|Home | About | Journals | Submit | Contact Us | Français|
We wanted to identify the most promising methylation marker candidates for cervical cancer early detection.
A systematic literature review was performed in Medline and weighted average frequencies for methylated genes stratified by tissue source and methods used were computed.
51 studies were identified analyzing 68 different genes for methylation in 4376 specimens across all stages of cervical carcinogenesis. 15 genes, DAPK1, RASSF1, CDH1, CDKN2A, MGMT, RARB, APC, FHIT, MLH1, TIMP3, GSTP1, CADM1, CDH13, HIC1, and TERT have been analyzed in 5 or more studies. The published data on these genes is highly heterogeneous; 7 genes (CDH1, FHIT, TERT, CDH13, MGMT, TIMP3, and HIC1) had a reported range of methylation frequencies in cervical cancers of greater than 60% between studies. Stratification by analysis method did not resolve the heterogeneity. Three markers, DAPK1, CADM1, and RARB, showed elevated methylation in cervical cancers consistently across studies.
There is currently no methylation marker that can be readily translated for use in cervical cancer screening or triage settings. Large, well-conducted methylation profiling studies of cervical carcinogenesis could yield new candidates that are more specific for HPV-related carcinogenesis. New candidate markers need to be thoroughly validated in highly standardized assays.
Persistent infections with carcinogenic human papilloma virus (HPV) types are causally linked to the development of cervical cancer . The development of invasive cancers from the initial viral infections takes decades, permitting detection and treatment of CIN2 and CIN3 (CIN2+). While cytological screening has substantially reduced cervical cancer incidence and mortality where it has been successfully implemented, it is limited by low single-test sensitivity and poor reproducibility for equivocal and minor abnormalities . Despite the recently introduced preventive vaccines against HPV16 and HPV18, screening needs to continue, since only about 70% of cervical cancers will be prevented. However, HPV vaccination will further reduce the efficiency of cytological screening. Therefore, new screening modalities need to be evaluated and pursued .
Large randomized trials have shown that adding HPV DNA testing to cytology greatly increases the sensitivity of primary screening for CIN2+ [3–5] and HPV testing alone is considered by some to be a plausible primary screening method . However, since a positive HPV DNA test almost always indicates a transient infection rather than risk of eventual invasive cervical cancer the positive predictive value (PPV) of HPV testing is low and a strategy needs to be developed to triage HPV DNA positive women.
Much effort has been put into identifying new biomarkers for CIN2+ to improve risk stratification, distinguishing women with benign infection from those requiring more intensive management .
Methylation of CpG islands within gene promoter regions can lead to silencing of gene expression. Methylation of tumor-relevant genes has been identified in many cancers: p16 methylation is the paradigm for epigenetic inactivation of a tumor suppressor gene, leading to abrogation of cell cycle control, escape from senescence, and induction of proliferation. MLH1 methylation has been identified as the first step in development of sporadic microsatellite unstable colorectal cancers. Likewise, many genes associated with tumor development have been found methylated in various cancer sites. Methylation has been detected already at precancerous stages, suggesting that methylation markers may have value in cervical cancer screening . Furthermore, methylated DNA is a stable target and allows for flexibility of assay development.
Over the last decade, a growing number of studies evaluating methylation of host genes in cervical tissue have been published. Most of the candidate methylation markers analyzed in cervical tissues were selected because altered methylation was previously observed in other types of cancer. Reflecting the technical development in the methylation field over the past 20 years, many different technologies were used in these studies, analyzing both biopsy specimens as well as cytology samples.
The aims of this review are (1) to summarize the results of published methylation studies analyzing cervical tissues and cells, including the specimen types, markers and assays evaluated and (2) to assess the opportunities and challenges facing this line of research.
A systematic literature research was performed of studies published in Medline until April 7, 2008, using the keywords: (methylation AND cervical),( methylation AND CIN),( methylation AND cervix). Only investigations that evaluated clinical specimens (either histologic or cytologic) were included and methylation frequencies for specific genes were considered; studies that analyzed methylation patterns for disease clustering without presenting individual frequency data by gene were not considered. Since we focused on host gene methylation, we did not include studies analyzing methylation of HPV genomes in cervical cancer progression.
The histologic diagnoses were cervical cancer, CIN3, CIN2, CIN1, or normal cervical tissue. The cytologic interpretations were high-grade squamous intraepithelial lesions (HSIL), low-grade squamous intraepithelial lesion (LSIL), atypical squamous cells- rule out high-grade lesion (ASC-H), atypical squamous cells of undetermined significance (ASC-US), or negative for intraepithelial lesion or malignancy (NILM) . Glandular lesions except for invasive adenocarcinoma were not included due to lack of research on these poorly-defined lesions. Several studies did not report exact cytologic and histologic categories, but grouped different categories (e.g. CIN1-3+ASC+SIL, CIN2-3+HSIL).
All gene names were checked on the homepage of the HUGO Gene Nomenclature Committee (www.genenames.org) and the official HUGO gene name was assigned. Multiple designations were found for several genes, including the following: CDH1 (cadherin, CDH1), CTNNB1 (b-catenin, CTNNB1), CADM1 (TSLC1, IGSF4, CADM1). Since the CDKN2A gene locus has 2 different promoters with CpG islands and encodes 2 different transcripts (p16INK4a, p14ARF), data on CDKN2A were used for the analysis of the p16INK4a CpG island while data on p14 was used for the p14ARF CpG island.
For each gene, we report the weighted average and the range of methylation frequencies observed in all included studies. To analyze host gene methylation in progression to cancer, we compared methylation frequencies of the 15 most frequently analyzed genes reported for normal tissue (including all studies reporting normal tissue, n=24 studies), cancer precursors (defined as high-grade cervical intraepithelial neoplasia (CIN) or squamous intraepithelial lesion, n=12 studies), and cancers (n=33 studies).
To analyze different methylation frequencies between squamous cell cancer (SCC) and adenocarcinoma (AC), we included 12 genes that were analyzed in at least two studies reporting frequencies for both differentiations.
Different methods were used to analyze gene methylation: Methylation specific PCR (MSP) is based on conversion of unmethylated Cytosin to Uracil by bisulfite treatment. Subsequently, amplification is performed using different primers designed for amplification of methylated or unmethylated CpGs. Quantitative MSP (Q-MSP, e.g. MethyLight) is based on the same principle but uses Real Time PCR amplification and increases specificity by adding a probe for the amplified sequence. Other methods used were bisulfite sequencing (analyzing a larger region for methylated CpGs than MSP/Q-MSP) and restriction based methylation detection (analyzing only single CpGs by methylation-sensitive restriction enzymes). Different sample types were analyzed in the studies, including fresh frozen tissue, exfoliated cells, and paraffin embedded tissue.
We stratified the reports on methylation frequencies by the following 9 combinations of assay and sample type: A) Fresh frozen tissue and MSP; B) Exfoliated tissue and MSP; C) Paraffin embedded tissue and MSP; D) Unspecified tissue and MSP; E) Fresh frozen tissue and Q-MSP; F) Exfoliated tissue and Q-MSP; G) Paraffin embedded tissue and Q-MSP; H) Unspecified tissue and Q-MSP; I) Fresh frozen tissue and other methods.
One study reported the highest frequencies for all 6 markers included in the study that belong to the 15 markers analyzed in detail . In sensitivity analysis, the exclusion of this study did not substantially change the results.
The initial Medline search yielded 3546 abstracts on methylation AND cervical/cervix/CIN. Fifty-one studies were identified that described methylation frequencies of human genes in cervical samples (Supplemental table 1).
The majority of the studies used MSP (32 of 51, 63%), followed by Methylight (7 of 51, 14%), other quantitative MSP protocols (5 of 51, 10%), bisulfite sequencing (4 of 51, 8%), and other methods (3 of 51, 6%). In 24 studies (47%), fresh frozen material was used for methylation analysis, followed by exfoliated cells (12 of 51, 24%), and paraffin embedded tissue (7 of 51, 14%). The most frequent combination of sample material and methods used in the studies was MSP with fresh frozen tissue samples (15 of 51, 29%) (Table 1). Eight studies reported the use of microdissected specimens [11–18].
47 of 51 studies (92%) analyzed cervical cancer samples, 26 of 51 (51%) normal cervical samples, and 16 of 51 (31%) included CIN or ASC/SIL (CIN1-3, ASCUS, LSIL, HSIL, in the following summarized as “CIN”). In total, 4376 samples covering 2836 cancer samples, 841 normal samples, and 709 CIN samples from the 51 studies are included in this review. Due to the heterogeneous CIN samples included in the different studies, no weighted mean and frequency ranges are presented for CIN.
In total, 68 different genes were analyzed in the studies, 31 genes were analyzed in more than 1 study, 15 in 5 or more studies, and 6 in ten or more studies (Table 2).
The genes analyzed in 5 or more studies (by descending frequency) were DAPK1, RASSF1, CDH1, CDKN2A, MGMT, RARB, APC, FHIT, MLH1, TIMP3, GSTP1, CADM1, CDH13, HIC1, and TERT. Gene names, gene ontology terms and data from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database on methylation in other cancer sites are summarized for these 15 genes in Supplemental Table 2.
Overall, there was a wide range of methylation frequencies in cervical cancers reported for most genes (Table 2, Figure 1). On the extremes, large variable frequencies were reported for TIMP3, MGMT, and CDH1 and somewhat less variation was observed for CADM1and GSTP1. Based on mean percentages weighted by study size, the most frequently methylated genes were CDH1 (58%), DAPK1 (57%), CADM1 (55%), and TERT (55%). MGMT and RASSF1 showed methylation frequencies below 20% and GSTP1 and MLH1 were rarely if ever methylated in cervical cancers.
In normal cervix tissue, all genes except for HIC1 had weighted mean methylation frequencies below 30% (Table 2, Figure 2). However, some genes showed a wide range of reported frequencies, including FHIT, HIC1, and CDH1. These three genes showed a high variation in both cancer and normal tissue.
We analyzed the reported range and the weighted mean methylation frequencies for the above described 15 genes in studies that explicitly presented results for the CIN2/CIN3 and HSIL categories. All genes with high differences of methylation frequencies between normal tissue and cancers showed intermediate frequencies in CIN2/CIN3/HSIL (Supplemental Figure 1). Of these genes, the highest mean frequency was reported for CADM1 (33%), followed by CDH1 (29%), DAPK1 (29%), and TERT (29%). The highest weighted mean methylation frequencies in CIN2/CIN3/HSIL were reported for HIC1 (59%) and APC (34%), but both genes had lower methylation frequencies in cancer (Supplemental Figure 1).
We observed a high degree of heterogeneity in the methylation frequencies reported for cervical cancers in the different studies. To determine if the variation of methylation frequencies was related to cancer type, we compared methylation frequencies between SCC and AC. Several studies did not specify the differentiation of cancers; others reported only SCCs. 19 studies reported methylation frequencies for SCCs and ACs. In total, 1062 SCCs and 396 ACs were included in these studies, indicating an oversampling of ACs compared with typical population diagnostic distributions. A total of 30 different genes have been compared between SCC and AC in these studies. 12 genes (RASSF1A, DAPK, CDKN2A, APC, CDH1, MGMT, TIMP3, HIC1, RARB, ESR1, FHIT, and MLH1) were analyzed in 2 or more studies (Table 2). Three genes, APC, TIMP3, and HIC1, showed >20% higher weighted mean methylation frequencies in AC as compared to SCC, while DAPK and CDH1 were more frequently found methylated in SCC. Still, the variation of frequencies reported for most of the 12 markers was similarly high in SCC and AC; only APC and TIMP3 had non-overlapping ranges of methylation frequencies in SCC and AC.
To analyze the influence of specimen type and methylation detection methods on the methylation frequencies in cancers, we stratified the frequencies by the nine combinations of tissue source and assay described in the Methods section (Table 1, Supplemental Figure 2). Most combinations were used only in few studies. Though, based on limited numbers within each stratum, the analysis showed frequency ranges that were similar to the overall range. For example, CDH1 was analyzed with 4 different combinations in at least 2 studies and showed a wide variation among studies sharing each of the combinations. Similarly, CDH13 and FHIT showed a wide variation in the sample source/assay combinations with 2 or more studies. In contrast, DAPK1 was analyzed in 4 combinations in at least 2 studies and showed very consistent results irrespective of sample source and assay used (Supplemental Figure 2). In the strata D and E (MSP with unspecified tissue, Q-MSP with frozen tissue), CDH1, CDH13, and DAPK1 were analyzed in an identical set of studies, permitting direct comparison. While CDH1 and CDH13 showed a high frequency range between methods, the same studies produced much more precise results regardless of method for DAPK1. Thus, DAPK1 results are not demonstrably methods-dependent.
The detection of methylated genes from cervical specimens is technically feasible and represents a source for detecting potential biomarkers of relevance to cervical carcinogenesis. In particular, there is the ultimate hope of finding methylation markers that, among HPV-infected women, would indicate the presence of CIN2+ and risk of cancer.
One striking conclusion of our survey of 51 studies is that methylation frequencies for the same gene vary widely between studies. This degree of heterogeneity combined with the low number of studies analyzing the same genes precluded us from performing a rigorous meta-analysis. In fact, we were unable to identify highly consistent results for most genes even when restricting analyses to studies of similar size or those that used common specimen sources or similar assays. These results suggest that the frequency of certain methylation markers may also vary for reasons related to differences in populations, specific features of assay protocols, chance or other unidentified factors. For example, CDH1 and FHIT, two genes with highly variable methylation frequencies in cancer between studies also show variable detection in normal tissue. In contrast, DAPK1 has been measured much more precisely over a wide range of heterogeneous studies, including those analyzing CDH1 and CDH13.
We described in detail the properties of 15 methylation markers that were analyzed in 5 or more studies. While we cannot discount the possibility of a promising biomarker among the remaining 53 candidates, there was insufficient information available to evaluate these candidates further.
The most important prerequisite for a potential biomarker is that it must be reliable in its measurement. Among the 15 genes analyzed in detail, 7 (CDH1, FHIT, TERT, CDH13, MGMT, TIMP3, HIC1) had a reported range of methylation frequencies in cervical cancers of greater than 60% between studies. Stratification for analysis method or specimen type used did not resolve the observed variations. However, since few studies used identical conditions to analyze the same markers, in most cases, numbers were insufficient to evaluate the influence of assay type and specimen on methylation frequencies. At the moment, assay reliability for these methylation markers therefore cannot be properly addressed since methods are poorly standardized. We acknowledge the possibility that the wide range of frequencies reported for some genes contrasting the more consistent measurement of few other genes in similar studies could be related to either unreliable assays for these specific genes or to biological variation.
Another prerequisite for a good biomarker is that it has high sensitivity and high specificity for disease detection, resulting in a high positive predictive value. Among the eight genes with limited variability, all but one (APC) had methylation frequencies below 5% in normal tissue. Of these, three had average frequencies of at least 30% in cancers (DAPK1, CADM1, and RARB). Based on these results, we conclude that there is currently no single methylation marker that has the appropriate performance to serve as a cervical cancer biomarker, especially since methylation frequencies in CIN2/CIN3/HSIL, the targets of most cervical cancer screening programs, were even lower for all genes.
Some groups have proposed the use of methylated gene panels to obtain an appropriate performance for cervical cancer screening. From the summary data in this review, the combination of DAPK1, CADM1, and RARB would appear the most promising. However, without a formal evaluation, it is unclear if methylation of these markers is mutually exclusive, entirely independent, or associated to some degree, which would affect the overall coverage of the marker panel and therefore has important impact on the sensitivity of marker combinations. We note that Feng et al. recently showed up to 74% sensitivity and 95% specificity for detecting cervical cancers using a panel of three candidate genes, DAPK1, RARB, and TWIST1 . Our review would support further evaluation of two of these genes (DAPK1 and RARB).
Finally, we consider the biological relevance to further inform our evaluation of candidate genes. Notably, a clear role of methylation in carcinogenesis has been demonstrated only for 6 genes (DAPK1, RASSF1, CDKN2A, RARB, MLH1, and GSTP1, see supplemental table 2). Two of the 15 genes, TERT and GSTP1, encode for proteins with oncogenic functions, so their methylation is not likely to drive carcinogenesis. CDKN2A (encoding p16), is methylated in several tumors early in carcinogenesis. In cervical cancer, however, p16 is found strongly overexpressed due to HP-oncogene mediated release of E2F from RB. Recently, it has been demonstrated that p16 methylation does not affect protein expression in cervical cancer . Thus, most of the 15 genes analyzed in this study have no strong a-priori for potentially being important in cervical carcinogenesis. Furthermore, none of the 7 genes with the highest variation of methylation frequencies is among the genes with a clearly defined role of methylation in the carcinogenesis of other tumors. This evaluation suggests that the current genes may not be optimal candidates for cervical carcinogenesis. To date, only few agnostic profiling studies for aberrant methylation in cervical carcinogenesis have been performed [21–23]. We therefore believe that, if methylation markers for cervical cancer screening are to be further pursued, future large well-conducted investigations should also incorporate a discovery effort specifically for methylated genes in cervical cancer. During the last years, several new platforms (e.g. microarray format, bead array format, 454 sequencing format) have been developed that allow for accurate high-throughput genome-wide DNA methylation profiling . Markers or marker panels identified in these approaches could be translated to smaller scaled assays such as Methylight to be used in cervical cancer screening.
In summary, to identify promising methylation candidates for cervical carcinogenesis, further tissue-based profiling studies using reliable and validated assays are needed. In addition, analyses of exfoliated tissue (e.g. stored in liquid based cytology media) will be critical as this would represent the most likely specimen used for cervical cancer screening and triage. Both efforts will require large well-powered epidemiologic studies designed to properly identify and then validate candidate methylation markers and panels of markers for utility in the early detection of cervical cancer.
The bars indicate the weighted average methylation frequencies. Vertical lines indicate range of reported frequencies. White: Normal cervical tissue. Grey: High grade cervical intraepithelial neoplasia (HGCIN)/High grade squamous intraepithelial lesion (HSIL). Black: Cervical cancer.
Precis: Although methylation markers for cervical cancer screening have been analyzed in multiple studies, there are currently no convincing methylation marker candidates to be used in cervical cancer screening.
Conflict of interest statement: All authors declare that there are no conflicts of interest.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.