A four-gene methylation marker panel as triage test in high-risk human papillomavirus positive patients

Authors


  • Conflict of interest: Prof. A.G.J. van der Zee is member of the scientific advisory board of Oncomethylome S.A., Liège, Belgium. V. Deregowski, G. Verpooten, L. Dehaspe and W. van Criekinge are employees of Oncomethylome BVBA, Liège, Belgium.

Abstract

Cervical neoplasia-specific biomarkers, e.g. DNA methylation markers, with high sensitivity and specificity are urgently needed to improve current population-based screening on (pre)malignant cervical neoplasia. We aimed to identify new cervical neoplasia-specific DNA methylation markers and to design and validate a methylation marker panel for triage of high-risk human papillomavirus (hr-HPV) positive patients. First, high-throughput quantitative methylation-specific PCRs (QMSP) on a novel OpenArray™ platform, representing 424 primers of 213 cancer specific methylated genes, were performed on frozen tissue samples from 84 cervical cancer patients and 106 normal cervices. Second, the top 20 discriminating methylation markers were validated by LightCycler® MSP on frozen tissue from 27 cervical cancer patients and 20 normal cervices and ROCs and test characteristics were assessed. Three new methylation markers were identified (JAM3, EPB41L3 and TERT), which were subsequently combined with C13ORF18 in our four-gene methylation panel. In a third step, our methylation panel detected in cervical scrapings 94% (70/74) of cervical cancers, while in a fourth step 82% (32/39) cervical intraepithelial neoplasia grade 3 or higher (CIN3+) and 65% (44/68) CIN2+ were detected, with 21% positive cases for ≤CIN1 (16/75). Finally, hypothetical scenario analysis showed that primary hr-HPV testing combined with our four-gene methylation panel as a triage test resulted in a higher identification of CIN3 and cervical cancers and a higher percentage of correct referrals compared to hr-HPV testing in combination with conventional cytology. In conclusion, our four-gene methylation panel might provide an alternative triage test after primary hr-HPV testing.

Infection with high-risk human papillomavirus (hr-HPV) is causally linked to cervical carcinogenesis.1 Cervical cancer incidence is reduced by cytologic screening, although cytologic morphologic assessment of cervical scrapings is not ideal, because its sensitivity is only ∼ 55% for CIN2+.2 Recently, preventive vaccines against hr-HPV-16 and hr-HPV-18 have been introduced in the Western world, which will reduce the incidence of cervical neoplasia significantly. However, these vaccines do not cover 100% of cervical cancers, while it will take over decades before HPV vaccination affects cervical neoplasia incidence. Therefore, screening needs to be continued. Simultaneously, efficiency of cytologic screening in population-based screening programs will be reduced by vaccination, due to a gradual decline of cervical neoplasia prevalence. Hr-HPV testing of cervical scrapings has been shown to improve sensitivity of cervical screening,3, 4 but is also associated with low specificity, especially in a young screening population.5 For that reason, other markers are needed especially to improve the positive predictive value for population-based screening of cervical neoplasia.

Promoter methylation of tumor suppressor genes has been reported to be an early event in carcinogenesis.6 Gene promoter methylation analysis of several cervical cancer-specific genes has been suggested as an alternative diagnostic tool for early detection of cervical neoplasia by QMSP.7, 8 Various methylated gene promoters for cervical neoplasia have been tested, mainly based on previously reported methylation status in cervical neoplasia or other tumor types. None of these markers have been validated in large population-based studies on cervical cancer screening so far, due to too low detection rates of cervical carcinoma in study populations enriched for cervical neoplasia.9

Different approaches for identification of new (cervical) cancer-specific methylated gene promoters have been described, such as pharmacological unmasking of cell lines with subsequent microarray analysis, restriction landmark genomic scanning and differential methylation hybridization.10–14 Previously, we identified 13 genes that are methylated specifically in cervical cancer.12 Of these genes, we have recently shown that C13ORF18 was the one of the best presently available methylation markers for early detection of cervical cancer because of its high specificity (97%) and sensitivity (71%).15 However, the sensitivity for CIN2 is 25% and CIN3 51% of C13ORF18 and therefore necessitates additional markers. The OpenArray™ (Biotrove, Inc.)-based QMSP experiments could be a novel high-throughput application for identifying specific (cervical) cancer methylation markers. This new methodology allows real-time, 33 nL PCRs on a 3,072 holes microarray with the size of a microscope slide.16 Therefore, this assay can be used to measure the methylation status of a considerable number of genes and patients samples in a relatively short time.

The aim of the present study was to identify new methylation markers for cervical neoplasia and to design and validate a methylation marker panel for triage of hr-HPV positive patients. Additionally, a scenario analysis for population-based screening programs for cervical neoplasia was performed to compare our four-gene methylation marker panel to conventional cytology as a triage test after primary hr-HPV testing.

Patients and methods

General strategy

For the identification and validation of new cervical cancer-specific methylation markers the following strategy was applied (see Fig. 1). First, OpenArray QMSP was performed with 424 primers (representing 213 genes) on DNA isolated from frozen tissue of cervical cancers (n = 84) and normal cervices (n = 106). Methylation markers were ranked based on their power to discriminate between cervical cancer and normal cervices. In the second step, the first 20 methylation markers, as derived from step 1, were selected for validation by LightCycler© MSP on DNA isolated from frozen tissue of cervical cancers (n = 27) and normal cervices (n = 20). In both steps, macrodissected frozen tissue sections were used, allowing methylation analysis of relatively pure neoplastic cells in sufficient amounts of DNA. After LightCycler MSP, methylation markers were ranked based on receiver operating curve (ROC) analysis, and therefore markers with the highest discriminative power between normal and cervical cancer tissue were ranked on top.

Figure 1.

Study strategy.

Together with C13ORF18 (previously identified by our group),15 the best three methylation markers from steps 1 and 2 were selected for first validation by QMSP on cervical scrapings from a large series of cervical cancer patients (n = 74) and healthy age-matched controls (n = 69). This third step enabled us to investigate the discriminative power of methylation analysis for cervical cancers compared to normal scrapings and to analyze if methylation is related to stage or histology (in the cervical cancer group) or to age (in the group of controls). The series of patients used for QMSP analysis was an independent series of patients, as only 13 patients (8 cervical cancer patients and 5 normal cervices) were overlapping with the LightCycler experiments. In a fourth step (second validation), the potential as a diagnostic tool of QMSP for these four genes was evaluated in a large series of scrapings (n = 148) from patients, referred to our department with an abnormal Pap smear, taken during population-based screening. Finally, a scenario analysis for population-based screening programs for cervical neoplasia was performed to compare our methylation marker panel to conventional cytology as a triage test after primary hr-HPV testing, as the population attending for screening has a different prevalence of CIN2+ in comparison with the study population used in step three. This analysis provides information on the expected performance of our methylation marker panel in a population-based screening program.

Patients

All patients referred to our outpatient clinic for (possible) cervical neoplasia are asked to participate in various studies on biomarkers in cervical neoplasia during their initial visit at the outpatient clinic in the University Medical Center Groningen (UMCG). Frozen tissue and cervical scrapings are prospectively collected and stored in our tissue bank from cervical cancer patients, from patients with normal cervices planned to undergo a hysterectomy for nonmalignant reasons and from patients referred with an abnormal Pap smear. All patients referred with an abnormal Pap smear were diagnosed by biopsy or large loop excision of the transformation zone, and all tissue samples were scored by an experienced gynecologic pathologist (H.H.). Clinicopathological data are retrieved from patient files and stored in a large anonymous database. For all cervical cancer patients, an examination under general anesthesia is performed for staging in accordance with the International Federation of Gynecology and Obstetrics (FIGO) criteria. All patients from whom material was obtained gave written informed consent. This study was approved by and followed the ethical guidelines of the Institutional Review Board of the UMCG.

For the OpenArray experiments, frozen tissue specimens from 84 cervical cancer and 106 normal cervices from our tissue bank were randomly selected. Stage of cervical cancer patients was 2 (2%) FIGO stage IA1, 53 (63%) FIGO stage IB, 16 (19%) FIGO stage IIA, 9 (11%) FIGO stage IIB, 2 (2%) FIGO stage IIIB and 2 (2%) FIGO stage IV. The histological classification of cervical cancer patients was 71 (85%) squamous cell carcinoma, 9 (11%) adenocarcinoma and 4 (4%) adenosquamous carcinoma. Median age of the cervical cancer patients was 50 years (range, 25–91). All normal cervix tissue samples were obtained from patients without a history of abnormal Pap smears who were planned to undergo a hysterectomy for nonmalignant reasons. Indications for hysterectomy were fibroids, prolaps uteri, adenomyosis, hypermenorrhea or a combination of these. All cervical tissues were judged as histopathological normal. Median age of the women with normal cervices was 49 years (range, 32–83). Samples for LightCycler MSP were randomly selected from these series of specimens (27 cervical cancers and 20 normal cervices).

QMSP analysis was performed on cervical scrapings randomly selected from our tissue bank (74 cervical cancer patients, 69 normal cervices and 148 patients referred with an abnormal Pap smear). Stage of cervical cancers patients was 38 (51%) FIGO stage IB, 10 (14%) FIGO stage IIA, 17 (23%) FIGO stage IIB, 7 (10%) FIGO stage IIIB and 2 (2%) FIGO stage IV. Histological classification of the cervical cancer patients was 61 (82%) squamous cell carcinoma and 13 (18%) adenocarcinoma. Patients referred with an abnormal Pap smear were divided in 39 (26%) without dysplasia (no dysplasia), 39 (26%) CIN1, 30 (20%) CIN2, 38 (26%) CIN3 and 2 (2%) microinvasive cervical carcinoma. Median age of cervical cancer patients was 47 years (range, 27–85) of controls 47 years (range, 30–68) and of patients referred with an abnormal Pap smear 35 years (range, 20–65).

Sample collection and DNA isolation

From all samples, 10 frozen tissue sections (10 μm) were cut, while for normal cervices, macrodissection was performed to enrich for epithelial cells. Before and after cutting, a hematoxylin and eosin slide was made per tumor and normal cervices. All slides were checked for proportion of cervical cancer cells (cut-off value > 50%) or presence of normal epithelium and the criteria as mentioned apply to all tissue samples used in this study. Cervical scrapings were collected using an Ayre's spatula and endocervical brush. The collected cervical cells were suspended in 5 ml of phosphate-buffered saline [(6.4 mM Na2HPO4; 1.5 mM KH2PO4; 0.14 M NaCl; 2.7 mM KCl (pH 7.2)] and kept on ice until further processing. Of these 5-ml cell suspensions, 1 ml was used to make cytospins for cytomorphologic assessment and 4 ml was centrifuged, washed and the cell pellet was snap-frozen in liquid nitrogen and stored at −80°C as described previously.8 DNA isolation was performed using standard salt-chloroform extraction and isopropanol precipitation. Precipitated DNA was resuspended in 150 μl of Tris–EDTA buffer (10 mM Tris; 1 mM EDTA, pH 8.0). Genomic DNA was amplified in a multiplex PCR according to the BIOMED-2 protocol to check the DNA quality.17

OpenArray™ and LightCycler® experiments

The OpenArray platform (Biotrove) consists of 3,072 through-holes loaded with 250 nM of each different primer per hole. Assays were custom made; representing 424 primers of 213 cancer-specific methylated genes; on average, one gene was represented by two different primers. These genes have been described to be discriminatively more methylated in cervical cancers compared to normal cervices as shown by us12, 14 and others.10, 18 After bisulfite treatment on denatured genomic DNA by the EZ DNA methylation kit according to manufacturer's protocol (Zymogen, BaseClear, Leiden, the Netherlands), β-actin copy number was determined by QMSP. The equivalent of 1,500 β-actin copies per sample was applied per subarray of an OpenArray plate. Real-time QMSP was carried out in a total volume of 33 nL based on SYBR® Green I chemistry in an Applied Biosystems 7900HT Sequence Detector System. Plates were cycled according to the manufacturer protocol (www.biotrove.com). In the final step, melting temperature (Tm) analysis was performed. Threshold cycles (Cts) were automatically calculated by the OpenArray qPCR analysis software. Supporting Information Table 1 shows the Ct and Tm values for all samples and all methylation markers.

Table 1. Sequences methylation markers
inline image

LightCycler® MSPs were based on SYBR® Green I chemistry, and the amplicon size was determined by capillary electrophoresis. An equivalent of 20 ng of bisulfite-treated genomic DNA was used per reaction, and the total reaction volume was 10 μl.

Quality control was performed using in vitro methylated and unmethylated leucocyte DNA samples. From the LightCycler platform, the Ct and Tm were calculated by the Roche LightCycler 480 software (Software release 1.5.0). From the capillary electrophoresis platform, the band sizes and band heights were calculated by the Caliper software (Caliper Labchip HT version 2.5.0, Build 195 Service Pack 2). A sample was called positive if the melting temperature and product size were within the specified boundaries of the measured in vitro methylated DNA sample. Boundaries were for melting temperature ±2°C and for product size ±10 bp. In addition, the Ct had to be under 40 cycles, and the correct band intensity height had to be higher than 20, the latter being a relative number calculated by the Caliper software.

OpenArray data analysis

A novel biomarker scoring strategy based on a weighted combination of a marker's discriminating power and its robustness was developed. The OpenArray system relates the presence of an amplification product to Ct and Tm. A sample was considered to be methylated, if Ct was less than 42 and Tm fell within an automatically derived marker-specific Tm interval. For a given candidate methylation marker, the Tm-interval was chosen, such that cancer samples tended to be classified as “methylated” (i.e., within the Tm-interval) and the controls as “unmethylated.” To rank candidate Tm-intervals, we used a scoring function—based on a one-sided binomial test—that favors intervals with significantly more cancer samples than could be expected from the overall percentage of cancer samples in the total set of samples. For each methylation marker, we retained the highest-ranking Tm-interval that includes the in vitro methylated samples.

Next, we challenged the robustness of the selected Tm interval by randomly perturbing the original dataset. For this, we used the Tm-variance (TmVar) derived from repeated measurements on in vitro methylated samples. More specifically, we generated for each marker a series of datasets in which the Tm values were replaced by random values selected from the normal distribution with mean Tm and variance TmVar/NLev, where NLev is a noise level ranging from 1 to 10. Per level, 1,000 datasets were generated, and, for each dataset, the quality of the Tm-interval was recorded using the above binomial scoring function. The average score—as measured over 10 times 1,000 random perturbations—was used to rank biomarkers in ascending order. In the resulting ranking, methylation markers with good discriminating power that is not affected by increasing levels of noise are sorted on top. So, essentially, we measure robustness by observing the discriminating power of the marker (with the above binomial scoring function, which favors markers with high sensitivity and specificity) as it changes when we replace the actual Tm values by increasingly different values. The markers that are least affected by these progressively more severe perturbations of the original data will be selected as the most robust ones.

QMSP on cervical scrapings

QMSP was performed after bisulfite treatment on denatured genomic DNA. Bisulfite treatment was performed with the EZ DNA methylation kit according to manufacturer's protocol (Zymogen, BaseClear, Leiden, The Netherlands). To correct for total DNA input, the housekeeping gene β-actin was used as a reference. QMSP was carried out in a total volume of 20 μl in 384 well plates in an Applied Biosystems 7900 Sequence Detector (Applied Biosystems, Nieuwekerk a/d IJsel, the Netherlands). Each sample was analyzed in triplicate. The final reaction mixture consisted of 600 nM of each primer, 250 nM probe, 1× QuantiTect Probe PCR Kit (Qiagen, Leiden, The Netherlands) and 50 ng of bisulfite converted genomic DNA. As a positive control, serial dilutions of genomic leukocyte DNA, in vitro methylated with the CpG methyltransferase (M.Sss I; New England Biolabs, Beverly, MA), were used in each experiment. A DNA sample was considered to be methylated if at least two of three exponential curves were visible with a Ct-value below 50 and DNA input of at least 225 pg β-actin. All amplification curves were visualized and scored without knowledge of the clinical data.

HPV detection and typing

In all samples, presence of hr-HPV was analyzed by PCR using HPV16 and HPV18 specific primers. For all HPV16 and HPV18 negative cases, a general primer-mediated PCR using primer set GP5+/6+ was performed, with subsequent nucleotide sequence analysis, as described previously.8 As control for the specificity and sensitivity of each HPV-PCR, serial dilutions of DNA extracted from HPV16-positive CaSki and HPV18-positive HeLa cell lines were included.

Statistical analysis

All analyses were carried out using the SPSS software package (SPSS 16.0, Chicago, IL). To determine the three best performing methylation markers after LightCycler® MSP, ROC analysis was performed. Presented are AUCs as well as sensitivity and specificities. Differences in detection rates between normal scrapings, cervical cancers scrapings and CIN scrapings by methylation markers were analyzed using the χ2 test. Diagnostic performance for methylation markers and hr-HPV DNA testing was expressed in sensitivity and specificity with a cut-off for CIN2+ or CIN3 or higher (CIN3+), respectively. Observed differences with a p value <0.05 were considered statistically significant.

Scenario analysis for population-based screening program

Scenario analysis was performed in a virtual population of 100,000 women. The following assumptions were based on two population-based screening studies concerning hr-HPV testing and cytomorphological assessment.3, 4 In this virtual population 1,100 CIN2+ patients were presumed, divided in 363 CIN2, 704 CIN3 and 33 cervical cancers. Hr-HPV testing was estimated to have a sensitivity of 95% and specificity of 94% for CIN2+. For conventional Pap smear as a triage test after hr-HPV testing, sensitivity for CIN2/CIN3 was estimated to be 70% and for cervical cancer 80%. The specificity of conventional Pap smear for CIN2+ after hr-HPV testing was estimated to be 82%. An analysis was performed for two scenarios: (1) primary hr-HPV testing followed by Pap smear and (2) primary hr-HPV testing followed by the methylation test. The scenario analysis was performed for one screening round without follow-up taken into account.

Results

Methylation marker selection

OpenArray data analysis representing 424 primers of 213 cancer-specific methylated genes revealed a ranked list of genes based on the discriminative power between normal cervices and cervical cancers including the robustness of the assay. The 20 highest ranked methylation markers were subsequently evaluated by LightCycler MSP (see Table 1 for sequences methylation markers and see Table 2 for OpenArray ranking). Table 2 shows LightCycler MSP analysis in frozen tissue of 27 cervical cancers and 20 normal cervices ranked on the AUC to represent the best combination of sensitivity and specificity. In addition, ROC analysis was performed for the early stage cervical cancers only (n = 19), because this group might be more relevant and meaningful than the total group of cervical cancers. This analysis revealed that the AUC was approximately the same resulting in the same ranking of genes (data not shown). Some genes (SLIT2, WT1 and DKK2) in Table 2 are enlisted by both primer pairs, indicating that a larger region in the CpG island was methylated discriminatively. Furthermore, also the markers EPB41L3, TERT, SOX1, LMX1A, SLIT1, ALX3 and RPRM were represented by multiple MSP primer pairs and were ranked on the OpenArray on place 44, 113, 58, 245, 146, 191 and 259, respectively. The first three markers JAM3, EPB41L3 and TERT were selected for further clinical validation.

Table 2. ROC analysis LightCycler MSP experiments1
inline image

QMSP and hr-HPV on cervical scrapings from cervical cancer patients and normal cervices

In this first validation step, 143 cervical scrapings (74 cervical cancer patients and 69 normal cervices) were included. Because of low-DNA input, 5 cervical cancer patients and 10 normal cervices were excluded from further analysis. Methylation markers were positive in cervical cancer scrapings in 83–90% and in normal cervices only in 5–14% (p < 0.0001, see Table 3). Our methylation panel (JAM3, EPB41L3, TERT and C13ORF18) detected 94% of cervical cancers. Hr-HPV was detected in 88% of cervical cancer scrapings and in 3% of normal cervical scrapings (Table 3). The two hr-HPV positive scrapings from normal cervices were not positive for one of the selected methylation markers (JAM3, EPB41L3 and TERT).

Table 3. Methylation markers and hr-HPV detection in cervical scrapings
inline image

QMSP and hr-HPV on cervical scrapings from CIN patients

In total, 148 cervical scrapings from patients referred with an abnormal Pap smear were included in the second validation step of the study. Because of low-DNA input, five patients were excluded from further analysis. Table 3 summarizes the results of QMSP for the four genes separately and combined. Methylation of EPB41L3 and JAM3 showed the highest detection level (65–68%) in CIN3 patients, while our methylation panel detected 81% of CIN3 patients. Methylation markers separately or the combination of four markers were highly discriminative between CIN1 or lower and CIN2+ patients (p < 0.0001). Hr-HPV was detected in 47% of patients with no dysplasia, 68% of CIN1, 72% of CIN2, 95% of CIN3 and 100% in microinvasive cervical carcinoma (Table 3).

Diagnostic performance of methylation markers and hr-HPV

In Table 4, the diagnostic performance of methylation markers is shown. The sensitivity of methylation markers separately and as a panel for CIN2+ varies between 37 and 65% and for CIN3+ between 54 and 82%. The specificity of methylation markers separately and as a panel varies between 79 and 100% for CIN2+. Sensitivity of hr-HPV test was 85% for CIN2+ with a low specificity 43%. The diagnostic performance of the methylation markers separately and as a panel were also analyzed in hr-HPV positive patients. The sensitivity of our panel in hr-HPV positive patients was 71% for CIN2+ patients and 84% for CIN3+ patients.

Table 4. Diagnostic performance for patients referred with an abnormal Pap smear
inline image

Scenario analysis population-based screening for detection cervical neoplasia

Because the population attending for screening has a different prevalence of CIN2+ in comparison with our enriched study population, a scenario analysis was performed (see Table 5). This analysis provides information on the expected performance of the methylation test in a population-based screening program. Overall, the detection of CIN2+ was almost equal between primary hr-HPV/Pap smear and primary hr-HPV/methylation test scenarios (p = 0.934), while the detection of CIN3 and cervical cancers was higher for hr-HPV testing in combination with our methylation panel (p = 0.021). The percentage correct referrals was also higher in the methylation test scenario (p < 0.001).

Table 5. Hypothetical scenario analysis population-based screening program
inline image

Discussion

In this study, OpenArray technology in combination with LightCycler® MSP experiments and QMSP identified three new methylation markers (JAM3, EPB41L3 and TERT), highly specific for the detection of cervical neoplasia. Together with C13ORF18, previously identified by our group,15 these genes were incorporated in a methylation panel that had a better diagnostic performance as a triage test after hr-HPV testing than currently widely applied cytology. Our hypothetical scenario analysis for population-based screening in silico evaluated a strategy of combining hr-HPV testing with our methylation panel and indicates that this approach may result in a higher detection of CIN3 and cervical cancers and a higher positive predictive value, when compared with hr-HPV with cytology as triage test. Our methylation test outperformed cytology as triage test, despite the fact that our hypothetical scenario analysis took a relatively high sensitivity of cytology for CIN2/3 (70%) and cervical cancer (80%) after hr-HPV testing into account, which is in line with the study of Bulkmans et al.4 However, sensitivity of conventional cytology and liquid-based cytology is ∼ 55%2 in routine screening, and the question remains if sensitivity indeed will improve more than 15% after positive hr-HPV testing in routine screening. Because of future impact of HPV vaccination on CIN prevalence, primary hr-HPV testing can be used to compose a high prevalence population. This will be needed for any triage test to maintain an acceptable NPV and PPV.19

To our knowledge, only one recent large study (n = 236) tested two methylation markers, CADM1 and MAL, in hr-HPV positive patients. In this study, different thresholds for methylation markers were used, resulting in different combinations of sensitivity and specificity. In our study, the methylation test, without setting any thresholds, reached a sensitivity of 84% for hr-HPV positive CIN3+ patients with a specificity of 69%. In the study of Hesselink et al.,20 a threshold resulting in a sensitivity of 84% for CIN3+ led to a specificity of 53%. The strength of the study of Hesselink et al.20 was that patients were selected from a population-based screening cohort. A limitation of this study compared to our study however was that, in our study, a larger cohort of CIN3+ patients including cervical cancer patients was analyzed.

Strengths of the current study are (1) all potential methylation markers for detection of cervical cancer as known from literature so far were selected for the OpenArray experiments, (2) the OpenArray platform is a high-throughput assay that needs only small amounts of DNA, (3) a validation strategy was developed to select the best performing markers and (4) the selected markers were validated in scrapings from patients with pre-malignant cervical neoplasia. Limitations of the current study are (1) the OpenArray assay analyzes only CpG islands in the promoter region (∼ 500 bp around the transcription start site) of a limited number of 213 genes, (2) in the discovery phase a relatively small number of patients was used, (3) for the discovery phase both squamous cell carcinomas as well as adenocarcinomas were used, which may decrease the power to identify methylation markers, specific for squamous cell carcinomas or adenocarcinomas. However, as a representative cohort of cervical cancers was used, we optimized our chances of finding biomarkers for both squamous cell carcinomas as well as adenocarcinomas, (4) validation of the markers on cervical scrapings was performed on a relatively small series of patients, enriched for cervical neoplasia and should be validated in an independent cohort of patients and (5) the scenario analysis is based on a hypothetical population and is performed for only one screening round and takes not into account the data of follow-up.

An important advantage of the methylation test is that it can be performed on the same sample as used for hr-HPV testing, and therefore methylation-based tests as triage tests are also promising in the context of self-sampling. Because self-sampling is suitable for hr-HPV testing but less appropriate for cytological examination,21 and probably other non-DNA-based assays (like p16 staining), methylation-based tests could also be an alternative in this approach. Recently, we investigated the feasibility of methylation testing in combination with cervico-vaginal lavages.22

Another advantage of methylation-based testing is that in future, this test may be further improved by identifying even more sensitive and specific methylated gene promoters, while such significant improvement is unlikely for the diagnostic performance of cytology. Hence, further effort will be put in reaching a preferable 100% sensitivity for CIN2 and CIN3 lesions, as the gynecologist treats all patients with CIN2+. This goal remains subject of further research. A possible strategy could be to discover new CIN2/3 methylation markers by exploiting new genome-wide methodologies such as MeDIP/MIRA in combination with microarray analysis or next-generation sequencing and to use our validation strategy to propose these new high-grade cervical neoplasia-specific methylation markers for evaluation in population-based screening trials.

Because methylation patterns of the promoter regions of the genes identified in our current study are specific for cervical cancer and not present in normal cervical epithelium, it is interesting to speculate on the role of these genes in cervical carcinogenesis. JAM3 belongs to the family of junctional adhesion molecules, but how these molecules are exactly involved in carcinogenesis is not known so far.23 Expression of TERT results in telomerase activity and telomerase is observed in particularly 79% of cervical cancers patients.24 Therefore, methylation of the gene promoter of TERT seems to be a contradiction in the concept of gene silencing due to methylation. However, it has been shown that binding of the CTCF protein to the first exon of the promoter region of TERT resulted in repression and methylation of the CTCF-binding site resulted in expression of TERT.25 The EPB41L3 gene, also known as DAL-1, is a tumor suppressor gene26; therefore, loss of EPB41L3 might be involved in tumor progression.27C13ORF18 encodes an open reading frame with unknown function presently.15 Currently, we are investigating the role of C13ORF18 in cervical carcinogenesis.

In conclusion, we identified three new high-grade cervical neoplasia-specific methylation markers for the detection of cervical neoplasia. Our methylation test might provide a potential novel triage test after hr-HPV testing in population-based screening programs. Its possible application deserves to be further explored in a population-based cohort.

Ancillary