Global methylation silencing of clustered proto-cadherin genes in cervical cancer: serving as diagnostic markers comparable to HPV

Epigenetic remodeling of cell adhesion genes is a common phenomenon in cancer invasion. This study aims to investigate global methylation of cell adhesion genes in cervical carcinogenesis and to apply them in early detection of cancer from cervical scraping. Genome-wide methylation array was performed on an investigation cohort, including 16 cervical intraepithelial neoplasia 3 (CIN3) and 20 cervical cancers (CA) versus 12 each of normal, inflammation and CIN1 as controls. Twelve members of clustered proto-cadherin (PCDH) genes were collectively methylated and silenced, which were validated in cancer cells of the cervix, endometrium, liver, head and neck, breast, and lung. In an independent cohort including 107 controls, 66 CIN1, 85 CIN2/3, and 38 CA, methylated PCDHA4 and PCDHA13 were detected in 2.8%, 24.2%, 52.9%, and 84.2% (P < 10−25), and 2.8%, 24.2%, 50.6%, and 94.7% (P < 10−29), respectively. In diagnosis of CIN2 or more severe lesion of the cervix, a combination test of methylated PCDHA4 or PCDHA13 from cervical scraping had a sensitivity, specificity, positive predictive value, and negative predictive value of 74.8%, 80.3%, 73%, and 81.8%, respectively. Testing of this combination from cervical scraping is equally sensitive but more specific than human papillomavirus (HPV) test in diagnosis of CIN2 or more severe lesions. The study disclosed a collective methylation of PCDH genes in cancer of cervix and other sites. At least two of them can be promising diagnostic markers for cervical cancer noninferior to HPV.


Introduction
Discovery and application of cancer biomarkers has been hampered by the extensive heterogeneities of tumorigenesis. While tumor-promoting genetic and epigenetic changes may differ among cancers of various sites and individual cancers of the same site [1], the universal carcinogenic effect of human papillomavirus virus (HPV) in virtually all cervical cancers stands out as an exception [2]. The invariable site of transformation at the squamous-columnar (SC) junction [3] allows a precise target for screening. Various diagnostic biomarkers such as high-risk HPV DNA, p16, and methylated genes have been discovered from scraping specimens at this site with unprecedented success [4][5][6][7].
In the decade-long evolution of carcinoma in situ (CIS) toward invasion, extensive epigenetic changes are

Cancer Medicine
Open Access required for epithelial to mesenchymal transition (EMT) [8] before invasion and metastasis. This involves methylation silencing of genes functioned in cell adhesion [9]. Among the superfamilies of cell adhesion molecules, clustered proto-cadherin (PCDH) is the largest one. The three clusters, namely a, b, and c PCDH gene clusters (Human Genome Organization nomenclature PCDHA@, PCDHB@, and PCDHG@, respectively), distribute in tandem at chromosome 5q31. PCDHA@ and PCDHG@ contain a tandem array of 15 and 22 variable exons, respectively, and each is controlled by a specific promoter. Expression of these genes involves both differential promoter activation and alternative pre-mRNA splicing to the downstream constant exon [10,11].
Taking cervical carcinogenesis as a model, this study aims to globally search for methylated genes that are specific for cancer invasion. Among the overrepresented genes were members of clustered PCDH genes, which were silence by methylation in cancer cell lines of multiple sites including the cervix. Progressive methylation of two flanking members of PCDHA@, PCDHA4, and PCDHA13 was observed in an independent cohort. In comparison with HPV test, methylation test of the two genes is more specific and equally sensitive in detecting in CIN2 or more severe lesions.

Studied subjects
This study was approved by Institutional Review Board of Tzu Chi General Hospital, Hualien, Taiwan, and informed consent was signed by each subject. Subjects including an investigation cohort and a testing cohort were enrolled from December 2005 to May 2013. In the investigation cohort, thirty six each of cases and controls were enrolled. The case group included cervical scrapings of 16 squamous cell carcinomas (SCC), 4 adenocarcinomas (AdenoCA), 14 CIS (carcinoma in situ), and 2 cervical intraepithelial neoplasia 3 (CIN3). The control group included 12 scrapings each of cytology/histology diagnosis of "negative for cervical neoplasia" (Normal), "reactive changes or inflammation" (Inflammation), and CIN1. The testing cohort was consisted of 107 Normal or Inflammation, 66 CIN1, 85 CIN2/3, and 38 invasive carcinomas (CA). Demographic data of these two cohorts were showed in Table 1. In both cohorts, subjects with newly diagnosed cervical neoplasia were enrolled and managed according to a standard guideline issued by the National Health Research Institute of Taiwan (http://tcog. nhri.org.tw/doc/gog_a.pdf). All subjects with CIN1 or more severe lesions were examined by colposcopy with a biopsy tissue proof. Healthy women undergoing regular Pap smears with a "Normal" or "Inflammation" results were recruited as controls.

Clinical specimens and nucleic acid extraction
Cervical tissues with CIN or invasive carcinoma were procured during the procedures of colposcopic biopsy, conization, and radical hysterectomy. Normal cervical tissues were obtained from patients receiving hysterectomy for benign tumors of the uterus. All the tissue specimens were collected within 20 min after resection and were frozen in À80°C. Cervical scrapings were collected by a cytobrush before colposcopy, cervical biopsy, or other procedures. The scrapings were transferred into a universal tube containing 3 mL of phosphate buffered saline. After thorough agitation, dispensing, they were stored at À80°C. Nucleic acids were extracted from cervical scrapings and the tissues by using DNA Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions.
DNA methylation array and data analysis DNA extracted from cervical scrapings of the case and control groups were each pooled and fragmented by probes which recognize CpG islands and CpG sites adjacent (at À250 to 500 bp) to transcription start sites genome-wide. From the scaled log2-ratio data of each probe, a fixed-length (750 bp) window was placed around each consecutive probe and the one-sided Kolmogorov-Smirnov (KS) test was applied to determine whether probes were drawn from a significantly more positive distribution of intensity log-ratios than those in the rest of the array. The resulting score (peak score) for each probe was the Àlog10 P-value from the windowed KS test around that probe.
Cell culture and 5 0 -aza-2 0 -deoxycytidine treatment All mediums contained 10% fetal bovine serum and 1% penicillin/ streptomycin. For demethylation treatment, 1 9 10 6 cells were seeded in medium containing 5 0 -aza-2 0 -deoxycytidine (5-aza) (Sigma) and the medium was replaced every day for a total of 4 days. Different concentrations of 5-aza (2.5, 5, 7.5, and 10 lmol/L) were tried in pilot studies for each cancer cell. The 10 lmol/L concentration turned out to give the best demethylation effect and did not affect cell survival.

RT-PCR analysis of clustered PCDH transcripts
Total RNA was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA) and was reverse transcribed using the First Strand cDNA Synthesis kit (Roche) and subjected to PCR with primers listed in Table S1. The following PCR condition was used with minor, uncritical variations was designed for each PCDH gene that gave amplification at the logarithm phase: one cycle at 95°C for 10 min, 35 cycles at 95°C for 30 sec/58-62°C for 30 sec/72°C for 30 sec, and one cycle at 72°C for 5 min. As the loading control, the b-actin gene was amplified under the same condition. The relative expression level was normalized by b-actin and the ratio of the density of PCR products in the agarose gel was calculated by using the ImageJ program (National Institutes of Health, Rockville Pike, Maryland).

Bisulfite genomic sequencing
Genomic DNA was bisulfite-modified using the DNA Methylation Gold Kit (Zymo Research). Based on the methylation array results, we focused on the hypermethylation region of each gene. The primers were designed from the MethPrimer website (http://www.urogene.org// methprimer/index1.html) (Table S1) and annealed to bisulfite DNA at 55°C. Amplified products were cloned into TA-cloning plasmids, transformed into competent cells, and the sequences were determined from randomly chosen colonies.

Methylation-specific PCR (MS-PCR)
Based on the BGS results, primers covering the highly distinctive methylation sites between normal and cancer specimens were designed for methylation-specific PCR (MS-PCR) via the MethPrimer website. MS-PCR was done by using Gold Taq DNA polymerase (Applied Biosystems, Foster City, CA). Table S1 lists the primers used. DNA of peripheral white blood cells was used for nonmethylated template control and those treated with CpG methyltranferase M. SssI (New England Biolab, Ipswich, MA) as the methylated control. The relative ratio of density of amplified product from methylation-specific PCR (MSP) and unmethylation-specific PCR (USP) resolved in the agarose gel was calculated by using the ImageJ program.

HPV detection and typing
The presence of HPV DNA and genotypes was determined by consensus PCR and reverse blot hybridization as described previously [12]. Briefly, the MY11 and biotinylated GP6+ consensus primers were used to amplify a fragment of~192 bp in the L1 open reading frame of the HPV genome for 40 cycles. The PCR products were then hybridized with an Easychip HPV Blot (King Car; Yuan-San, I-Lan County, Taiwan), which included oligonucleotide probes of 39 different HPV types, and were visualized with biotinylated antibodies and alkaline phosphatase conjugation.

Statistical analysis
Data was analyzed by using SPSS 15.0 (SPSS, Inc., Chicago, IL). The relationship between cervical neoplasia and biomarkers was analyzed by Fisher's exact test.

Widespread methylation of clustered PCDH genes in cervical cancers
In total, 2152 peaks representing 2227 genes were found to be specifically methylated in the cancer/precancer group with an "average peak score" over 1.3 (or P < 0.05) ( Table 2 and Data S1). Gene ontology analysis for biological process revealed enrichment of genes involved in cell adhesion signaling (n = 67) and cell adhesions (n = 78) with Enriched Score of 6.94 and 2.98, respectively ( Table 2). A lot of these cell adhesion genes are members of clustered PCDH genes. They include four members in each of the PCDHA@ (PCDHA4, A8, A10, A13), PCDHB@ (PCDHB3, B6, B7, B14), and PCDHG@ (PCDHGA12, GB6, GB7, GC3) cluster (Table 3). In particular, individual PCDH genes in clusters are hypermethylated consecutively. For instance, at a less stringent criterion of "any peak score" of >1.3 (or P < 0.05), nine consecutive members (from PCDHA4 to PCDHA13) of the PCDHA@ were hypermethylated in the case group but not the control group (Fig. 1). These data suggest that clustered PCDH genes may be hypermethylated in cervical cancer/precancer lesion.
Clustered PCDH genes are methylationsilenced in multiple carcinoma cells The twelve methylated members of clustered PCDH were underexpressed in a wide variety of cancer cell lines including carcinoma of the cervix (4/4), endometrium (2/ 2), liver (3/3), head and neck (3/3), breast (1/3), and lung (1/2). A majority of these underexpressed genes could be reexpressed upon treating the cells with 10 lmol/L 5-aza, a demethylation agent (Table 4, Fig. 2A). To further exploration of this PCDHA@ cluster, we chose the two flanking members, PCDHA4 and PCHA13, for further characterization. The methylation status of these two genes, before and after demethylation treatment, was confirmed by MS-PCR, in which methylation product was reduced and unmethylation one was increased after demethylation treatment (Fig. 2B). Detailed mappings of methylation at CpG islands around the first exon were shown in Figure 3. All the CpG sites, including five each recognized by the MS-PCR primers for PCDHA4 and PCDHA13, were hypermethylated in the SCC or CIS samples, whereas low methylation densities were noted in the normal samples. These results suggest most of clustered PCDH genes, especially PCDHA4 and PCDHA13, were methylation-silenced, not only in cervical cancer cells but also in other cancer cells.
Methylation of the PCDHA4 and PCDHA13 genes are strongly correlated with severity of cervical neoplasia In an independent testing cohort of cervical scrapings of different severity, progressive methylation of the PCDHA4 was noted from 2.8% in Normal or Inflammation, 24.2% in CIN1, 52.9% in CIN2/3, to 84.2% in CA (P < 10 À25 ). For PCDHA13, methylation was noted from 2.8%, 24.2%, 50.6%, to 94.7%, respectively (P < 10 À29 ). As a comparison, high-risk HPV was detected in 77.6% of CIN2/3 and 86.8% of CA, but was also found in 15.9% of Normal or Inflammation and 45.5% of CIN1 (P < 10 À21 , Table 5). Methylation of these two genes was further verified in tissue specimens in this cohort. As showed in Table 6, methylation of either of the two genes was noted in 12.5% of Normal cervix, 11.1% of  Figure 1. Methylation of members of the PCDHA@ gene cluster in cervical cancer. Genomic organization of the 13 members of the PCDHA@ gene cluster reveals a gene-specific exon 1 followed by shared exon 2, 3, and 4 (boxes). Peak score and P-value of the most significant peak of each gene are given. The first exons with a most significant peak scores higher than 1.3 (P < 0.05) are indicated by gray boxes, and those with an average peak score higher than 1.3 are indicated by black boxes. Peaks are detected (from the P-value data) from at least two probes that have P < 0.05 which is equal to a Àlog(10) value or peak score of 1.3. 2 The feature is the transcription start site. CIN1, 57.1% of CIN2/3, 78.9% of invasive CA and 100% of metastatic CA (P < 10 À6 ). The data demonstrate a positive correlation of frequency of methylation of PCDHA4 and PCDHA13 with severity of cervical neoplasia, and suggest a potential to be applied in clinical diagnosis.
Testing of methylated PCDHA4 or PCDHA13 is more specific than HPV test in detecting CIS and invasive cervical cancer We analyzed the performance of testing methylated PCDHA and high-risk HPV in detection of cervical neoplasia of different severity in the testing cohort. Both methylated PCDHA tests had a lower sensitivity but higher specificity than HPV test in detecting CIN2+, CIN3+ or CA. A combination test of methylation of PCDHA4 or PCDHA13 achieved a better sensitivity (97.4%) and specificity (65.5%) than the HPV test in detecting CA. For early detection of CIN2 lesion, testing of methylated PCDHA4 or PCDHA13 is equally sensitive (74.8%) but more specific (80.3%) than HPV test ( Table 7). Given that HPV type 16 and 18 are the most oncogenic HPV types causing over 70% of cervical cancer worldwide and may serve as a specific diagnostic marker [13], we tested the performance of combination of HPV16/18 with the two DNA methylation markers. As showed in Table  S2, testing for the presence of any of these four markers in cervical scrapings increased the sensitivity but decreased the specificity for detecting CIN2 or more severe lesions. The effect is more prominent in detection of CIN2 and CIN3 than of CA. A benefit of increasing sensitivity with less compromise of specificity was found when HPV16/18 was added to methylated PCDHA4, but not to the methylated PCDHA13 or the combination of two PCDHA@ members.

Discussion
In this genome-wide promoter methylation study, collective methylation silencing of clustered PCDH genes was observed in cervical cancers. In addition, carcinoma cells of different origins including the endometrium, the head and neck, the liver, the lung, and the breast were each found to harbor silenced PCDH genes of differing spectra. This finding is consistent with the long range epigenetic silencing (LRES) of clustered gene loci in cancers, as illustrated in breast cancer [14] and in Wilm's tumor [15]. Recently, the cis-regulatory elements of PCDHA@ were identified and the possible mechanism of competition between individual variable exon promoters was characterized [16]. Interestingly, as Table 4. Methylation silencing of PCDH genes in cancer cells of the cervix, liver, head and neck (H&N), endometrium, lung, and breast. shown in Table 4, not all of the silenced members within the PCDH locus can be reexpressed by demethylation treatment. This is in accordance with the findings of coordinate suppression of a 4-Mb segment in colon cancer where silencing of individual genes within a LRES locus appears to be dependent on a domain-wide nonpermissive chromatin configuration rather than the methylation status [17]. Differing from previous genome-wide studies of DNA methylation in cancers, which used cancer tissues as sources of exploration [18][19][20], this study discovered methylated markers from swab samples of the spectrum   of cervical carcinogenesis. Tumor tissues, especially those from precancerous lesions and the normal epithelial tissue, are comprised mainly of stromal cells and typically do not represent the neoplastic or epithelial cells. In contrast, cervical scrapings are comprised mainly of epithelial cells with a minor population of inflammatory cells, and can be a better source to search for biomarkers. A design of pooling samples of normal, inflammatory, and mildly dysplastic scrapings may reduce the individual variations of cell components and more specifically target on severe neoplasia. Indeed, candidate genes were validated in a high yield. Eleven of the 12 candidate clustered PCDH genes were confirmed in cervical neoplasia and most of them were also methylation-silenced in other cancer cells tested (Table 4).
In accordance with a general phenomenon of EMT with loss of cell-cell adhesions during the invasion of cancer, numerous cell adhesion genes have been reported as silenced by DNA methylation in carcinomas. These include classical cell adhesion genes such as CDH1 (E-cadherin) [21], CDH4 (R-cadherin) [22], and CDH13 (T-cadherin) [23], and nonclustered PCDH genes such as PCDH10 [24,25] PCDH20 [26], FAT4 [27]. Methylations of clustered PCDH genes are much less reported previously [15,28]. The present study showed a widespread methylation silencing of clustered PCDH genes in multiple cancer cells. The finding that PCDHA4 and PCDHA13 are methylated drastically in CIN2/3 and invasive CA stages suggests a role of these clustered adhesive molecules in maintaining the integrity of normal and dysplastic cells.
Clustered PCDH genes are predominantly expressed in the brain and are sophisticatedly expressed during development of central nerve system [10]. A complex system of gene regulation, expression, protein cleavage, and interaction among different members was observed [10,11]. Studies of functions of these clustered cell adhesion molecules are complicated by their molecular diversity with cis-homodimers and cis-heterodimers on the same cell surface and by cross-regulations of the encoding genes [29,30]. Recently, the mosaic or mixed DNA methylation states of PCDHA@ genes have been characterized in mouse neural cells and in vivo [31]. This individual cell-specific and allele-specific control of these clustered genes may be responsible for development of cell identities of complex and highly organized tissues. Echoing to this notion, this study revealed down-regulation of specific spectra of clustered PCDH genes in cancer cells of differing sites, suggesting tissuespecific adhesion profiles that are overridden in cancer invasion.
Each one or specific combinations of hyper-methylated genes unveiled in this study may be effective biomarkers of cancers. As shown in this study, the frequency of methylation of PCDHA4 and PCDHA13 correlates well with the progression of cervical carcinogenesis. Compar- Fisher's exact test. 100.0 P-value 1 P = 5.6 9 10 À6 P = 5.7 9 10 À6 P = 1.3 9 10 À7 CIN3, cervical intraepithelial neoplasia 3. 1 Fisher's exact test. ing to high-risk HPV test, methylated PCDHA4 and PCDHA13 were as frequently found in invasive cervical cancer but was much less frequently found in normal or CIN1 (Table 5). In detection of CIN2/3 or more severe lesions, a combination test of methylation of either of the two genes achieved a better specificity (80.3% vs. 72.8%) and equal sensitivity (74.8% vs. 75.6%) as compared with the HPV test. In detection of CIN3 or more severe lesions, it even surpasses the HPV test (specificity 77.6% vs. 67.7%; sensitivity 79.8% vs. 75.0%). Currently, the main concern of using HPV test as screening and triage tool for high-grade cervical lesions is its lower specificity, due to the fact that most HPV infections are transient and only the persistent infection is associated with an increased risk of high-grade CIN and cancer. While specifically testing HPV type 16 and 18 may be more specific and adding HPV16/18 to the methylated gene markers has been showed to improve the sensitivity of detecting HSIL or above lesions [32], our data showed an improvement of performance when it was combined with PCDHA4 methylation marker, but not for PCDHA13, which performed better than PCDHA4, or the combination of the two PCDHAs. Our data reveal the combination test of these two methylated genes can be a promising test in early detection of cervical cancer, which is at least as good as HPV test and even better when specifying the diagnostic target to mare severe lesions of CIS or CA ( Table 7). Dozens of methylated genes have been discovered to aid the screening and diagnosis of cervical cancer [33]. Among them, CADM1, MAL, and PCDH10 have been established as tumor suppressor genes, and their methylation are highly related to the severity of cervical noeplasia [33][34][35]. They have been investigated as triage tool for equivocal Pap smear result (PCDH10) [34], or for high-risk HPV infection (CADM1 and MAL) [35,36] with performances better than or noninferior to the traditional triage method of HPV test and Pap smear, respectively. Comparing to methylated PCDH10 [24], our markers are more sensitive (63-64% vs. 40%) and less specific (89% vs. 100%) in detecting CIN2 or more severe lesions. Comparing to the data of methylated CADM1 and MAL in a trial of high-risk HPV-infected women [36], the two markers are slightly more sensitive and specific than methylated MAL and far more better than methylated CADM1. Given that the majority of methylated PCDHAs detected in CIN1 were of low intensity in MSP assay, an adoption with quantitative measurement may further improve the specificity of the test. Further head-to-head comparison is required to see which marker is better.
In summary, this study identified collective methylation of clustered PCDH genes in cervical cancer. Testing these methylated genes may be valuable in screening or diagnosing cervical neoplasia. Further studies are encouraged to see whether they can also be biomarkers in cancers of other sites.

Supporting Information
Additional Supporting Information may be found in the online version of this article: Table S1. Primer sequences of PCDH genes.