Cervical cancer is an important cause of death in women worldwide.1 Cervical carcinogenesis is highly associated with (high-risk) human papilloma virus (HPV) infections.2 Cytomorphological examination of cervical smears is a widely applied, though not ideal screening method for cervical cancer and its precursors, since the Pap smear has false negative rates of 2–40%, due to a combination of sampling error, processing artifacts and the nature of subjective interpretation.1, 3, 4 False-negative cytology can also be found in about 50% of cases when previous negative smears are reviewed from the small proportion of screened women who develop invasive cancer.4 Moreover, as many as 20% of all Pap smears are interpreted as atypical squamous cells of undetermined significance (ASCUS) or borderline dyskaryotic, leading to increased surveillance and more invasive tests in many of these patients.3, 5 The incidence of squamous cell carcinoma of the cervix has decreased since introduction of nation wide screening programs, compared to a relative increased incidence of adenocarcinoma of the cervix.6 The efficacy of cytological screening appears to be diminished for adenocarcinomas, possibly because glandular atypia is more difficult toassess than dyskariosis of squamous cells.7 High-risk HPV (Hr-HPV) testing has been suggested to improve cervical cancer screening8, 9; however, the specificity of Hr-HPV testing, especially in a young screening population is relatively low10 and adding more HPV subtypes to the test will lower the specificity even more.11 Therefore, new objective diagnostic methods, based on molecular changes specific for cervical carcinogenesis are needed.
Silencing of (candidate) tumor suppressor genes by hypermethylation of CpG islands, located in the promoter regions of many genes, is a common feature of human cancers.12 CpG island hypermethylation is often associated with a transcriptional block and loss of the relevant protein and is an early event in carcinogenesis.12 In addition to the functional implications of gene inactivation in tumor development, these aberrant methylation patterns represent excellent targets for novel diagnostic approaches based on methylation sensitive PCR techniques. In previous studies, several genes were identified as being aberrantly methylated in cervical cancer. Most of these studies focused on a single gene and only a few studies investigated several genes.13, 14, 15, 16, 17, 18, 19 Virmani et al.,17 Dong et al.19 and Narayan et al.18 studied methylation patterns of several genes in cervical cancer and found 74, 79 and 87% of cervical cancers to be aberrantly methylated, respectively. It seems that by combining more genes or cervical cancer-specific genes, an increasing number of cervical cancers should be identified through hypermethylation analysis. These previous studies were carried out using conventional methylation specific PCR (MSP) rather than real-time quantitative MSP (QMSP), which permits reliable quantification of methylated DNA.20 QMSP is reported to be more specific and more sensitive than conventional MSP and allows for high throughput analysis, making it more suitable as a screening tool.20 Very few studies investigated the use of (Q)MSP as a diagnostic tool for cervical neoplasia using cervical scrapings.21, 22 Recently, we reported that hypermethylation of genes in cervical scrapings reflect the methylation status of underlying tissue, by comparing methylation ratios of paired fresh frozen tissue samples and cervical scrapings of both cervical cancer patients and healthy controls.21 Feng et al.22 also found a strong concordance between cervical scrapings and biopsy specimens for DAPK, RARβ and TWIST1.
In the present study, we investigated promoter hypermethylation of 12 genes in cervical scrapings, obtained from cervical cancer patients (squamous cell carcinomas as well as adenocarcinomas) and controls. For the present study, 12 genes were chosen for a variety of reasons: CALCA and TIMP3 because these genes were reported to be frequently methylated in cervical cancer, but so far the methylation status of these genes was unknown in normal cervices.23 Hypermethylation patterns of ESR1 and β-Catenin were unknown in cervical tissue at the time of our analysis. However, based on differential mRNA or protein expression levels (e.g. higher expression in controls compared to cervical cancer specimens using either RT-PCR, Western blotting or immunohistochemistry), we hypothesized that this could be due to hypermethylation and therefore these genes were selected.24, 25 Muller et al. recently observed hypermethylation of ESR1 in 48/65 cervical cancers.26APC, CDH1, DAPK, FHIT, HIC1, MLH1, RAR-β2 and RASSF1A were selected based on differential hypermethylation patterns between cervical cancer specimens and normal cervices.13, 14, 17, 18, 19, 27 Previously, it was reported that APC, HIC1 and RASSF1A were more often methylated in adenocarcinomas compared to squamous cell cervical cancer.19, 27 Our aims were to evaluate genes that may distinguish cervical cancer cases from controls as well as those genes that are specifically hypermethylated in adenocarcinoma, since one of the pitfalls of the current screening method is the unreliable identification of cervical adenocarcinomas.
Patients and methods
Cervical scrapings were collected from cervical cancer patients and controls.
All cervical cancer patients referred between March 2001 and September 2003 because of cervical cancer were asked to participate in our study during their initial visit to the outpatient clinic of the University Medical Center Groningen. Gynecological examination under general anesthesia was performed in all cervical cancer patients for staging in accordance with the International Federation of Gynecology and Obstetrics (FIGO) criteria.28 Cervical scrapings were collected during the initial visit to the outpatient department or at gynecologic examination under general anesthesia.
For the present study, we selected 30 cervical cancer patients; 20 with squamous cell carcinoma (67%) and 10 with adenocarcinoma (33%); 1 FIGO stage IA (3%), 10 FIGO stage IB (33%), 4 FIGO stage IIA (13%), 2 FIGO stage IIA/B (7%), 9 FIGO stage IIB (30%), 2 FIGO stage III (2%) and 2 FIGO stage IV (7%). The median age of the cervical cancer patients was 42 years (IQ range 35–58 years). There were no differences between the adenocarcinomas and squamous cell cancers regarding stage and/or age. Control scrapings were obtained from patients (n = 19) without a history of abnormal Pap smears or any form of cancer and planned to undergo a hysterectomy for nonmalignant reasons during the same period. Indications for hysterectomy were fibroids (n = 7), prolaps uteri (n= 6), adenomyosis (n = 1), hypermenorrhea (n = 1) or a combination of fibroids and either prolaps or adenomyosis (n = 4). Cervical scrapings were collected after anesthesia and just before surgery. Median age for controls was 49 years (IQ range 44–57 years). All specimens were judged as benign at pathology.
Informed consent was obtained from all patients and controls participating in this study. The study was approved by the ethics committee of our hospital.
Sample collection and DNA extraction
The cervical scrapings were collected as described previously.21 In brief, after scraping of the cervix, cells were resuspended in 5 ml PBS. One milliliter was used to make cytospins for cytomorphological assessment in ethanol-carbowax (7% polyethylene glycol, 50% ethanol). Cytospins were Pap-stained and routinely classified by 2 independent pathologists without knowledge of the molecular and clinical data according to a modified Papanicolaou system.29 The remaining part (4 ml) was used for DNA isolation using standard salt-chloroform extraction and ethanol precipitation.
HPV detection and typing
For the detection of the presence of Hr-HPV, genomic DNA was first analysed with HPV16- and HPV18-specific primers. In short, the HPV16-specific PCR was performed as described previously.30 The HPV18-specific PCR was performed as described by Baay et al.31 using 2 sets of primers: HPV18-long (217 bp), 5′-AAG GAT GCT GCA CCG GCT GA-3′and 5′-CAC GCA CAC GCT TGG CAG GT-3′; and HPV18-short (115 bp), 5′-CCT TGG AGG ACG TAA ATT TTT GG-3′ and 5′-CAC GCA CAC GCT TGG CAG GT-3′. On all HPV16- or HPV18-negative cases, a general primer-mediated PCR was performed using 2 HPV consensus primer sets, CPI/CPIIG and GP5+/6+, with subsequent nucleotide sequence analysis as we described previously.32 For nucleotide sequence analysis and comparisons, the programs Blast2-WU and Align of the EMBL-EBI sequence analysis software package were used (www.ebi.ac.uk). As a control for specificity and sensitivity of each HPV-PCR, a serial dilution of DNA was extracted from HPV16-positive CaSki and HPV18-positive HeLa cell lines. All standard precautions were taken to avoid contamination of amplification products. For quality control, genomic DNA was amplified in a multiplex PCR containing a control gene primer set resulting in products of 100, 200, 300, 400 and 600 bp according to the BIOMED-2 protocol.33 Only DNA samples with PCR products of 300 bp and larger were used for the detection of HPV.
Real-time quantitative methylation-specific PCR
QMSP was performed after bisulfite treatment on denatured genomic DNA as previously reported.21, 34, 35 Primer pairs, amplicon size and Genbank accession number QMSP primers and probes are listed in Table I. As internal reference gene, β-actin was chosen. For the TaqMan-based QMSP each sample was analyzed in triplicate. As a positive control, serial dilutions of in vitro-methylated genomic DNA with Sss I (CpG) methyltransferase (New England Biolabs, Beverly, MA) was used in each experiment. For quality control, all amplification curves were visualized and scored without knowledge of the clinical data. A DNA sample was considered hypermethylation positive for a certain gene if at least 2 of 3 triplicates showed a Ct-value below 50 and DNA input was at least 225 pg β-actin (equivalent to a Ct-value of 34). Samples with DNA input below 225 pg frequently showed stochastic amplification.
QMSP values were adjusted for DNA input by expressing results as ratios between 2 absolute measurements ((average DNA quantity of methylated gene of interest/average DNA quantity for internal reference gene β-actin) × 10,000).21, 20 Samples were analyzed by plotting hypermethylation ratios for each sample in a scatter plot and choosing a cut-off ratio above the highest control ratio observed for each gene, to set specificity at 100%.21 Differences in prevalence of cancers hypermethylated above the highest control ratio between groups were tested using the chi-square test with Yates' correction for small numbers. Hypermethylation ratios for each gene were compared between cancer cases and controls, and furthermore between adenocarcinomas and squamous cell carcinomas with the Mann-Whitney U test. Once the best individually discriminating genes were found, 2-gene, 3-gene and 4-gene combinations were tested to identify the highest sensitivity with specificity set at 100% for each gene. To compare QMSP as a diagnostic tool with “classic” cytomorphologic assessment, differences in sensitivity and specificity between these assays were analyzed with the chi-square test with Yates' correction for small numbers. All analyses were carried out using the SPSS software package (SPSS 11.5, Chicago, IL). All observed differences were considered to be significant when associated with p < 0.05.
QMSP as a diagnostic tool
DNA quality was sufficient to perform QMSP on 47 cervical scrapings (20 SCC, 8 adenocarcinomas and 19 controls). CDH1, FHIT, HIC1, MLH1 and RASSF1A promoters were hypermethylated in cervical cancers as often as in controls, whereas no hypermethylation was observed for β-Catenin promoter (data not shown). Figure 1 shows that CALCA, ESR1, DAPK, TIMP3, APC and RAR-β2 promoters were more hypermethylated in cervical cancers than in controls. RAR-β2 was the only gene promoter out of these 6 genes that was never found to be hypermethylated in controls. APC and TIMP3 promoters were also more hypermethylated in adenocarcinomas than in squamous cell cervical cancers (for p-values, see legend to Fig. 1).
To determine discriminative ability between normal and cancer cases for each gene, specificity was set at 100% by choosing a cut-off value above the highest control ratio for each gene. Table II shows the number of cervical cancer scrapings that were hypermethylated above the highest control ratio. CALCA, DAPK, ESR1, TIMP3, APC and RAR-β2 promoter hypermethylation discriminated between cervical cancer cases and controls. In an analysis of different histological subtypes, in adenocarcinomas hypermethylation above the highest control ratio for the APC, TIMP3 and RASSF1A promoters was significantly more often compared to squamous cell carcinomas (Table II).
Table II. Hypermethylation above the Highest Control Ratio in all Cervical Cancer Scrapings (CC) and Squamous Cell Carcinomas (SCC) Versus Adenocarcinomas (AC)
Number of cancers hypermethylated above the highest control ratio.
χ2, comparing number of hypermethylated cases.
Setting specificity at 100%, CALCA, DAPK, ESR1, TIMP3, RAR-β2 and APC had a sensitivity for the detection of cancer of 68, 46, 32, 21, 18 and 18%, respectively (Table II). Table III shows the sensitivity of different combinations of these genes. Combinations of 2, 3 or 4 genes were tested while maintaining perfect specificity for each gene. Our best result was obtained by combining 4 genes, CALCA, DAPK, ESR1 and APC, reaching a sensitivity of 89.3%. Addition of more genes did not increase sensitivity. The best 3-gene combination was CALCA combined with ESR1 and DAPK (sensitivity of 85.7%). The sensitivity of the 2-gene combination CALCA together with either ESR1 or DAPK was 78.6, while ESR1 together with DAPK reached a sensitivity of 67.8%.
Table III. Sensitivity and 95% Confidence Interval (CI) for Combination of Hypermethylated Genes, with Specificity set at 100% for each Gene Promoter
CALCA, DAPK, ESR1, APC
CALCA, DAPK, ESR1
Cytomorphology and HPV analysis
Cytomorphological analysis could be performed on 18 of 19 control samples (1 inadequate sample). Cervical tissue of all controls was histologically diagnosed as normal cervical epithelium. Nevertheless, 2 scrapings were classified as borderline dyskaryotic and 1 as severe dyskaryotic (Table IV). Specificity for cytomorphology was therefore 83.3% (15/18) in our study.
Table IV. Morphological Assessment Compared with Hypermethylation of at least one Gene in Cervical Scrapings of Histologically Confirmed Normal Controls and Cervical Cancer Patients
Sufficient DNA for QMSP
Hypermethylation is defined as methylation above highest control ratio.
High-risk HPV include HPV16, HPV18, HPV31 and HPV45.
Controls are by definition negative for hypermethylation.
Borderline dysplasia is Pap II and Pap IIIA mild dyskaryosis.
Severe dysplasia/CIS is Pap IIIA moderate dyskaryosis, Pap IIIB and Pap IV.
All 30 cervical cancer scrapings were morphologically analyzed. Three scrapings were classified as no or borderline dyskaryosis, of which 2 were obtained from adenocarcinoma cases. Furthermore, 3 scrapings were inadequate for morphological diagnosis because they contained an insufficient amount of cells (Table IV). Since 1 cancer case was scored as normal and 2 as borderline dyskaryosis (according to Dutch screening guidelines leading to a delay in diagnosis of at least 6 months) sensitivity for cytomorphology was calculated to be 88.9%(24/27).
HPV analysis was performed on all cervical scrapings (19 controls and 30 cervical cancers) (Table IV). Hr-HPV was detected in 6 of 19 controls and in 27 of 30 cervical cancers, resulting in a specificity and sensitivity of 68.4% (13/19) and 90% (27/30), respectively.
Cytomorphology versus QMSP and HPV
Two cervical cancer scrapings (both adenocarcinomas) had insufficient DNA to perform QMSP. Of the 28 remaining cervical cancer samples, 25 were hypermethylated above the highest control ratio for at least 1 gene (sensitivity 89.3% (25/28)). Of the 3 cervical cancer scrapings that were morphologically classified as borderline or no dyskaryosis 2 had sufficient DNA input for QMSP analysis and 1 was hypermethylated above the highest control ratio for at least 1 gene. Of the 3 scrapings that were insufficient for cytomorphological assessment, 2 had sufficient DNA input and were also hypermethylated above the highest control ratio for at least 1 gene (Table IV). All of these 6 cervical cancer scrapings were positive for Hr-HPV.
In summary, our data reveal that the combination of 4 genes (CALCA, DAPK, ESR1 and APC) had a theoretical sensitivity of 89.3%, equal to cytomorphology (88.9%) and Hr-HPV (90.0%) with a specificity of QMSP (each gene cutoff set at 100%) compared to cytomorphology (83%) and Hr-HPV (68%) (both p < 0.05).
Current screening for cervical cancer by Pap smear analysis is associated with significant false positive and false negative rates3 and especially adenocarcinomas are easily missed.7, 6 Previously, we and others have shown that detection of hypermethylated gene promoters in cervical scrapings using (Q)MSP is a promising tool for identification of squamous cell cervical cancer patients.21, 22 However, the published sensitivity of (Q)MSP for detection of cervical cancer (67–74%) needs to be improved and more data on the detection of adenocarcinoma are needed.21, 22 In the present study, we evaluated the hypermethylation status of the promoters of 12 genes. These 12 genes were selected, because they were either previously shown to be frequently hypermethylated in cervical cancer specimens or to have frequently downregulated mRNA or protein expression in cervical cancer. We hypothesized that sensitivity could be increased by analyzing genes frequently hypermethylated in cervical cancer and we demonstrate that this is indeed the case. We identified 89% of cervical cancer scrapings as cases, by combining ESR1, DAPK, APC and CALCA, with high theoretical specificity. Furthermore, we identified 3 genes, APC, TIMP3 and RASSF1A that could distinguish adenocarcinomas from squamous cell carcinomas, despite the fact that we only analyzed a relatively small series of cervical cancers. In fact, QMSP identified all 8 adenocarcinomas as cancers with high sensitivity and specificity, which well exceeded the performance of cytomorphology in identifying adenocarcinomas. Overall, our results suggest that QMSP of gene promoters, frequently hypermethylated in cervical cancer (squamous cell and adenocarcinoma) show promise in augmenting current cervical cancer screening based on cytomorphology. However, our study has some important limitations; the number of patients is relatively small, only cancer cases were analyzed and patients already known to have cervical cancer were included in the study, which is of course quite different from a real screening population where cases and controls are unknown at presentation. Moreover, despite the fact that we evaluated a panel of genes in a very selected group of patients, no 100% sensitivity could be reached, pointing to the fact that the ideal combination of genes still needs to be defined. After we performed our present study, several studies showed other genes to be potential candidate markers for QMSP in cervical cancer screening, such as Tumor Suppressor Lung Cancer 1 (TSLC1), the anti-apoptotic decoy receptor 1 (DcR1) and the cell differentiation gene TWIST140, 41, 22. In future studies, these genes need to be further evaluated and when a sensitivity of >95% for cervical cancers will be reached, it will be interesting to expand these studies also to patients with higher grades CIN.
Ideally hypermethylation of 1 marker should be able to identify most (perhaps >90%) cancers as cases and none of controls, as appears to be the case for GSTP1 hypermethylation in adenocarcinoma of the prostate.42 Many studies have been conducted trying to find such markers for cervical cancer using either MSP or QMSP.40, 41, 43, 44 Of the 12 genes we analyzed, RAR-β2 was the only gene not hypermethylated in controls (Fig. 1), suggesting that it may be difficult to find a marker that is highly sensitive and still rarely hypermethylated in controls. As a consequence, one may imagine that only a quantitative assay will be able to really distinguish cancer cases from controls by setting cut-off values. An important quality of a candidate gene should then be that there is a large difference in median hypermethylation ratios between cancer cases and controls. Markers such as ESR1 might also make good candidates. Very recently Müller et al. observed hypermethylation of ESR1 in 48/65 (73%) cervical cancers.26 In our study 64% of cervical cancers were hypermethylated for ESR1 versus only 1 out of 19 controls (Fig. 1). If larger studies show that ESR1 indeed is only rarely hypermethylated in controls it might be possible to lower the positive threshold for this gene, thus increasing sensitivity.
The results of our analyses of hypermethylation of APC, DAPK, MLH1, RAR-β2 and RASSF1A were similar to those observed in other studies in cervical cancer and normal cervices13, 16, 17, 18, 22, 27, 45 while hypermethylation patterns observed in our study for CDH1, FHIT and HIC1 were not comparable to other studies.17, 18, 19 Other studies found aberrant methylation in 16–58% of cervical cancers versus none of controls, while we observed 75–100% of hypermethylation in both cancer and controls. These differences can be explained by the fact that in these other studies conventional MSP was performed, which is reported to be less sensitive than QMSP.20 We showed that APC, TIMP3 and RASSF1A were more often hypermethylated in adenocarinomas compared to squamous cell cervical cancers, which is in agreement with other reports.19, 27, 45
Currently, implementing HPV DNA testing in cervical cancer screening programs is being considered.8 Infection with Hr-HPV is mandatory for cervical cancer to develop and therefore screening for HPV DNA seems logical. However, screening for Hr-HPV will result in informing many women with normal Pap smears that they are at risk of developing cervical cancer, causing unnecessary anxiety.8, 11 Our study also illustrates that even in our very selected, relatively small study population Hr-HPV testing indeed was less specific than testing for hypermethylation. Subclinical HPV infection is relatively common and often transient, especially among younger, sexually active women, making cervical cancer a rare complication of an HPV infection.10 The percentage of women who are HPV positive but have a cytologically and/or histologically confirmed normal cervix varies from 5% in Europe to 26% in Sub-Saharan Africa.46 One potential advantage of QMSP over HPV detection for cervical cancer identification is that QMSP will hopefully identify clonal lesions that are already present, while HPV detection alone identifies many patients at higher risk, most of whom will never develop cervical neoplasia. The data on the potential of HPV screening are much larger and more mature than that presently available for QMSP. In light of the length of the “HPV experience” with respect to the implementation of HPV detection in cervical screening lessons should be learnt and comparable mistakes should be avoided when designing future studies on the application of QMSP in cervical cancer screening.
In conclusion, our pilot-study on cervical scrapings indicates that a QMSP combination of 4 genes, frequently hypermethylated in cervical cancer appears to have similar sensitivity, but better specificity in comparison to “classic” cytomorphological assessment and Hr-HPV detection. Future studies will uncover whether QMSP on cervical scrapings will be able to augment or improve current screening approaches.
Prof. A.G.J. van der Zee is a paid consultants for OncoMethylome Sciences S.A., Liège, Belgium. Under a licensing agreement between Oncomethylome Sciences, SA and the Johns Hopkins University, Dr. D. Sidransky is entitled to a share of royalty received by the University upon sales of diagnostic products described in this article. Dr. D. Sidransky owns Oncomethylome Sciences, SA stock, which is subject to certain restrictions under the University policy. Dr. D. Sidransky is a paid consultant to Oncomethylome Sciences, SA and is a paid member of the company's Scientific Advisory Board. The Johns Hopkins University in accordance with its conflict of interest policies is managing the terms of this agreement.