DNA methylation may be used a potential biomarker for detecting cervical cancer. The authors of this report used quantitative methylation analysis of 4 genes in a full spectrum of cervical lesions to test its potential clinical application.
This hospital-based, retrospective, case-control study was conducted in 185 patients and included patients who had a normal uterine cervix (n = 53), cervical intraepithelial neoplasm type 1 (CIN1) (n = 37), CIN2 (n = 22), CIN3 (n = 24), carcinoma in situ (CIS) (n = 22), squamous cell carcinoma (SCC, n = 20), and adenocarcinoma (AC) (n = 7). Methylation levels of the genes sex-determining region Y, box 1 (SOX1); paired box gene 1 (PAX1); LIM homeobox transcription factor 1α (LMX1A), and NK6 transcription factor-related locus 1 (NKX6-1) were determined by using real-time methylation-specific polymerase chain reaction (PCR) amplification. Cutoff values of the percentage of methylation reference (PMR) for different diagnoses were determined to test the sensitivity and specificity and to generate receiver operating characteristic (ROC) curves. Two-sided Mann-Whitney U tests were used to test differences in PMR between groups.
The PMRs of the 4 genes were significantly higher in CIN3 and worse (CIN3+) lesions than the PMRs in specimens of normal cervix and CIN1 or CIN2 (P < .001). ROC curve analysis demonstrated that the sensitivity, specificity, and accuracy for detecting CIN3+ lesions were 0.88, 0.82, and 0.95, respectively, for SOX1; 0.78, 0.91, and 0.89, respectively, for PAX1; 0.77, 0.88, and 0.90, respectively, for LMX1A; and 0.93, 0.97, and 0.97, respectively, for NKX6-1.
Since the introduction of the Papanicolaou test (Pap smear) decades ago, the mortality and morbidity for patients with invasive cervical cancer have reduced greatly, especially in developed countries.1-6 The achievement of Pap smear testing for public health is obvious. However, there are still 500,000 cases of cervical cancer and 275,000 cervical cancer deaths among women worldwide, and the incidence rates in developing countries are particularly high.7 Sensitivity of the Pap smear varies substantially in areas with different screening infrastructures.8 In several meta-analyses of the accuracy of Pap smears, the sensitivity ranged from approximately 50% to 80% but could be as low as 20%,9-11 which limits the efficacy of cancer detection.12 Because infection with the oncogenic human papillomavirus (HPV) causes cervical cancer,13-15 HPV DNA testing is appealing as a method of molecular screening.11, 16, 17 Despite the greater sensitivity of DNA testing compared with the Pap smear, the common nature of this virus makes the specificity low, leading to unnecessary referrals for colposcopy18, 19 and needless worry.20 The transient nature of the age-dependent prevalence of HPV infection also diminishes the value of HPV testing in cervical cancer screening.14, 16, 21, 22 Efforts have been made to circumvent the drawbacks of HPV testing using a higher viral load threshold and the detection of E6/E7 messenger RNA rather than DNA, but those efforts still produced unsatisfactory results.23 Thus, there is a need to develop other biomarkers for cervical cancer screening.
Epigenetic studies during the past decades have demonstrated that epigenetic silencing, such as DNA methylation of tumor suppressor genes (TSGs), can serve as a mechanism of carcinogenesis.24, 25 The addition of a methyl group to the 5′ position of cytosine in a cytosine-guanine (CpG) dinucleotide by DNA methyltransferase and the accumulation of this methyl cytosine in dense CpG regions, termed CpG islands, result in chromatin remodeling and subsequent gene silencing. This can lead to the functional loss of important TSGs.26 Because epigenetic silencing of TSGs by promoter hypermethylation is observed commonly in human cancers, it is possible that DNA methylation could serve as a marker for the early diagnosis of cancers and as a means of assessing the prognosis for patients with cancer.27
The feasibility of using methylation markers in cervical malignancy screening has been appraised critically.28-30 Current methylation markers are insufficient for clinical application, because they have only moderate sensitivity and specificity, and it is hard to convince health providers to use these markers in a clinical setting. However, DNA methylation would have potential as a biomarker for detecting cervical cancer if genes with high sensitivity or specificity could be discovered. Nontargeted epigenomic approaches might help discover potential candidate methylation biomarkers for validation in different populations and with different analytic techniques. Recently, we reported potential methylation biomarkers using differential methylation hybridization coupled with a CpG island microarray.31 To further test the feasibility of these new biomarkers in the detection of cervical neoplasms, we converted the methylation analysis to a quantitative methylation polymerase chain reaction (QMSP) approach and tested its diagnostic performance in an independent clinical cohort.32 The results reported here demonstrate the promise of quantitative methylation analysis for the following genes in the detection of cervical intraepithelial neoplasm (CIN) type 3 (CIN3) and worse (CIN+): sex-determining region Y, box 1 (SOX1); paired box gene 1 (PAX1); LIM homeobox transcription factor 1 α (LMX1A); and NK6 transcription factor-related locus 1 (NKX6-1).
MATERIALS AND METHODS
A hospital-based, retrospective, case-control study was conducted on 185 patients, including women who had a normal uterine cervix (n = 53), CIN1 (n = 37), CIN2 (n = 22), CIN3 (n = 24), carcinoma in situ (CIS) (n = 22), squamous cell carcinoma (SCC) (n = 20), or adenocarcinoma (AC) (n = 7) of the uterine cervix diagnosed according to histologic reports. Cervical scrapings were used for the analysis. A cervical brush (Pap Brush; Young Ou Company, Ltd., Yongin City, Korea) was used to collect cervical scrapings. The brush was preserved in phosphate-buffered saline solution at 4°C until DNA extraction. All patients were diagnosed, treated, and had their tissues banked at the National Defense Medical Center, Taipei, Taiwan, as described previously.33 Patients who had low-grade and high-grade lesions identified by cytology underwent colposcopic cervical biopsy with subsequent conization or major surgery when the biopsy results revealed CIN2+. The final diagnosis was made by tissue-proven pathology rather than cytology except for controls. Controls were recruited from healthy women who underwent routine Pap screening. Informed consent was obtained from all patients and controls. Exclusion criteria included pregnancy, chronic or acute systemic viral infections, a history of cervical neoplasia, skin or genital warts, an immunocompromised state, the presence of other cancers. or a history of surgery to the uterine cervix. The Institutional Review Board of the Tri-Service General Hospital approved this study.
Clinical Specimens and Preparation of Genomic DNA
Genomic DNA was extracted from specimens using an established protocol for tissue banking.33 The concentration of DNA was determined using the PicoGreen fluorescence absorption method, and DNA quality was verified using agarose gel electrophoresis. Samples with a DNA yield >500 ng were considered for further testing. The quality of DNA was not limited in the current project.
Bisulfite Conversion and Quantitative Methylation Polymerase Chain Reaction
A DNA modification kit (Chemicon International Inc., Temecula, Calif) was used according to the manufacturer's recommendations. TaqMan-based QMSP (MethyLight) was performed after bisulfite treatment on denatured genomic DNA.32 The primers and probes for each gene were named me1SOX1, me1PAX1, me1LMX1A, and me1NKX6-1 (sequences will be provided on request). The collagen type II α1 gene (COL2A) was used as an internal reference gene by amplifying non-CpG sequences. Each sample was analyzed in duplicate. In vitro methylated genomic DNA with M.SssI methyltransferase (New England Biolabs, Beverly, Mass) was used as a positive control, because it is considered to result in 100% methylation of each gene. QMSP was done in a total volume of 20 μL that contained 2 μL modified template DNA, 1 μL 20 × custom TaqMan reagent, and 10 μL Universal polymerase chain reaction (PCR) Master Mix (No AmpErase UNG, no. 4324018; Applied Biosystems, Foster City, Calif). The reactions were subjected to an initial incubation at 95°C for 10 minutes, followed by 50 cycles at 95°C for 15 seconds, and annealing and extension at the appropriate temperature for 1 minute at 60°C, then detected using the 7500 Real-Time PCR System (Applied Biosystems). The DNA methylation level was assessed as the percentage of methylation reference (PMR) using the formula: (sample_gene/sample_COL2A) × 100/(SssI_gene/SssI_COL2A).32, 34 Testing results with cycle threshold values of COL2A >40% were defined as detection failures.
Two-sided Mann-Whitney U tests were used to evaluate differences in PMR between groups. Age differences between controls and cases were compared using independent sample Student t tests. The trend of age variation from CIN1 to invasive cancer was tested using ordinal logistic regression analysis. Receiver operating characteristic (ROC) curves were generated to confirm the accuracy of diagnosis of each gene, and sensitivity and specificity were computed for each combination. Twenty-six samples per group were required to achieve adequate power (0.90) with an α level <.05 to test for a 0.35 difference in the area under the ROC curve from 0.5 (a straight line from bottom left to top right corners, implying a decision rule no better than pure chance). All differences were considered statistically significant at P < .05. All analyses were carried out using the SPSS software package (version 13.0; SPSS Inc., Chicago, Ill).
The characteristics of enrolled patients and the numbers with successful methylation detection are shown in Table 1. The mean patient age increased in accordance with disease severity (P < .001), as expected. Quantitative assessment of DNA methylation for each gene in accordance with disease severity is shown in Table 2 and Figure 1. The median PMR was low in samples with normal cytology and in CIN1 samples for all genes (<1%). The median PMR was high in most CIN3+ samples (including CIN3, CIS, SCC, and AC) and malignant smear samples. CIN2 samples demonstrated low but heterogeneous PMR values for the SOX1, PAX1, and LMX1A genes. When dichotomized as CIN3+ and CIN3− (including CIN2, CIN1, and normal cervix), the difference in PMR was statistically significant for each gene (all P < .001).
To assess the clinical application, ROC curves were generated, and the areas under the ROC curve were calculated to discriminate between the CIN3− group and the CIN3+ group (Fig. 2). Accuracy ranged from 0.89 to 0.97. Sensitivity and specificity were calculated at the best cutoff PMR values for each gene (Fig. 1). The sensitivities and specificities ranged from 0.76 to 0.93 and from 0.81 to 0.97 for the CIN3− group and the CIN3+ group, respectively. NKX6-1 conferred the best performance with sensitivity, specificity, and accuracy of 0.93, 0.97 and 0.97, respectively, for the detection of CIN3+ lesions. The performance of combined testing also was calculated (Table 3). Combined testing of either SOX1 or NKX6-1 had greater sensitivity (0.96) and compromised specificity (0.81). Combined testing of both SOX1 and NKX6-1 also had greater specificity (0.99) and compromised sensitivity (0.85).
Table 3. Combined Testing of DNA Methylation in the Detection of Cervical Intraepithelial Neoplasia Type 3 or Worse
Any 1 Gene Methylated
Both Genes Methylated
SOX1 indicates sex-determining region Y, box 1; PAX1, paired box gene 1; LMX1A, LIM homeobox transcription factor 1α; NKX6-1, NK6 transcription factor-related locus 1.
By using the cutoff values for the detection of CIN3+, the methylation rates in cervical scrapings from CIN2 were 28.6%, 47.1%, and 40.9% for SOX1, PAX1, and LMX1A, respectively (Fig. 3). Six of 22 CIN2 tumors (27.3%) were negative for methylation of all 3 of these genes. Only 1 patient (4.5%) had positive methylation for all 3 of these genes.
DNA methylation has been proposed as a potential biomarker for the detection of cervical cancer.29 Initial attempts at using methylation markers for detecting cervical cancer were hampered by the lack of consistent QMSP results and moderate sensitivity and specificity using the genes available.29 In the latest review, which included 68 genes from 51 studies of cervical cancer,29 there was wide variation in the methylation frequencies of most genes. Only limited numbers of genes (15 of 68 genes) were tested in >5 studies. Most studies (32 of 51 studies) used methylation-specific PCR (MSP), and only 7 of 51 studies used QMSP. Because of the inconsistent methylation analysis results from different studies, the authors concluded that, currently, there is no single methylation biomarker that has the appropriate performance to serve as a cervical cancer biomarker. Indeed, testing methods and tissue sources do affect testing results. For clinical applications, it is more logical to test DNA methylation using cervical scrapings rather than tissues. Genes with high methylation rates in cancer tissues may not necessarily have high methylation rates in cervical scrapings.31, 35 The reasons for this discrepancy are not clear but may include sampling issues, detection methods, or biologic mechanisms.36 MSP identifies CpG sites on primers, which usually cover only 2 or 3 CpG sites each. The call of methylation or not is made based on semiquantitative gel electrophoresis, which is not sensitive. For QMSP, a pair of primers and a probe are used. Different designs of primer and probe sequences may cause different testing results. Real-time PCR provides quantitative results. The cutoff values for methylation may depend on clinical intentions. The probes may cover CpG sites, which further increase the specificity of the testing results. In the review by Wentzensen et al, only 12 studies used exfoliated cells, and only 7 of those studies used QMSP. In the current study, we used QMSP with primers and probes different from those used for MSP and produced a better performance in detecting CIN3+, which, to our knowledge, is the best performance of DNA methylation biomarkers in cervical cancer detection reported to date. A reliable and validated assay is needed to expand these results.
Efforts to discover genes that undergo methylation silencing in cervical cancers have been made using various genome-wide strategies and have suggested the potential for some newly discovered genes.31, 37, 38 In the current study, we validated the application of our previously discovered methylation levels of SOX1, PAX1, LMX1A, and NKX6-1 by using QMSP in an independent clinical cohort and demonstrated high accuracy for detecting CIN3+ lesions.31 In a recent study of 28 genes, including the same 4 genes that we analyzed in the current study, MethyLight and liquid-based cytology specimens were used to independently validate the accuracy of SOX1 in the detection of high-grade squamous intraepithelial lesions in a United Kingdom population despite a limited number of patients, no tissue-proven pathology, and differences in primer and probe designs.39 The differences between studies in methylation results from analyses using QMSP may lie in different designs of primers and probes, different analytic platforms, different cutoff values, and different ethnic or environmental backgrounds. A panel of genes with moderate sensitivity and specificity is needed to achieve high accuracy for detecting cervical lesions.29, 35 The current study demonstrated that the use of a limited number of genes or possibly a single gene rather than a large panel could be satisfactory to achieve diagnostic efficacy. The reason why NKX6-1 was better than PAX1 in the current study, contrary to our previous results, lies in the designs of primers and probes for QMSP. Selected testing of different genes might yield different sensitivities and specificities, and such genes should be evaluated for different clinical settings. For patients who have greater medical demands or who are in developed countries that have a low incidence of cervical cancer, a screening test might need greater sensitivity, albeit a minor compromise with specificity, such as the combination of SOX1 and NKX6-1. This means that women with CIN3+ are less likely to be missed on screening and, conversely, that more women without disease also may have positive methylation testing results and will need more medical resources to clarify their diagnoses. Women in developed countries or those who are reluctant to take the risk of false-negative results and are willing to pay for more medical evaluations may be suitable for this combination. However, in a developing country, a more specific screening method may be chosen to avoid the resource-demanding management of false-positive results, suggesting the alternative testing of SOX1 with NKX6-1, a combination that yielded a specificity of 0.99 and a sensitivity of 0.85 for CIN3+ lesions, or even the combination of PAX1 with NKX6-1, which yielded 100% specificity and 78% sensitivity.
The current study also demonstrated the heterogeneity of CIN2 regarding methylation profiles. According to 2006 consensus guidelines for the management of women with abnormal cervical cancer screening tests, CIN2 is categorized as a high-grade lesion, and patients who have CIN2 lesions should be referred for colposcopy and tissue biopsy, including adolescent and pregnant women.40 Cold-knife conization and loop electrosurgical-excisional conization are suggested for high-grade lesions after unsatisfactory colposcopy; these lesions reportedly are associated with cervical stenosis and cervical incompetence and the resultant preterm premature rupture of membranes and preterm delivery.41-44 Conservative cytologic and colposcopic follow-up is an alternative for adolescents and young women with CIN2.45 Indeed, the diagnosis of CIN2 has been a gray area and is the most difficult for pathologists to reproduce among all cervical smear diagnoses.46, 47 For a long time, CIN2 was considered an intermediate entity that may be over called as CIN1 or under called as CIN3. Some pathologists even use “CIN1-2 or CIN2-3” to equivocate the classification. In fact, CIN2 lesions differ from CIN3 lesions according to their natural history.48 Approximately 40% of CIN2 lesions regress, and 5% progress to invasive cancer. The corresponding approximations for CIN3 are 33% and 12%, respectively.49 A follow-up study also demonstrated that 40% of undiagnosed CIN2 will regress over 2 years.50 The detection of HPV infection cannot solve this problem satisfactorily.51 Unnecessary referral and repeat diagnostic procedures not only waste medical resources but also generate needless worries. Therefore, a better prognostic risk evaluation for CIN2 is needed. The current study demonstrated a heterogeneous methylation profile among CIN2 lesions, including only 4.5% that were positive for 3 methylation markers, which implies that DNA methylation may serve as an auxiliary evaluation when there are equivocal results. A larger longitudinal study may be warranted to elucidate the natural history of CIN2 in relation to DNA methylation.
SOX1, PAX1, LMX1A, and NKX6-1 are transcription factors with known roles in the developmental processes. Little is known about the roles of these transcription factors in cancers. Recently, for the first time to our knowledge, we characterized the role of LMX1A as a true metastasis suppressor in cervical cancers through a bone morphogenetic protein-mediated, incomplete epithelial-mesenchymal transition.36 Methylation of LMX1A also was observed in a colon cancer cell line (HCT-116) and was demethylated in the DKO cell line, which is genetically disrupted for DNA methyltransferase 1 (DNMT1) and DNMT3b.52 A recent report using a CpG island microarray indicated that NKX6-1 is methylated in some subtypes of lymphoma.53 We anticipate that further functional characterization of these developmental genes in cancer will increase our understanding of development and cancer biology and lead to new diagnostics or therapeutics.
Despite the promising results, the current study does have limitations. The cutoff value for each gene was based on a hospital-based, retrospective case-control study from a research platform, which may not be applied directly to clinical settings from wider populations. A multicenter study is ongoing to validate these results using standardized reagents and platforms with diagnostic certifications. Population-based studies using a standardized DNA methylation detection method are warranted to evaluate its impact and cost effectiveness in clinical practice. In addition, all patients who we included in the current study were Asian. The extent to which the current results can be applied to other ethnic populations remains to be determined, although the results from a United Kingdom-based case-control study have supported the potential of this approach.39
In summary, adding to our previous work on the discovery phase of these developmental genes that undergo methylation silencing in cervical cancer, the current study validated the clinical potential of a quantitative methylation analysis in the detection of CIN3+ lesions. The application of these new biomarkers in the triage of mildly abnormal smears, as an adjunct to HPV testing for primary screening or as a new generation of cervical cancer screening, warrants further investigation.
CONFLICT OF INTEREST DISCLOSURES
Supported by grants TSGH-C98-07-S01, S02, and S03 from the Tri-Service General Hospital; grant NSC98-2314-B-016-030-MY3 from the National Science Council, Republic of China; grant DOH98-TD-I-111-TM003 from the Department of Health, Taiwan, ROC; the C. Y. Chai Foundation for Advancement of Education, Sciences, and Medicine; and the Teh-Tzer Study Group for Human Medical Research Foundation.