Cervical dysplasia: Assessing methylation status (Methylight) of CCNA1, DAPK1, HS3ST2, PAX1 and TFPI2 to improve diagnostic accuracy

Results A total of 165 subjects provided cytologically proven 63 HSIL, 49 LSIL and 53 normal samples. All patients with HSIL and LSIL underwent colposcopic examination. Patients with LSIL were all found to be CIN1; patients with HSIL were subsequently subdivided into 10 squamous cell carcinoma (SCC), 31 CIN3, 10 CIN2 and 12 CIN1. For each gene, there was increasing frequency of methylation from normal and LSIL (CIN1), through HSIL (CIN2 and CIN3), to SCC. Methylation of ≥ 1 of genes investigated was observed in 88% of combined HSIL (CIN2 and CIN3) and SCC cases. All genes showed significant increase in methylation level (PMR value) with increasing disease grade ( p < 0.005). CCNA1 was the only gene that was able to distinguish CIN2 from CIN3 specimens ( p = 0.016). Based on receiver operating characteristic (ROC) analysis, HS3ST2 was the most significant candidate in segregating HSIL/SCC from normal/LSIL cases ( p < 0.0001); at an optimal cutoff value, sensitivity and specificity between 70% and 80% were obtained. Conclusions Development of DNA methylation status of a gene panel to improve diagnostic accuracy in cervical neoplasia is warranted. Keywords Cervical cancer Cervical dysplasia CIN HPV Methylation Methylight Introduction Cervical cancer is the 2nd commonest cancer in women worldwide, and is ranked 3rd in cause of cancer death [1] . These ominous statistics are largely due to the developing countries, which stand in contrast with those of developed countries, where the declining trends in incidence and mortality rates are largely attributable to cervical smear screening and early intervention. Cervical cancer has a long pre-invasive phase, spanning years. The different grades of dysplasia are defined using descriptive terminology based on the Bethesda system [2] . For example, low-grade squamous intraepithelial lesion (LSIL) includes mild dysplasia and cervical intraepithelial neoplasia grade 1 (CIN1), whereas high-grade squamous intraepithelial lesion (HSIL) comprises moderate to severe dysplasia, CIN2 and CIN3. In practice, a diagnosis of CIN 3 or CIN 1 is more reproducible than CIN 2, so CIN 2 may include overcalled CIN 1 and undercalled CIN 3. The medical community continues to grapple with the issue of reproducibility of cervical cytological and histological interpretations, as evidenced by the variability found in the ASCUS-LSIL Triage Study [3] and the Gardasil FUTURE trials [4] . The limited reliability of colposcopy and colposcopic-directed biopsies also has been problematic [5–7] , although studies have suggested that this is partly ameliorated by taking additional biopsies [8,9] . The question arises whether more precise determination of mild, moderate or severe pre-invasive disease can be achieved from molecular data, as there are treatment and follow-up implications. Currently, most patients with CIN 2 and CIN 3 have surgical intervention [10,11] . If left untreated, it is estimated that for CIN 2, 40–60% of lesions regress, 20% progress to CIN 3 and 5% progress to invasive cancer, whereas for CIN 3, 30–50% regress, 10–40% progress to invasive cancer [12–17] . For CIN 1, regression to normal is seen with most women and about 10% progress to CIN2 or CIN 3 [7,12,18,19] . For the majority of CIN 1 patients, a watchful waiting approach is adopted. This dichotomy in clinical follow-up based on the diagnostic accuracy of microscopic visual inspection raises issues of over- and under-treatment, and whether there are other independent diagnostic methods that can complement the traditional approach. Cervical neoplasia has an etiologic association with human papilloma virus (HPV) [20,21] . However, being an external stimulus, the presence or absence of HPV does not directly reflect the cervical epithelial cell's molecular status, which may better predict for the risk of disease progression. Less than 10% of HPV-related cervical dysplasia progress to higher grade lesions or invasive cancer, inferring that HPV is not the sole determinant in disease progression [22–27] . Since the discovery of aberrant DNA methylation patterns in cancer cells in the 1980s, where hypomethylation was observed compared with their normal tissue counterparts [28] , several studies have emerged, reporting atypical promoter hypermethylation of tumor suppressor genes and other cancer-associated genes in a variety of human cancers, including cervical cancer [29–36] . Genomic microarray studies found extensive hypomethylation in gene-poor regions but higher methylation levels in gene-rich areas [37] . Hypotheses put forward to explain hypomethylation in carcinogenesis include promotion of chromosomal rearrangements, disruption of genomic imprinting and reactivation of transposable elements [38–40] . Hypermethylation of CpG island promoters can inactivate tumor suppressor genes, affecting genes of the cell cycle, DNA repair, cell-to-cell interactions, apoptosis and angiogenesis, all of which are involved in cancer development [31,41] . It has been hypothesized that hypomethylation is an inconsequential side effect of carcinogenesis, but there is accumulating evidence regarding the simultaneous spreading of demethylation and de novo methylation in cancer cells, suggesting close interplay between both dynamic processes [42,43] . The development of genome-wide screening approaches such as CpG island microarray and restriction landmark genomic scanning (RLGS) has increased the chances of identifying novel biomarkers that can improve diagnostic accuracy and prognostication in cervical neoplasia [44–46] . In societies where cervical screening programmes are ongoing, high-throughput platforms of analysis offer practical advantage in dealing with the clinical load. Here, we utilize a quantitative, high-throughput methylation assay, MethyLight [47–49] to assess the ability of a gene panel to separate normal, LSIL, HSIL and squamous cell carcinoma (SCC) from one another, by using the logistic regression model that incorporates the best gene(s) combination [50,51] . Materials and methods Clinical specimens Approvals for this study were obtained from the Institutional Review Boards of the National University Hospital and Kandang Kerbau Women's and Children's Hospital, Singapore. These hospitals are referral centres for abnormal cervical screening results. Samples were obtained from patients with informed consent. Residual cells from cervical smear samples collected in SurePath liquid-based cytology containers were analyzed. A total of 165 samples were selected for analysis. These were consecutive specimens collected from 2007 to 2008, specifically enriched for HSIL, LSIL and normal. Cytological examination determined 53 were normal, 49 were LSIL and 63 were HSIL. HPV status was not routinely checked. Subsequent histological examination showed that 61 were CIN1, 10 were CIN2, 31 were CIN3 and 10 were SCC. Colposcopic examination included conventional visual assessment, application of 5% acetic acid, identification of the squamocolumnar junction and transformation zone, identification of suspected CIN for biopsy and overall colposcopic impression. Biopsies were taken from the worst of abnormal-looking areas, and additional biopsies from suspicious areas, according to the clinician's judgement. DNA preparation and bisulfite treatment Cells from cervical smear samples were centrifuged and washed twice with phosphate-buffered saline (PBS) to remove residual preservatives before extraction of genomic DNA using the DNeasy kit (Qiagen, Valencia, CA), according to manufacturer's protocol. DNA quantification was done using the NanoDrop spectrophotometer (Nanodrop Technologies, Wilmington, DE). Approximately 1 μg of genomic DNA was bisulfite-treated with the EZ DNA Methylation-GOLD kit (ZYMO Research, Orange, California), according to manufacturer's protocol. The bisulfite-modified DNA was stored at −20 °C and used within a week for analysis. Real-time quantitative methylation-specific PCR (QMSP) TaqMan-based QMSP analysis was performed using the ABI 7900HT Fast Real-Time PCR System (Applied Biosystems, CA). Supplementary Table 1 summarizes the QMSP primers and probes used. In brief, the PCR assay was performed in a final reaction volume of 20 μl, containing the forward and reverse primers (600 nM each), probe (250 nM) and 4 μl bisulfite-converted genomic DNA (corresponds to 60 ng DNA initially used for conversion) in 1X TaqMan FAST Universal PCR master mix. PCR was performed under the following conditions: 95 °C for 20 s followed by 45 cycles of 95 °C for 1 second and 60–62 °C (depending on the primer set) for 20 s. The primer and probe sequences used were published previously except for CCNA1, PAX1 and TFPI2 where minor modifications were made. Amplification of β-actin was used for internal reference. Samples that were negative for β-actin were excluded in the methylation analysis. Human genomic DNA was methylated in vitro with M.SssI CpG methyltransferase (New England Biolabs, Beverly, MA), according to manufacturer's protocol, to generate fully methylated DNA for use as a calibrator. Fluorescence data were collected during the annealing/extension step for determination of the cycle threshold (Ct). Samples were analyzed in triplicate. A sample was considered hypermethylated for that specific gene when at least 2 of 3 replicates showed Ct-value that did not differ by more than 1 unit. Relative quantification (RQ) was used to estimate the methylation level of a particular gene for a given sample. RQ value was calculated using the 2 −ΔΔCt method, in which the ΔCt value was calculated by subtracting the Ct value of the target gene from that of β-actin (reference gene). ΔΔCt value was generated by subtracting the ΔCt value from that of the calibrator ( in vitro methylated human genomic DNA). RQ value was then multiplied by 100 to get the “percentage methylated reference” (PMR) value. A gene is deemed methylated if the PMR value > 0. Statistical analysis Distribution of the different methylation status of the different genes for the different grades of dysplasia is illustrated with box plots. Comparison of differences in methylation levels between two or more groups was done using the Mann–Whitney U test and Kruskal–Wallis test, respectively. Receiver operating characteristic (ROC) curve and area under the ROC curve (AUC) were used as a measure of accuracy for each gene. All statistical analyses were carried out using SPSS version 16.0 software package and Analyze-IT version 2.12. Statistical significance was set at p -value < 0.05 and all tests performed were two sided. Results Study population The participating hospitals are referral centres for abnormal cervical screening results for the country where the national cervical cancer screening programme has been in operation for several decades. The study cervical samples were obtained consecutively from newly referred patients. There was specific enrichment for HSIL, LSIL and normal categories. Of 63 patients with cytologically proven HSIL, 10 had SCC, 31 had CIN3, 10 had CIN2 and 12 had CIN1 on immediate follow-up with colposcopic biopsies (see Table 1 a ). This represents a 65% concordance between HSIL and histologically proven CIN3 + CIN2, i.e. (31 + 10) of 63, or a concordance of 81% if SCC is included, i.e. SCC + CIN3 + CIN2 = 10 + 31 + 10 of 63. There was 100% concordance for LSIL, where all were confirmed CIN1 on colposcopic biopsy. CCNA1, PAX1, HS3ST2, DAPK1 and TFPI2 are methylated more frequently and to a greater extent in SCC/HSIL Based on a literature search and a trial run (Methylight) on 80 cervical smear (liquid cytology) samples of shortlisted candidates, CCNA1, PAX1, HS3ST2, DAPK1 and TFPI2 were chosen for analysis. Supplementary Table 2 summarizes their known functions. The literature search was based on a PubMed search for genes known to be hypermethylated in the context of cervical cancer and increasing grades of cervical dysplasia, including those found to have aberrant expression  [33–36,44–46,52–55] . Apart from the 5 genes used in this study, we had tested SFRP1 , WT1 , MT1G , DcR1 and SPARC . We decided on the final 5 genes because they showed increasing frequency and extent of methylation with increasing grades of dysplasia ( DAPK1, HS3ST2, TFPI2 ) and consistently distinguished HSIL cases from LSIL/normal cases ( CCNA1, PAX1 ). They are also genes with functional representation from cell cycle regulation, developmental control, apoptosis and the extracellular matrix. The others were excluded as they yielded very low PMR values and did not distinguish HSIL from LSIL/normal cases. We examined the methylation status of CCNA1, PAX1, HS3ST2, DAPK1 and TFPI2 in residual cells from liquid-based cytology samples, obtained from 165 patients with cytologically proven HSIL, LSIL and normal, of whom there were 63, 49 and 53, respectively. All patients with abnormal cytology had subsequent colposcopic examinations, so these samples were then re-designated as 10 squamous cell carcinomas (SCC), 41 HSIL (31 CIN3 and 10 CIN2), 61 LSIL (12 from cytologic HSIL) and 53 normal cases. Residual cells from liquid-based cytology samples were subjected to real-time quantitative methylation-specific PCR (QMSP) analysis using Methylight. A sample was read as methylated if PMR > 0 (see Materials and methods ). If > 0, the PMR value, being a continuous variable, indicated the extent or level of DNA methylation, with respect to an internal reference (see Materials and methods ). There were 2 samples excluded from analysis due to negative β-actin results, both normal samples. Fig. 1 shows the distribution of PMR values in SCC, HSIL, LSIL and normal specimens for each of the 5 genes in the form of box plots. By visual inspection, the PMR values are highest in SCC and HSIL samples combined, compared with LSIL and normal ones. Pair-wise comparison by the Kruskal–Wallis test confirms that SCC/HSIL combined is significantly different from LSIL/normal ( p < 0.005) for all genes. There was no significant difference between the LSIL and normal categories ( p = 1.000). There was also no difference between LSIL and the group that was cytological HSIL but reclassified as LSIL/CIN1 upon histological confirmation ( p = 1.000). Methylation of at least one of the genes investigated was detected in 45 of 51 (88%) combined SCC/HSIL cases. Table 1 b shows the frequency and level of methylation of each gene in SCC, HSIL, LSIL/normal categories. In SCC, overall high frequency of methylation was observed, where 100% were methylated in CCNA1, TFPI2 and HS3ST2 , 90% in PAX1 and 80% in DAPK1 . A decreasing trend of methylation frequency was shown in HSIL followed by LSIL/normal. In HSIL, methylation frequency had dropped to 41%, 46%, 61%, 73% and 85% for PAX1 , CCNA1 , DAPK1 , HS3ST2 and TFPI2 , respectively. In LSIL/normal, the methylation frequency fell further to 2%, 5%, 25%, 30% and 72% for PAX1 , CCNA1 , HS3ST2, DAPK1 and TFPI2 , respectively. Although the methylation frequencies of TFPI2 were similar in HSIL and LSIL/normal (85% vs. 72%), the mean PMR value and PMR range illustrate a higher methylation level in HSIL. An increasing pattern of mean PMR values was observed from LSIL/normal to HSIL, then to SCC. The mean PMR values were very low in the LSIL/normal samples, being < 0.1 for all genes except TFPI2 (0.129) and HS3ST2 (0.312), indicating an overall low degree of methylation. In HSIL, the mean PMR values ranged from 2.867 to 8.533 for all genes, except DAPK1 , which had a mean PMR value of 0.858. In SCC, the mean PMR values were > 22 for all 5 genes. The range of PMR values detected in LSIL/normal for each gene was 0.000–15.054. In HSIL/SCC combined, however, the range stretched to as high as 286.603 (see PAX1 ) . CCNA1 methylation status may distinguish CIN2 from CIN3 We tested the ability of the methylation status of these genes to distinguish CIN1, CIN2 and CIN3 from one another. Pair-wise comparison using the Kruskal–Wallis test (see Table 2 ) showed that CIN3 was significantly different from CIN1 for all genes tested ( p < 0.005). There was no significant difference between CIN1 and CIN2, except for a marginal difference in DAPK1 ( p = 0.049). CCNA1 was the only gene that was able to distinguish CIN2 from CIN3 ( p = 0.016). HS3ST2 singly and in combination can similarly distinguish SCC/HSIL from LSIL/normal In order to determine the combination of genes whose methylation status best distinguishes SCC/HSIL from LSIL/normal samples, we performed ROC analysis and used area under the ROC curve (AUC) to test the performance of each gene separately and in combination ( Table 3 ). Overall, the AUC values exceeded 0.5, demonstrating the positive association of methylation status and the diagnosis. The methylation status of each of the 5 genes was able to segregate SCC/HSIL from LSIL/normal, with AUC values ranging from 0.719 to 0.822 and HS3ST2 registering the greatest AUC (AUC = 0.822, p = 4.04E-11). The AUC values for the best 2-gene, 3-gene, 4-gene and 5-gene combinations ranged between 0.836- 0.860. We compared the ROC curves using Analyze-IT. The AUC for the best 2-gene combination, HS3ST2/PAX1 (AUC = 0.836), was greater than that of the best-performing single gene HS3ST2 (AUC = 0.822), but the difference was not statistically significant ( p = 0.261), suggesting that the addition of PAX1 did not significantly improve the test performance. Further addition of a third and fourth gene showed no statistical difference ( p > 0.5) as well. Using the logistic regression equation, we analyzed the PMR values of HS3ST2 obtained from all samples and generated a curve based on SCC/HSIL = 1 and LSIL/normal = 0 (see Fig. 2 ). We evaluated the “risk probability of SCC/HSIL” by plugging the individual HS3ST2 PMR values into the fitted logistic regression model equation. After fitting the logistic regression model, we obtained β -coefficient = 0.263, p = 0.016 and intercept α = −1.167, p < 0.001. By inspection of the graphical display, the optimal cutoff value of the risk score for distinguishing SCC/HSIL from LSIL/normal was 0.237, where sensitivity is 78.4% and specificity is 74.6%. Discussion The global incidence of cervical cancer has decreased, largely attributed to cervical smear screening programmes. Given the large majority of cases that are HPV-related, the advent of cervical cancer vaccines has raised the hopes of eradicating all HPV-related cervical cancer [56,57] . This view is somewhat tempered by issues of access to the vaccines and the vaccine coverage of viruses. Regardless, it is important to elucidate the underlying molecular pathogenesis of cervical cancer, both virus- and non-virus-related. From the therapeutic standpoint, there may be common molecular targets for the development of novel agents. On a broader scale, we may also discover common points of attack for other cancers that are thought to be virally-induced. Cervical cancer is, in fact, an ideal disease model for investigation, given the easy and safe access to cells/tissue from the relevant anatomical location, and the sizeable numbers of inbuilt normal controls from cervical cancer screening programmes. Methylight [47–49] was chosen because of several inherent advantages that make it well-suited for clinical application. Its sensitive, specific and quantitative technology, absence of need for downstream manipulations and high-throughput nature are features that position it favorably for further clinical development. Cervical smear samples are heterogeneous mixtures of abnormal cells and normal cells, hence, the ability of Methylight to detect methylated alleles in the presence of 10,000-fold excess of unmethylated alleles is crucial. The presence of CpG dinucleotides in the design of PCR primers and probe serves to increase the stringency of the assay by allowing sequence discrimination at the PCR amplification level, as well as at the probe hybridization step. The trade-off in this highly specific test is that other methylation permutations will therefore be missed. As the Methylight readout is a continuous variable, its quantitative nature has the potential for even finer risk estimation. For example, a risk score can be derived from the absolute PMR values for each gene. The clinical significance of this would need to be verified on a clinical trial basis. From the practical perspective, the 1-step sample loading for analysis obviates the risk of human error. Methylight's high-throughput nature also lends itself to greater overall efficiency in dealing with clinical volume. Cervical smear screening has made a significant impact on gynaecological health, but can be fine-tuned. The distinction between CIN1 and CIN3 is robust. However, ambiguity exists in calling a specimen CIN2. There are studies looking into augmenting diagnostic accuracy with molecular data—these may be more critical in predicting cell behaviour and hence may complement or even modify usual cytological/histological definitions. Perhaps “low-risk” and “high-risk” categories can be defined, based on molecular characteristics, where “high-risk” would point towards surgical intervention, whereas “low-risk” would support watchful waiting, even lengthening the follow-up interval. For reasons of decreasing individual anxiety, clinic visits and consequent healthcare cost, it would be sensible to maximise the yield of information in a cervical smear or colposcopic biopsy. This present study examines the correlation of DNA methylation with disease severity as defined by cytology/histology. The intention is to work towards improving diagnostic accuracy and hence prognostication. The explosion of new molecular evaluation techniques have resulted in molecular subtyping of malignancies beyond traditional clinical and histopathological factors. This has enabled more accurate understanding of disease behaviour, thereby guiding treatment decisions, e.g. breast cancer [58,59] , non-Hodgkin's lymphoma [60] . Similarly in cervical neoplasia, molecular biology provides the opportunity to tailor treatment to the individual. The discordance between screening cytology and subsequent histology is commonly observed and may be explained by sampling error in the concurrent presence of several grades of dysplasia, or inter-operator variability in microscopic examination. For a patient with cytological HSIL but subsequent CIN1 on colposcopic biopsy, the whole experience can be emotionally unsettling with lingering doubt as to whether she might indeed harbour more severe disease than CIN1. Objective molecular diagnostic data may serve to reassure or even prompt earlier action. For patients with cytological HSIL/CIN2 who are not keen on surgery, e.g. adolescents, added knowledge of one's risk status may help clinical decision-making. Our study showed that the methylation status of all 5 genes is able to segregate HSIL/SCC from LSIL/normal. Statistical analysis showed that the methylation status of HS3ST2 alone performed well in segregating HSIL/SCC from LSIL/normal, without significant improvement by the addition of other genes. Hypothetically, an LSIL or normal sample that is methylated in any of these 5 genes could be at greater risk of disease progression, warranting closer follow-up. Conversely, patients without marks of methylation may abide by standard screening intervals, e.g. cytological HSIL but bearing methylation features of LSIL/CIN1. These hypotheses would require prospective clinical trials to confirm or refute. The methylation status of CCNA1 was found to be able to distinguish CIN2 from CIN3. In the context of a patient who wishes not to have surgical intervention, this might eventually prove to be useful in predicting disease progression. Again, prospective clinical trials are required to assess the clinical utility of such a test. Functionally, HS3ST2 encodes an O-sulfotransferase that is involved in the final modification step of glycosaminoglycan (GAG) chains of heparan sulfate proteoglycans. High incidence of aberrant methylation in the promoter region of HS3ST2 has been shown in breast, colon, lung and pancreatic cancers, supporting our current finding of HS3ST2 's role in cervical neoplasia [61,62] . CCNA1 is a cell cycle regulator and has been previously implicated in cervical neoplasia [34,35] . Limitations in our study include the following. Methylation analysis still depends on harvested cells, so it does not address the issue of sampling error, whether by cervical smear or colposcopy. However, the advantage Methylight offers is its high sensitivity and specificity in detecting methylated alleles in the midst of unmethylated alleles, possibly improving diagnostic accuracy. Perhaps this can partly offset the inherent variability in visual interpretation and limited reliability of colposcopy and colposcopic-guided biopsies. In fact, this lack of an accepted gold standard against which to compare methylation results limits the reliability of future work in this field. Despite this, the clinical utility of methylation in cervical neoplasia can be assessed through prospective clinical trials. To answer the question of whether the methylation status of a panel of genes can accurately predict disease progression or regression will initially involve subjecting patients’ cervical cells to usual microscopic inspection, as well as methylation analysis, at fixed time intervals, then determining patient outcomes by longitudinal follow-up. Patients whose cytological/histological and methylation results are discordant would be especially interesting. If methylation analysis turns out to accurately predict disease progression or regression, then the follow-up algorithm for these patients may be investigated: depending on the risk level, there may be earlier intervention or lengthening of follow-up interval. In summary, we have demonstrated the feasibility of using Methylight to detect the methylation status of a spectrum of cervical neoplasia. We also confirmed that higher frequency and greater extent of methylation in the genes examined corresponded with increasing disease severity: methylation status could distinguish HSIL/SCC from LSIL/normal, and may provide finer distinction within HSIL. Therefore, methylation analysis may heighten the diagnostic accuracy and complement the current diagnostic workflow. Before considering clinical trial testing, genome-wide screening approaches need to be used to identify novel biomarkers for detailed assessment, thereby formulating a robust gene panel for testing its risk prediction capability in women with cervical neoplasia. Conflict of interest statement All authors declare that there are no conflicts of interest. Acknowledgments This work was supported by a National Medical Research Council (Singapore) grant ( NMRC/EDG/0002/2007 ) awarded to E.H. Lim. We thank Annie Koh and Chai Hong Soh for providing logistical help. Appendix A Supplementary data Supplementary Table 1 Primers and TaqMan probes for QMSP assays. Supplementary Table 2 Function of genes used for quantitative methylation specific PCR (QMSP). Appendix A Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.ygyno.2010.07.028 . References [1] D.M. Parkin F. Bray J. Ferlay P. Pisani Global cancer statistics, 2002 CA Cancer J Clin 55 2005 74 108 [2] D. Solomon D. Davey R. Kurman The 2001 Bethesda System: terminology for reporting results of cervical cytology JAMA 287 2002 2114 2119 [3] M.H. Stoler M. Schiffman Atypical Squamous Cells of Undetermined Significance-Low-grade Squamous Intraepithelial Lesion Triage Study (ALTS) Group. Interobserver reproducibility of cervical cytologic and histologic interpretations: realistic estimates from the ASCUS-LSIL Triage Study JAMA 285 2001 1500 1505 [4] M.H. Stoler M.D. Vichnin A. Ferenczy FUTURE I, II, and III Investigators The accuracy of colposcopic biopsy: analyses from the placebo arm of the gardasil clinical trials Int J Cancer May 20 2010 [Epub ahead of print] [5] M. Sideri F. Schettino L. Spinaci Operator variability in disease detection and grading by colposcopy in patients with mild dysplastic smears Cancer 76 1995 1601 1605 [6] E.H. Hopman F.J. Voorhorst P. Kenemans Observer agreement on interpreting colposcopic images of CIN Gynecol Oncol 58 1995 206 209 [7] J.T. Cox M. Schiffman D. Solomon ASCUS-LSIL Triage Study (ALTS) Group Prospective follow-up suggests similar risk of subsequent cervical intraepithelial neoplasia grade 2 or 3 among women with cervical intraepithelial neoplasia grade 1 or negative colposcopy and directed biopsy Am J Obstet Gynecol 188 2003 1406 1412 [8] R.G. Pretorius W.H. Zhang J.L. Belinson Colposcopically directed biopsy, random cervical biopsy, and endocervical curettage in the diagnosis of cervical intraepithelial neoplasia II or worse Am J Obstet Gynecol 191 2004 430 434 [9] J.C. Gage V.W. Hanson K. Abbey ASCUS LSIL Triage Study (ALTS) Group Number of cervical biopsies and sensitivity of colposcopy Obstet Gynecol 108 2006 264 272 [10] T.C. Wright Jr. L.S. Massad C.J. Dunton 2006 consensus guidelines for the management of women with cervical intraepithelial neoplasia or adenocarcinoma in situ Am J Obstet Gynecol 197 2007 340 345 [11] R. Guido M. Schiffman D. Solomon L. Burke Postcolposcopy management strategies for women referred with low-grade squamous intraepithelial lesions or human papillomavirus DNA-positive atypical squamous cells of undetermined significance: a two-year prospective study Am J Obstet Gynecol 188 2003 1401 1405 [12] A.G. Ostor Natural history of cervical intraepithelial neoplasia: a critical review Int J Gynecol Pathol 12 1993 186 192 [13] M.F. Mitchell G. Tortolero-Luna T. Wright Cervical human papillomavirus infection and intraepithelial neoplasia: a review J Natl Cancer Inst Monogr 21 1996 17 25 [14] J.K. Chan B.J. Monk C. Brewer HPV infection and number of lifetime sexual partners are strong predictors for 'natural' regression of CIN 2 and 3 Br J Cancer 89 2003 1062 1066 [15] P.E. Castle M. Schiffman C.M. Wheeler D. Solomon Evidence for Frequent Regression of Cervical Intraepithelial Neoplasia-Grade 2 Obstet Gynecol 113 2009 18 25 [16] L. Gustafsson H.O. Adami Natural history of cervical neoplasia: consistent results obtained by an identification technique Br J Cancer 60 1989 132 141 [17] M.R. McCredie K.J. Sharples C. Paul Natural history of cervical neoplasia and risk of invasive cancer in women with cervical intraepithelial neoplasia 3: a retrospective cohort study Lancet Oncol 9 2008 425 434 [18] R.G. Pretorius P. Peterson F. Azizi R.J. Burchette Subsequent risk and presentation of cervical intraepithelial neoplasia (CIN) 3 or cancer after a colposcopic diagnosis of CIN 1 or less Am J Obstet Gynecol 195 2006 1260 1265 [19] J. Melnikow J. Nuovo A.R. Willan Natural history of cervical squamous intraepithelial lesions: a meta-analysis Obstet Gynecol 92 1998 727 735 [20] M.H. Schiffman H.M. Bauer R.N. Hoover Epidemiologic evidence showing that human papillomavirus infection causes most cervical intraepithelial neoplasia J Natl Cancer Inst 85 1993 958 964 [21] R.H. Kaufman E. Adam J. Icenogle Relevance of human papillomavirus screening in management of cervical intraepithelial neoplasia Am J Obstet Gynecol 176 1997 87 92 [22] N. Wentzensen M. von Knebel Doeberitz Biomarkers in cervical cancer screening Dis Markers 23 2007 315 330 [23] J.T. Horng K.C. Hu L.C. Wu Identifying the combination of genetic factors that determine susceptibility to cervical cancer IEEE Trans Inf Technol Biomed 8 2004 59 66 [24] H.W. Haverkos G. Soon S.L. Steckley W. Pickworth Cigarette smoking and cervical cancer: Part I: a meta-analysis Biomed Pharmacother 57 2003 67 77 [25] G.Y. Ho R. Bierman L. Beardsley C.J. Chang R.D. Burk Natural history of cervicovaginal papillomavirus infection in young women N Engl J Med 338 1998 423 428 [26] P.E. Castle D. Solomon M. Schiffman C.M. Wheeler Human papillomavirus type 16 infections and 2-year absolute risk of cervical precancer in women with equivocal or mild cytologic abnormalities J Natl Cancer Inst 97 2005 1066 1071 [27] A.C. Rodriguez M. Schiffman R. Herrero Rapid clearance of human papillomavirus and implications for clinical focus on persistent infections J Natl Cancer Inst 100 2008 513 517 [28] A.P. Feinberg B. Vogelstein Hypomethylation distinguishes genes of some human cancers from their normal counterparts Nature 301 1983 89 92 [29] P.A. Jones S.B. Baylin The fundamental role of epigenetic events in cancer Nat Rev Genet 3 2002 415 428 [30] M. Esteller Epigenetics in cancer N Engl J Med 358 2008 1148 1159 [31] J.G. Herman S.B. Baylin Gene silencing in cancer in association with promoter hypermethylation N Engl J Med 349 2003 2042 2054 [32] P.W. Laird The power and the promise of DNA methylation markers Nat Rev Cancer 3 2003 253 266 [33] N. Wentzensen M.E. Sherman M. Schiffman S.S. Wang Utility of methylation markers in cervical cancer early detection: appraisal of the state-of-the-science Gynecol Oncol 112 2009 293 299 [34] N. Yang J.J. Eijsink A. Lendvai Methylation markers for CCNA1 and C13ORF18 are strongly associated with high-grade cervical intraepithelial neoplasia and cervical cancer in cervical scrapings Cancer Epidemiol Biomarkers Prev 18 2009 3000 3007 [35] N. Yang E.R. Nijhuis H.H. Volders Gene promoter methylation patterns throughout the process of cervical carcinogenesis Cell Oncol 32 2010 131 143 [36] J.H. Kim Y.D. Choi J.S. Lee Assessment of DNA methylation for the detection of cervical neoplasia in liquid-based cytology specimens Gynecol Oncol 116 2010 99 104 [37] M. Weber J.J. Davies D. Wittig Oakeley Chromosome-wide and promoter-specific analyses identify sites of differential DNA methylation in normal and transformed human cells Nat Genet 37 2005 853 862 [38] A. Eden F. Gaudet A. Waghmare R. Jaenisch Chromosomal instability and tumors promoted by DNA hypomethylation Science 300 2003 455 [39] H. Cui M. Cruz-Correa F.M. Giardiello Loss of IGF2 imprinting: a potential marker of colorectal cancer risk Science 299 2003 1753 1755 [40] T.H. Bestor Transposons reanimated in mice Cell 122 2005 322 325 [41] M. Esteller Cancer epigenomics: DNA methylomes and histone-modification maps Nat Rev Genet 8 2007 286 298 [42] L. Wild J.L. Flanagan Genome-wide hypomethylation in cancer may be a passive consequence of transformation Biochim Biophys Acta Apr 14 2010 [Epub ahead of print] [43] M. Ehrlich DNA hypomethylation in cancer cells Epigenomics 1 2009 239 259 [44] P. Sova Q. Feng G. Geiss Discovery of novel methylation biomarkers in cervical carcinoma by global demethylation and microarray analysis Cancer Epidemiol Biomarkers Prev 15 2006 114 123 [45] H.C. Lai Y.W. Lin T.H.M. Huang Identification of novel DNA methylation markers in cervical cancer Int J Cancer 123 2008 161 167 [46] S.S. Wang D.J. Smiraglia Y.Z. Wu Identification of novel methylation markers in cervical cancer using restriction landmark genomic scanning Cancer Res 68 2008 2489 2497 [47] C.A. Eads K.D. Danenberg K. Kawakami MethyLight: a high-throughput assay to measure DNA methylation Nucleic Acids Res 28 2000 E32 [48] B.N. Trinh T.I. Long P.W. Laird DNA methylation analysis by MethyLight technology Methods 25 2001 456 462 [49] S. Ogino T. Kawasaki M. Brahmandam Precision and performance characteristics of bisulfite conversion and real-time PCR (MethyLight) for quantitative DNA methylation analysis J Mol Diagn 8 2006 209 217 [50] M.W. McIntosh M.S. Pepe Combining several screening tests: optimality of the risk score Biometrics 58 2002 657 664 [51] C. Stephan S. Wesseling T. Schink K. Jung Comparison of eight computer programs for receiver-operating characteristic analysis Clin Chem 49 2003 433 439 [52] A.K. Virmani C. Muller A. Rathi Aberrant methylation during cervical carcinogenesis Clin Cancer Res 7 2001 584 589 [53] K.S. Gustafson E.E. Furth D.F. Heitjan DNA methylation profiling of cervical squamous intraepithelial lesions using liquid-based cytology specimens: an approach that utilizes receiver-operating characteristic analysis Cancer 102 2004 259 268 [54] Q. Feng A. Balasubramanian S.E. Hawes Detection of hypermethylated genes in women with and without cervical neoplasia J Natl Cancer Inst 97 2005 273 282 [55] S.L. Kahn B.M. Ronnett P.E. Gravitt K.S. Gustafson Quantitative methylation-specific PCR for the detection of aberrant DNA methylation in liquid-based Pap tests Cancer 114 2008 57 64 [56] FUTURE II Study Group Quadrivalent vaccine against human papillomavirus to prevent high-grade cervical lesions N Engl J Med 356 2007 1915 1927 [57] J. Paavonen P. Naud J. Salmeron Efficacy of human papillomavirus (HPV)-16/18 AS04-adjuvanted vaccine against cervical infection and precancer caused by oncogenic HPV types (PATRICIA): final analysis of a double-blind, randomised study in young women Lancet 374 2009 301 314 [58] C.M. Perou T. Sorlie M.B. Eisen Molecular portraits of human breast tumours Nature 406 2000 747 752 [59] T. Sorlie C.M. Perou R. Tibshirani Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications Proc Natl Acad Sci USA 98 2001 10869 10874 [60] A.A. Alizadeh M.B. Eisen R.E. Davis Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling Nature 403 2000 503 511 [61] N. Perrimon M. Bernfield Specificities of heparan sulphate proteoglycans in developmental processes Nature 404 2000 725 728 [62] K. Miyamoto K. Asada T. Fukutomi Methylation-associated silencing of heparan sulfate D-glucosaminyl 3-O-sulfotransferase-2 (3-OST-2) in human breast, colon, lung and pancreatic cancers Oncogene 22 2003 274 280