. Tobacco consumption (current or past) was analyzed as a categorical variable and was noted in 67.3% of patients in the benign category and 69.8% and 74.3% in the dysplasia and cancer categories respectively. Alcohol consumption was analyzed as a binary variable where use was noted in 70.8%, 74.4%, and 77.1% of patients in the benign, dysplasia, and cancer categories respectively. An exact chi-square test, however, did reveal a statistically significant association between risk classification and histologic diagnosis in our 161 patient cohort (p<0.008).
Table 1.
Univariate analysis of association between predictors and histology (n=191)
| Variable | Benign (n=113) | Premalignant (n=43) | Malignant (n=35) | Total |
|---|---|---|---|---|
| Age (years) | ||||
| Mean (±SD) | 52 (± 13) | 56 (± 14) | 58 (± 14) | 54 (± 13) |
| Median (range) | 52 (18 – 78) | 57 (26 – 82) | 56 (24 – 90) | 54 (18 – 90) |
| Gender, n [%] | ||||
| Female | 36 (31.9) | 14 (32.6) | 9 (25.7) | 59 (30.9) |
| Male | 77 (68.1) | 29 (67.4) | 26 (74.3) | 132 (69.1) |
| Race, n [%] | ||||
| African-American | 28 (24.8) | 9 (20.9) | 7 (20.0) | 44 (23.0) |
| Caucasian | 77 (68.1) | 32 (74.4) | 24 (68.6) | 133 (69.6) |
| Other | 8 (7.1) | 2 (4.7) | 4 (11.4) | 14 (7.3) |
| Tobacco, n [%] | ||||
| Never user | 37 (32.7) | 13 (30.2) | 9 (25.7) | 59 (30.9) |
| Former user | 20 (17.7) | 7 (16.3) | 11 (31.4) | 38 (19.9) |
| Current user | 56 (49.6) | 23 (53.5) | 15 (42.9) | 94 (49.2) |
| Ethanol, n [%] | ||||
| Never used | 33 (29.2) | 11 (25.6) | 8 (22.9) | 52 (27.2) |
| Used | 80 (70.8) | 32 (74.4) | 27 (77.1) | 139 (72.8) |
| Risk classification, n [%] | ||||
| Low risk | 75 (66.4) | 21 (48.8) | 0 (0) | 96 (50.3) |
| High risk | 38 (33.6) | 21 (48.8) | 6 (17.1) | 65 (34.0) |
| Cancer | 0 (0) | 1 (2.3) | 29 (82.9) | 30 (15.7) |
| EDNRB, n [%] | ||||
| Unmethylated | 88 (77.9) | 26 (60.5) | 14 (40.0) | 128 (67.0) |
| Methylated | 25 (22.1) | 17 (39.5) | 21 (60.0) | 63 (33.0) |
| KIF1A, n [%] | ||||
| Unmethylated | 84 (74.3) | 31 (72.1) | 17 (48.6) | 132 (69.1) |
| Methylated | 28 (24.8) | 12 (27.9) | 18 (51.4) | 58 (30.4) |
| Missing | 1 (0.9) | 0 (0) | 0 (0) | 1 (0.5) |
The presence of methylated EDNRB promoter is associated with oral premalignancy and malignancy
A univariate analysis was then conducted to evaluate the association between histologic diagnosis and variables including age, gender, race, tobacco consumption, and alcohol consumption. In the analysis, we also included methylation status of EDNRB and KIF1A as potential predictors of histology (Table 2). Based on the univariate modeling we observed that age was significantly associated with the diagnosis of benign, dysplasia, or cancer (OR=1.3, 95% CI=(1.1–1.6), p=0.014). Associations between histologic diagnosis and EDNRB and KIF1A methylation status were also found to be statistically significant (OR=3.6, 95% CI=(2.0–6.4), p<0.0001 and OR=2.2, 95% CI=(1.2–3.9), p=0.011 respectively). We did not observe a significant association with gender (p=0.618) or race (p=0.730) in the analysis. Although tobacco consumption (p=0.372) and alcohol consumption (p=0.435) did not suggest a significant association with histologic diagnosis in our cohort of patients, these variables have a well established role in the progression to oral cancer.
Table 2.
Analyses of association between predictors and histology (n=191) -EDNRB
| Variable | Univariate analysis |
Multivariable analysis |
||
|---|---|---|---|---|
| OR1 (95% CI) | P1 | Adjusted OR2 (95% CI) | Adjusted P2 | |
| Age (years) | 1.33 (1.1 – 1.6) | 0.014 | 1.23 (1.0 – 1.6) | 0.088 |
| Gender | ||||
| Female | Ref | Ref | ||
| Male | 1.2 (0.6 – 2.1) | 0.618 | 0.9 (0.5 – 1.9) | 0.896 |
| Race | ||||
| African-American | Ref | Ref | ||
| Caucasian | 1.2 (0.6 – 2.5) | 0.730 | 1.1 (0.6 – 2.4) | 0.745 |
| Other | 1.5 (0.5 – 4.9) | 1.6 (0.5 – 5.5) | ||
| Tobacco | ||||
| Never user | Ref | Ref | ||
| Former user | 1.7 (0.8 – 3.8) | 0.372 | 1.7 (0.7 – 3.9) | 0.454 |
| Current user | 1.1 (0.6 – 2.2) | 1.4 (0.7 – 3.0) | ||
| Ethanol | ||||
| Never used | Ref | Ref | ||
| Used | 1.3 (0.7 – 2.4) | 0.435 | 1.2 (0.6 – 2.3) | 0.663 |
| EDNRB | ||||
| Unmethylated | Ref | Ref | ||
| Methylated | 3.6 (2.0 – 6.4) | < .0001 | 3.1 (1.7 – 5.8) | 0.0003 |
Univariate proportional odds model;
Multivariate proportional odds model;
Unit of 10 years
EDNRB and KIF1A were highly correlated (p<0.0001) so that their individual effects on the observed histology could not be clearly separated if they were included simultaneously in the model. We then chose EDNRB along with other variables of significance suggested by the univariate model (e.g., age) as well as those felt to play a clinically and biologically significant role in the development of oral cavity cancer (tobacco and alcohol consumption) to perform a multivariable proportional odds logistic regression analysis (Table 2). Once again, tobacco (p=0.454) and alcohol (p=0.663) consumption did not reach statistical significance. Age on multivariable analysis had an odds ratio of 1.2 (1.0–1.6) but only reached borderline significance (p=0.088). EDNRB methylation, however, remained significantly associated with histologic diagnosis (OR=3.1, 95% CI=(1.7–5.8), p=0.0003). These data indicate that EDNRB hypermethylation is an independent predictor of histologic diagnosis. Patients with EDNRB methylated were about 3 times as likely to have a diagnosis of premalignancy or malignancy as compared to those with EDNRB unmethylated, after adjusting for age, tobacco use, and alcohol consumption.
EDNRB salivary rinse methylation analysis can improve risk classification when combined with clinical risk classification
A subset of our patient cohort initially presented with a known diagnosis of cancer. To obtain insight into the performance of clinical risk classification and salivary rinse methylation, we excluded those patients who had a diagnosis of known cancer on presentation (n=30) from the cohort. The remaining patients (n=161) were classified using consensus risk classification methods based on clinical appearance of the oral lesion into were high risk (n=21) and low risk (n=140) lesions.16
We used the same predictors as used in our initial analysis and added risk classification designated during the initial examination to construct a univariate logistic regression model as shown in Table 3. Age, gender, race, tobacco, and alcohol consumption, and KIF1A methylation status were not significantly associated with diagnosis. Risk classification, described as low risk or high risk, based on clinical examination was significantly associated with histologic diagnosis (OR=2.5, 95% CI=(1.3–5.1), p=0.008). EDNRB methylation status on univariate modeling was significant (OR=2.1, 95% CI=(1.0–4.4), p=0.046). We expanded the analysis to determine if there was any correlation between EDNRB and KIF1A. A Spearman correlation coefficient was calculated as R=0.39 (p<0.0001) and a scatter plot was constructed to better visualize if there appeared to be a correlation between the two markers (Figure 1). The correlation between EDNRB and KIF1A was further evaluated by calculating the Phi Coefficient to measure the degree of correlation between two categorical variables which was 0.225. Although these results may be limited because of the use of EDNRB and KIF!a as binary variables (methylated vs. unmethylated) and the sample size our findings suggested only a small relationship between EDNRB and KIF1A.
Table 3.
Univariate analysis of association between predictors and histology – excluding patients with known cancer at presentation (n=161)
| Variable | Benign (n=113) | Dysplasia/cancer (n=48) | Odds ratio1 (95% CI) | P1 |
|---|---|---|---|---|
| Age (years) | ||||
| Mean (± SD) | 52 (± 13) | 56 (± 15) | 1.22 (0.9 – 1.6) | 0.110 |
| Median (range) | 52 (18 – 78) | 57 (24 – 86) | ||
| Gender, n [%] | ||||
| Female | 36 (31.9) | 15 (31.3) | Ref | |
| Male | 77 (68.1) | 33 (68.7) | 1.0 (0.5 – 2.1) | 0.940 |
| Race, n [%] | ||||
| African-American | 28 (24.8) | 10 (20.8) | Ref | |
| White | 77 (68.1) | 36 (75.0) | 1.3 (0.6 – 3.0) | 0.639 |
| Other | 8 (7.1) | 2 (4.2) | 0.7 (0.1 – 3.9) | |
| Tobacco, n [%] | ||||
| Never user | 37 (32.7) | 14 (29.2) | Ref | |
| Former user | 20 (17.7) | 10 (20.8) | 1.3 (0.5 – 3.5) | 0.855 |
| Current user | 56 (49.6) | 24 (50.0) | 1.1 (0.5 – 2.5) | |
| Ethanol, n [%] | ||||
| Neverused | 33 (29.2) | 12 (25.0) | Ref | |
| Used | 80 (70.8) | 36 (75.0) | 1.2 (0.6 – 2.7) | 0.587 |
| Risk classification, n [%] | ||||
| Low risk | 75 (66.4) | 21 (43.8) | Ref | |
| High risk | 38 (33.6) | 27 (56.2) | 2.5 (1.3 – 5.1) | 0.008 |
| EDNRB, n [%] | ||||
| Unmethylated | 88 (77.9) | 30 (62.5) | Ref | |
| Methylated | 25 (22.1) | 18 (37.5) | 2.1 (1.0 – 4.4) | 0.046 |
| KIF1A, n [%] | ||||
| Unmethylated | 84 (74.3) | 35 (72.9) | Ref | |
| Methylated | 28 (24.8) | 13 (27.1) | 1.1 (0.5 – 2.4) | 0.782 |
| Missing3 | 1 (0.9) | 0 (0) | ||
Univariate logistic regression model;
Unit of 10 years;
Excluded from the analysis of association
Figure 1.
Correlation between EDNRB and KIF1A
We then proceeded to construct a multivariable model where we analyzed risk classification, age, gender, race, tobacco, ethanol, and EDNRB methylation status (Table 4). Once again, risk classification remained independently associated with a histologic diagnosis (OR=2.6, 95% CI=(1.3–5.2), p=0.008). And similarly, EDNRB methylation demonstrated statistical significance in being independently associated with a diagnosis of dysplasia or cancer (OR=2.1, 95% CI=(1.0–4.6), p=0.047).
Table 4.
Risk factors for histologic diagnosis by multivariable analysis(n=161)
| Variable | Adjusted OR1 (95% CI) | Adjusted P1 |
|---|---|---|
| Age (years) | 1.12 (0.8 – 1.5) | 0.455 |
| Gender | ||
| Male vs. female | 0.9 (0.4 – 2.0) | 0.763 |
| Race | ||
| Caucasian vs. African-American | 1.3 (0.5 – 3.2) | 0.547 |
| Othervs. African-American | 0.7 (0.1 – 4.0) | 0.659 |
| Tobacco | ||
| Former uservs. never user | 1.5 (0.5 – 4.4) | 0.457 |
| Currentuservs. never user | 1.6 (0.7 – 3.8) | 0.296 |
| Ethanol | ||
| Used vs. never used | 1.3 (0.6 – 3.0) | 0.561 |
| Risk classification | ||
| High risk vs. low risk | 2.6(1.3 – 5.2) | 0.008 |
| EDNRB | ||
| Methylated vs. unmethylated | 2.1 (1.0 – 4.6) | 0.047 |
Multivariate logistic regression model;
Unit of 10 years
Through logistic regression modeling and ROC analyses, we then calculated the sensitivities and specificities of risk classification strategy and EDNRB methylation status (Table 5). Quantitative EDNRB was also considered to assess whether a cutoff other than zero may result in better predictive accuracy. The area under the curve (AUC) was also calculated with a 95% confidence interval. Risk classification as a sole predictor of histology outcome had a sensitivity of 71% (95% CI=56–83%), and a specificity of 58% (95% CI=48–67%). The AUC for risk classification was 0.65 (95% CI=0.56-.075). EDNRB methylation in salivary rinse samples as a predictor of histologic diagnosis had a sensitivity of 65% (95% CI= 49–78%) specificity of 51% (95% CI=42–61%) and AUC of 0.61 (95% CI=0.51–0.71) when treated as a binary value. As described in Table 5, the ROC curves with adjustment for other covariates are shown in Figure 2 demonstrating the AUC for EDNRB alone (0.61), risk classification alone (0.65), and the combination of risk classification and EDNRB (0.68).
Table 5.
Predictive accuracy of risk classification and EDNRB and combination after adjusting for other predictors1 associated with head and neck cancer (n=161)
| Predictor | Cutoff2 | Sensitivity (%, 95% CI) | Specificity (%, 95% CI) | PPV3 (%, 95% CI) | NPV4 (%, 95% CI) | AUC (95% CI) |
|---|---|---|---|---|---|---|
| Risk classification | 0.267590 | 71(56 – 83) | 58(48 – 67) | 41 (31 – 53) | 77 (66 – 86) | 0.65(0.56 – 0.75) |
| EDNRB | 0.274459 | 65(49 – 78) | 51(42 – 61) | 36 (26 – 47) | 82 (72 – 90) | 0.61 (0.51 – 0.71) |
| Risk and EDNRB | 0.2467921 | 75(60 – 86) | 50(41 – 60) | 39 (29 – 50) | 83 (72 – 91) | 0.68(0.58 – 0.77) |
| combined | 0.4501881 | 21 (10 – 35) | 92(85 – 96) | 53 (29 – 76) | 73 (65 – 80) |
Other predictors include age, sex, race, tobacco, and ethanol use;
Based on predicted probability of high grade dysplasia/cancerusing multivariable logistic regression model;
Positive predictive value, depending on the prevalence of the disease (high grade dysplasia/cancer) which was 13% for this study population.
Negative predictive value, depending on the prevalence of the disease (high grade dysplasia/cancer) which was 13% for this study population.
Note: The cutoffs are not EDNRB methylation values but a predicted probability from the logistic regression model that simultaneously includes risk classification and EDNRB.
Figure 2.
ROC curves with adjustment for other covariates as described in Table 5
We also included the positive (PPV) and negative (NPV) predictive values in our analysis. Individuals with a positive test had a 41% (95% CI=31–53%) chance of having high grade dysplasia/cancer based solely on risk classification. The PPV increased to 53% (95% CI=29–76%) with the combination of risk classification and EDNRB when optimizing for specificity. We saw a similar increase in the NPV with the combination of risk classification and EDNRB from 77% (95% CI=66–86%) to 83% (95% CI=72–91%) when optimizing for sensitivity.
We then performed logistic regression analysis and analyzed the variables by combining risk classification and EDNRB methylation to determine if there was any improvement in the predictive capability. Optimal cutoffs to maximize either sensitivity or specificity were obtained based on the predicted probability of high grade dysplasia/cancer from the multivariable logistic regression model. Using the combination of risk classification and EDNRB at the selected cutoff to maximize specificity resulted in a specificity of 92% (95% CI=85–96%) and sensitivity of 21% (95% CI=10–35%) with the AUC of 0.68 (95% CI=0.58–0.77). At a cutoff threshold to maximize sensitivity, addition of EDNRB methylation improved sensitivity to 75% (95% CI=60–86%) but decreased specificity from to 50% (95% CI=41–60%). In practical terms, the application of EDNRB salivary rinse methylation to define a high risk category of patients within a cohort of 161 patients without prior biopsy would have changed the clinical low risk assessment of 5 patients with mild dysplasia and 3 patients with severe dysplasia to high risk based on methylation analysis of salivary rinses.
DISCUSSION
Aberrant promoter hypermethylation has been recently proposed as a means for detection of HNSCC in salivary rinses. 3–7 We have studied a large cohort of prospectively collected salivary rinses obtained from patients with benign, dysplastic, and cancer diagnoses to determine the ability of Q-MSP to detect EDNRB or KIF1A promoter hypermethylation in high-risk patients. Due to the sensitivity of the Q-MSP technique used to detect the presence of EDNRB gene or KIF1A gene methylation, this enabled us to accurately correlate the risk classification strategy to the methylation status of the samples. This is the first report demonstrating use of molecular markers in salivary rinse samples for detection of premalignant oral disease.
Our group previously published the utility of evaluating promoter region methylation status of various genes as a tool for detection of HNSCC.12 EDNRB hypermethylation has been studied extensively in prostate cancers with the potential of diagnostic as well as prognostic value.17–23 In addition, EDNRB methylation status has been studied in a variety of other cancers including lung cancer24, bladder cancer25, 26, hepatocellular carcinoma27, nasopharyngeal carcinoma28 and others29, 30. More recently, our lab demonstrated the presence of KIF1A promoter hypermethylation in breast cancer.31 We have also been successful in discovering that EDNRB and KIF1A are preferentially methylated in salivary rinse samples of HNSCC (see supporting manuscript 1 available as online submission)15 and aberrant methylation of these genes is also highly prevalent in a cohort of Indian oral squamous cell carcinoma (see supporting manuscript 2 available as online submission). Based on these prior data and data reported in various other cancers including HNSCC we selected EDNRB and KIFIA as our primary genes of interest for our study. This study evaluates promoter hypermethylation of the EDNRB gene and KIF1A gene in salivary rinse samples from patients with benign, dysplastic, and malignant lesions in combination with a risk classification strategy. Furthermore, it demonstrates the effectiveness of quantitative measurement of promoter hypermethylation in a significant sized cohort of oral cavity salivary rinse samples as a potential tool for assessing risk of malignancy, and detecting dysplastic or cancer cells.
We observed that the presence of EDNRB promoter methylation in salivary rinses was associated with the presence of dysplasia or invasive cancer, and that this was independent of clinical covariates including age and exposure history. This confirms that epigenetic alterations specific to dysplasia or invasive cancer can be detected in salivary rinses in the context of a dental clinic designed to assess patients at risk for oral cancer.
In our analysis we found that risk classification by a specialist resulted in a 71% sensitivity in screening indiviuals with oral cavity lesions for dysplasia or cancer. This underscores the significance of clinical examination and risk classification as a gold standard of initial screening performed by highly trained individuals such as otolaryngologists/dentists. In addition, the combination of EDNRB to risk classification when using a selected cutoff threshold from the logistic regression analysis allowed increased sensitivity of 75% with moderate decrease in specificity from 58% to 50%. In our analysis, we did observe considerable variablility in the sensitivity and specificity based on selected cutoff values. Ideally, we would like to observe an increase in the sensitivity without a substantial drop in the specificity with the addition of EDNRB as a predictive variable. However, we noted that even in clinical situations where a skilled professional has designated a risk classification based on clinical examination, the addition of EDNRB methylation status may change risk assessment of a lesion, prompting biopsy of a dysplastic or malignant oral lesion that would have otherwise not met clinical criteria for biopsy. Our examined cohort has several demographic and exposure characteristics associated with oral cancer, including tobacco and ethanol exposure and advanced age. Therefore, EDNRB promoter region methylation in salivary rinses status may be useful as a risk assessment tool in patients evaluated for potential oral malignancy based on exposure history and age, but presenting with a clinically low risk lesion.
The fundamental necessity of early detection in oral cancer can be confounded my numerous barriers. This can be to due the lack of a trained dentist/head and neck surgeon/otolaryngologist in the community as well as a basic lack of education/awareness in the public and health professionals.32, 33 In addition, lack of access to health care can also prevent patients from seeking care to facilitate earlier detection.34 The ideal test for oral premalignancy and cancer would be available for administration to a high risk population, and administered by health care workers without specialized training, yet still provide predictive outcome results. Q-MSP provides a cost-effective easy to carry-out method that allows high-throughput and rapid analysis. This would indicate the potential use of this technique as a means for early detection of dysplastic oral cavity lesions and reinforces the potential usefulness of obtaining salivary rinses as a screening and surveillance strategy.
In a clinical setting, highly trained professionals examine patients with oral cavity lesions through conventional techniques of physical examination. Other established adjuncts include oral cytology, toluidine blue, and light-based detection systems.35 Based on history, clinical/environmental risk factors, and oral examination parameters, the clinician sets a threshold whereby those patients that are deemed to be at a higher risk of having oral cancers undergo the gold standard scalpel biopsy. This methodology often times results in a population of patients that are categorized as low risk patients clinically, however may in fact harbor dysplastic lesions.36 Our study unveils the presence of false negatives with using a clinically based risk classification system alone by identifying patients who had a low risk clinical lesion but had a histologic diagnosis of dysplasia. We established that the increase in sensitivity by combining both risk classification and EDNRB methylation status resulted from recognizing those clinical low risk patients that in fact had dysplastic lesions. Due to the lack of long term clinical outcomes for our patient cohort at this time, the true clinical implications of identifying more dysplastic lesions than would be discernable by a clinical risk stratification strategy alone needs to be interpreted with caution. Although it is not apparent which patients with dysplastic lesions may proceed to developing cancer, we acknowledge that these patients do most certainly warrant a more vigilant follow up.
The use of molecular markers in salivary rinses for the detection of cancer or those harboring occult cancers has been explored with the intent to improve screening accuracy and cost effectiveness. Salivary rinse samples potentially carry whole cells, DNA, RNA and proteins which allows for the capability of detecting alterations leading to cancer. Increasingly, saliva has been used to diagnose infectious diseases, hereditary disorders, autoimmune diseases, and endocrine disorders. Rosas et al. published the first study to demonstrate detection of aberrant promoter hypermethylation in saliva from HNSCC patients.6 Carvalho et al. identified differential hypermethylation patterns in salivary rinses and serum of patients with HNSCC in a panel of eight genes by Q-MSP.12 Similarly, Lallemant et al. analyzed the expression levels of nine genes in HNSCC and control salivary rinse samples by rT-qPCR.4 Zhao et al. explored the feasibility of DNA PCR to screen for HPV in salivary rinse samples of head and neck cancer patients.3 The novelty of this paper, however, is its prospective nature and the inclusion of salivary rinse samples from patients with premalignant lesions in the oral cavity. Further investigations of the functionality of EDNRB and downstream pathways may yield additional insight to its role in oral cavity lesions.
Future studies using CpG island microarray technologies may be useful in creating helpful panels of genes with increased sensitivities and retained specificities. Future studies to investigate the progression from premalignant changes to malignant transformation and the timing of the aberrant hypermethylation will also be of great value. Nonetheless, the use of salivary rinse molecular analysis may offer a feasible, rapid and cost-effective tool for stratification of high-risk patients and early detection of premalignant lesions.
Supplementary Material
Acknowledgments
FUNDING: This work was supported by National Institute of Dental and Craniofacial Research and National Institutes of Health SPORE GRANT 5P50DE019032 and NIDCR/NIH: U54 DE14257
ACKNOWLEDGMENTS: None
This work was supported by the National Institute of Dental and Craniofacial Research and National Institutes of Health (SPORE GRANT 5P50DE019032), the Early Detection Research Network (EDRN) grant U01-CA084986. The funding agency had no role in the design of the study; data collection, or analysis; in the interpretation of the results; or in the preparation of the manuscript; and the decision to submit the manuscript for publication. D. Sidransky owns Oncomethylome Sciences, SA stock, which is subject to certain restrictions under University policy. Dr. Sidransky is a paid consultant to Oncomethylome Sciences, SA and is a paid member of the company’s Scientific Advisory Board.
ABBREVIATIONS
- EDNRB
endothelin receptor type B
- KIF1A
kinesin chain member 1A
- HNSCC
head and neck squamous cell carcinoma
- Q-MSP
quantitative methylation-specific PCR
Footnotes
Conflict of interest: Dr. Califano is the Director of Research of the Milton J. Dance Head and Neck Endowment. The terms of this arrangement are being managed by the Johns Hopkins University in accordance with its conflict of interest policies.
The Johns Hopkins University in accordance with its conflict of interest policies is managing the terms of this agreement.
References
- 1.American ACSCFFA, 2008. CS.
- 2.Speight PM, Palmer S, Moles DR, et al. The cost-effectiveness of screening for oral cancer in primary care. Health Technol Assess. 2006;10(14):1–144. iii–iv. doi: 10.3310/hta10140. [DOI] [PubMed] [Google Scholar]
- 3.Zhao M, Rosenbaum E, Carvalho AL, et al. Feasibility of quantitative PCR-based saliva rinse screening of HPV for head and neck cancer. Int J Cancer. 2005;117(4):605–10. doi: 10.1002/ijc.21216. [DOI] [PubMed] [Google Scholar]
- 4.Lallemant B, Evrard A, Combescure C, et al. Clinical relevance of nine transcriptional molecular markers for the diagnosis of head and neck squamous cell carcinoma in tissue and saliva rinse. BMC Cancer. 2009;9:370. doi: 10.1186/1471-2407-9-370. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Nunes DN, Kowalski LP, Simpson AJ. Detection of oral and oropharyngeal cancer by microsatellite analysis in mouth washes and lesion brushings. Oral Oncol. 2000;36(6):525–8. doi: 10.1016/s1368-8375(00)00045-2. [DOI] [PubMed] [Google Scholar]
- 6.Rosas SL, Koch W, da Costa Carvalho MG, et al. Promoter hypermethylation patterns of p16, O6-methylguanine-DNA-methyltransferase, and death-associated protein kinase in tumors and saliva of head and neck cancer patients. Cancer Res. 2001;61(3):939–42. [PubMed] [Google Scholar]
- 7.El-Naggar AK, Mao L, Staerkel G, et al. Genetic heterogeneity in saliva from patients with oral squamous carcinomas: implications in molecular diagnosis and screening. J Mol Diagn. 2001;3(4):164–70. doi: 10.1016/S1525-1578(10)60668-X. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Herman JG, Jen J, Merlo A, Baylin SB. Hypermethylation-associated inactivation indicates a tumor suppressor role for p15INK4B. Cancer Res. 1996;56(4):722–7. [PubMed] [Google Scholar]
- 9.Merlo A, Herman JG, Mao L, et al. 5′ CpG island methylation is associated with transcriptional silencing of the tumour suppressor p16/CDKN2/MTS1 in human cancers. Nat Med. 1995;1(7):686–92. doi: 10.1038/nm0795-686. [DOI] [PubMed] [Google Scholar]
- 10.Ha PK, Califano JA. Promoter methylation and inactivation of tumour-suppressor genes in oral squamous-cell carcinoma. Lancet Oncol. 2006;7(1):77–82. doi: 10.1016/S1470-2045(05)70540-4. [DOI] [PubMed] [Google Scholar]
- 11.Palmisano WA, Crume KP, Grimes MJ, et al. Aberrant promoter methylation of the transcription factor genes PAX5 alpha and beta in human cancers. Cancer Res. 2003;63(15):4620–5. [PubMed] [Google Scholar]
- 12.Carvalho AL, Jeronimo C, Kim MM, et al. Evaluation of promoter hypermethylation detection in body fluids as a screening/diagnosis tool for head and neck squamous cell carcinoma. Clin Cancer Res. 2008;14(1):97–107. doi: 10.1158/1078-0432.CCR-07-0722. [DOI] [PubMed] [Google Scholar]
- 13.Herman JG, Graff JR, Myohanen S, Nelkin BD, Baylin SB. Methylation-specific PCR: a novel PCR assay for methylation status of CpG islands. Proc Natl Acad Sci U S A. 1996;93(18):9821–6. doi: 10.1073/pnas.93.18.9821. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Harden SV, Tokumaru Y, Westra WH, et al. Gene promoter hypermethylation in tumors and lymph nodes of stage I lung cancer patients. Clin Cancer Res. 2003;9(4):1370–5. [PubMed] [Google Scholar]
- 15.Demokan S, Xiaofei Chang, Chuang A, Mydlarz WK, Kaur J, Huang P, Khan Z, et al. KIF1A and EDNRB are differentially methylated in primary HNSCC and salivary rinses. Int J Cancer. 2009 doi: 10.1002/ijc.25248. submitted. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Kramer IR, Pindborg JJ, Bezroukov V, Infirri JS. Guide to epidemiology and diagnosis of oral mucosal diseases and conditions. World Health Organization. Community Dent Oral Epidemiol. 1980;8(1):1–26. doi: 10.1111/j.1600-0528.1980.tb01249.x. [DOI] [PubMed] [Google Scholar]
- 17.Bastian PJ, Ellinger J, Heukamp LC, Kahl P, Muller SC, von Rucker A. Prognostic value of CpG island hypermethylation at PTGS2, RAR-beta, EDNRB, and other gene loci in patients undergoing radical prostatectomy. Eur Urol. 2007;51(3):665–74. doi: 10.1016/j.eururo.2006.08.008. discussion 74. [DOI] [PubMed] [Google Scholar]
- 18.Bastian PJ, Palapattu GS, Yegnasubramanian S, et al. CpG island hypermethylation profile in the serum of men with clinically localized and hormone refractory metastatic prostate cancer. J Urol. 2008;179(2):529–34. doi: 10.1016/j.juro.2007.09.038. discussion 34–5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Ellinger J, Bastian PJ, Jurgan T, et al. CpG island hypermethylation at multiple gene sites in diagnosis and prognosis of prostate cancer. Urology. 2008;71(1):161–7. doi: 10.1016/j.urology.2007.09.056. [DOI] [PubMed] [Google Scholar]
- 20.Jeronimo C, Henrique R, Campos PF, et al. Endothelin B receptor gene hypermethylation in prostate adenocarcinoma. J Clin Pathol. 2003;56(1):52–5. doi: 10.1136/jcp.56.1.52. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Woodson K, Gillespie J, Hanson J, et al. Heterogeneous gene methylation patterns among pre-invasive and cancerous lesions of the prostate: a histopathologic study of whole mount prostate specimens. Prostate. 2004;60(1):25–31. doi: 10.1002/pros.20013. [DOI] [PubMed] [Google Scholar]
- 22.Nelson JB, Lee WH, Nguyen SH, et al. Methylation of the 5′ CpG island of the endothelin B receptor gene is common in human prostate cancer. Cancer Res. 1997;57(1):35–7. [PubMed] [Google Scholar]
- 23.Yegnasubramanian S, Kowalski J, Gonzalgo ML, et al. Hypermethylation of CpG islands in primary and metastatic human prostate cancer. Cancer Res. 2004;64(6):1975–86. doi: 10.1158/0008-5472.can-03-3972. [DOI] [PubMed] [Google Scholar]
- 24.Chen SC, Lin CY, Chen YH, et al. Aberrant promoter methylation of EDNRB in lung cancer in Taiwan. Oncol Rep. 2006;15(1):167–72. [PubMed] [Google Scholar]
- 25.Friedrich MG, Weisenberger DJ, Cheng JC, et al. Detection of methylated apoptosis-associated genes in urine sediments of bladder cancer patients. Clin Cancer Res. 2004;10(22):7457–65. doi: 10.1158/1078-0432.CCR-04-0930. [DOI] [PubMed] [Google Scholar]
- 26.Yates DR, Rehman I, Abbod MF, et al. Promoter hypermethylation identifies progression risk in bladder cancer. Clin Cancer Res. 2007;13(7):2046–53. doi: 10.1158/1078-0432.CCR-06-2476. [DOI] [PubMed] [Google Scholar]
- 27.Hsu LS, Lee HC, Chau GY, Yin PH, Chi CW, Lui WY. Aberrant methylation of EDNRB and p16 genes in hepatocellular carcinoma (HCC) in Taiwan. Oncol Rep. 2006;15(2):507–11. [PubMed] [Google Scholar]
- 28.Lo KW, Tsang YS, Kwong J, To KF, Teo PM, Huang DP. Promoter hypermethylation of the EDNRB gene in nasopharyngeal carcinoma. Int J Cancer. 2002;98(5):651–5. doi: 10.1002/ijc.10271. [DOI] [PubMed] [Google Scholar]
- 29.Hsiao PC, Liu MC, Chen LM, et al. Promoter methylation of p16 and EDNRB gene in leukemia patients in Taiwan. Chin J Physiol. 2008;51(1):27–31. [PubMed] [Google Scholar]
- 30.Zhao BJ, Sun DG, Zhang M, Tan SN, Ma X. Identification of aberrant promoter methylation of EDNRB gene in esophageal squamous cell carcinoma. Dis Esophagus. 2009;22(1):55–61. doi: 10.1111/j.1442-2050.2008.00848.x. [DOI] [PubMed] [Google Scholar]
- 31.Ostrow KL, Park HL, Hoque MO, et al. Pharmacologic unmasking of epigenetically silenced genes in breast cancer. Clin Cancer Res. 2009;15(4):1184–91. doi: 10.1158/1078-0432.CCR-08-1304. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Ahluwalia KP, Yellowitz JA, Goodman HS, Horowitz AM. An assessment of oral cancer prevention curricula in U.S. medical schools. J Cancer Educ. 1998;13(2):90–5. doi: 10.1080/08858199809528523. [DOI] [PubMed] [Google Scholar]
- 33.Baysac MA, Horowitz AM, Ma DS. Oral cancer information in health education textbooks. J Cancer Educ. 2004;19(1):12–6. doi: 10.1207/s15430154jce1901_07. [DOI] [PubMed] [Google Scholar]
- 34.Day TA, Chi A, Neville B, Hebert JR. Prevention of head and neck cancer. Curr Oncol Rep. 2005;7(2):145–53. doi: 10.1007/s11912-005-0041-x. [DOI] [PubMed] [Google Scholar]
- 35.Lingen MW, Kalmar JR, Karrison T, Speight PM. Critical evaluation of diagnostic aids for the detection of oral cancer. Oral Oncol. 2008;44(1):10–22. doi: 10.1016/j.oraloncology.2007.06.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Thomson PJ. Field change and oral cancer: new evidence for widespread carcinogenesis? Int J Oral Maxillofac Surg. 2002;31(3):262–6. doi: 10.1054/ijom.2002.0220. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.