Triage of high-risk human papillomavirus-positive women by methylated POU4F3

Insufficient specificity of the high-risk human papillomavirus (hrHPV) assay in primary cervical cancer screening results in unnecessary referral. Additional assays to triage hrHPV-positive women are needed to improve molecular cervical cancer screening. DNA methylation is a promising biomarker in cervical cancer. We evaluated the clinical performance of potentially methylated genes as a triage assay for hrHPV-positive women. We conducted a retrospective hospital-based case–control study in Taiwan. Cervical scrapings were collected before colposcopy for hrHPV testing and quantitative methylation-specific PCR (QMSP) of 16 genes. Five genes, POU4F3, HS3ST2, AJAP1, PAX1, and SOX1, were prioritized for the clinical performance to triage hrHPV-positive women. Two hundred cervical scrapings were randomly classified into a training set (n = 111) and testing set (n = 89). All samples were tested for hrHPV using a Hybrid Capture II (HCII) assay. HrHPV-positive women were subjected to DNA methylation analysis by QMSP. In the training set, the receiver operating characteristic (ROC) curves defined the optimal methylation index (M-index) cutoff values for discriminating CIN3+ from CIN1/normal, which then were applied to the testing set. Among the five genes, POU4F3 revealed the highest area under the ROC curve (AUC) (0.86; 95 % CI, 0.78–0.95) in detecting CIN3+. In the testing set, POU4F3 revealed the best clinical performance in triage of hrHPV-positive women with a sensitivity of 74 % and specificity of 89 % for detecting CIN3+. POU4F3 methylation analysis is a potential molecular tool for triage in detecting CIN3+ in hrHPV-positive women. The combined use of broad-spectrum HPV assay and POU4F3 methylation analysis as a new generation of molecular cervical cancer screening warrants further population-based study.


Background
Cervical cancer is a common medical problem in women, with 528,000 new cases and 266,000 deaths globally in 2012 indicating the need to develop and implement an effective cancer screening strategy [1]. The Papanicolaou (Pap) smear for cytological examination has been used for the detection of precancerous cellular abnormalities of cervical cells for decades and has lessened the disease burden by reducing the mortality and morbidity of cervical cancer [2]. The Pap smear or cytology test has high specificity for cervical intraepithelial neoplasia; however, it has many drawbacks such as suboptimal sensitivity [3] and moderate accuracy to detect relevant lesions and subjective diagnosis of cervical abnormalities with poor reproducibility [4]. Oncogenic high-risk human papillomavirus (hrHPV) infection is a well-known etiology of cervical cancer [5]. Because the duration of the initial hrHPV infection until the development of invasive cancer is long, assay of HPV DNA as a screening tool is appealing [6,7]. However, HPV infection is transient in nature, and only few infected lesions further progress as invasive cancer [8]. Insufficient specificity of the HPV DNA assay results in a high false-positive rate and extra medical burden because of the consequent high colposcopy referral rate [6]. Findings of HPV-positive assay results also cause adverse psychosocial impact [9]. Therefore, an additional triage assay is required to improve HPV-based molecular cervical cancer screening [10,11].

Generation of methylation cutoff values for triage of hrHPV-positive women in the training set
We tested the clinical performance of POU4F3, HS3ST2, and AJAP1 methylation for the triage of hrHPV-positive women (Fig. 1). The independently enrolled 200 women were randomly classified into two groups with a trainingto-testing ratio of 1:1 ( Table 2). There was no significant difference in the age (P = 0.17) and diagnosis distribution in the training set and testing set. Methylation levels of POU4F3, HS3ST2, and AJAP1 in hrHPV-positive women increased with disease severity (Fig. 2a-c). The optimal methylation index (M-index) cutoff values for detecting CIN3 + were 38 for POU4F3 and 2 for HS3ST2 and AJAP1 as defined by receiver operating characteristic (ROC) curves. The areas under the ROC curves (AUC) were 0.86 (95 % CI, 0.78-0.95) for POU4F3, 0.82 (95 % CI, 0.71-0.92) for HS3ST2, and 0.71 (95 % CI, 0.59-0.83) for AJAP1 ( Fig. 2d-f). Because we have previously discovered and tested SOX1 and PAX1 genes as potential biomarkers [29], we also included the data of these two genes in this study to compare their clinical performance. At the optimal M-index cutoff values, the sensitivities of POU4F3, HS3ST2, AJAP1, SOX1, and PAX1 in discriminating CIN3 + among hrHPV-positive women were 79, 67, 63, 78, and 70 %, respectively, whereas the specificities were 78, 89, 64, 71, and 89 %, respectively ( Table 3).

Validation of the clinical performance of methylated genes in the testing set
Cervical scrapings of 55 hrHPV-positive women out of 89 women were analyzed further in the testing set for DNA methylation levels (Fig. 1). The testing set validated that POU4F3 methylation analysis conferred the best clinical performance among five potential candidates with 74 % sensitivity and 89 % specificity (Table 3). When stratified by histology, POU4F3 and AJAP1 methylation testing did not miss any invasive cancer patients (Table 4). AJAP1 methylation had better performance in detecting CIN3/CIS lesions than POU4F3 (70.8 vs. 62.5 %). However, more CIN1 lesions were detected using AJAP1.

Discussion
Previous studies support the concept that DNA methylation could be a potential molecular biomarker for detection of cervical lesions [12,15,16,28,30]. An ideal methylation biomarker should have better specificity than HPV testing and better sensitivity than cytology when applied as a primary screening tool. Recent studies proposed the alternative role of methylation biomarkers as a triage method for hrHPV-positive women [22-24, 26, 27]. More high-risk HPV genotype detection may have a better chance to include more women at risk for triage in the primary screening. In addition, the distribution of HPV type varies across continents because 16, 31, 33, and 18 are prevalent in Europe, and 16, 58, 52 and 18 are prevalent in the Asia-Pacific region [31,32]. We used the Hybrid Capture II (HCII) assay for hrHPV testing, which assays 13 high-risk genotypes simultaneously [7,32,33]. The present study demonstrated that DNA methylation analysis as a triage for hrHPV-positive women is feasible. The POU4F3 methylation analysis confers the best clinical performance when combined with the HCII assay. In this study, the primary objective was to use broad-spectrum hrHPV testing capable of detecting more susceptible women for further triage with POU4F3 methylation to achieve a better sensitivity. Further hrHPV subtype analysis may clarify typespecific correlation with POU4F3 methylation, which may be useful in estimating the impact of molecular screening strategy using HPV detection followed by methylation triage in post-vaccination era. POU4F3 is located on chromosome 5q32 and plays various biological functions, such as regulation of transcription, cellular and metabolic processes, organ development, cellular differentiation, nervous system development, neurogenesis, and generation of neurons [34]. The function of POU4F3 in cancer biology remains largely unknown. POU4F3 hypermethylation in cervical cancer and glioma suggests its suppressor role in cancer [28,34]. This study supports the concept that POU4F3 could be a potential triage biomarker for hrHPV-positive women.
In the present study, a single gene, POU4F3, has a specificity of 89 % in detecting CIN3 + in hrHPV-positive women with limited compromise in sensitivity (79 to 74 %), which is better than the specificities previously published using FAM19A4 (67 %) [27], or a panel of two genes CADM1-M18/MAL-M1 (71-83 %) [22,23,35,36], or a panel of at least two out of five methylated biomarkers (77 %) [26], or a panel with four methylated biomarkers (69 %) [24] or comparable to the specificity of JAM (88 %) [37]. We propose a scenario for the combination of HPV assay and POU4F3 methylation analysis for cervical cancer screening   (Fig. 3). However, it requires further independent validation together with additional standalone biomarker. Women without hrHPV infection undergo follow-up 3 to 5 years later [38]. Women with hrHPV infection will undertake POU4F3 methylation analysis. Women having positive POU4F3 methylation are referred for colposcopy.
Because POU4F3 methylation analysis did not miss any invasive cancer, POU4F3 methylation-negative, hrHPVpositive women may repeat HPV assay and DNA methylation analysis 1 year later. This strategy may substantially reduce the referral rate. However, a longitudinal follow-up study is needed to clarify the natural history of those  infected with hrHPV, but without POU4F3 hypermethylation to determine if a longer interval between screenings is also safe. The high negative predictive value in hrHPVnegative women is well documented, which means a longer screening interval is safe. However, determining POU4F3 methylation in HPV-negative women to assess POU4F3 as independent from HPV as a marker for cervical neoplasia/ CIN/CIN3/cancer may also be a consideration. Because HPV assay and methylation analysis can be conducted in the same self-collected cervical sample, the application of this approach may improve the participation of women for screening [39,40], especially those in low-resource areas. The application of a DNA methylation analysis using self-collected vaginal samples warrants further evaluation. In addition, this is a retrospective hospital-based study, which did not follow up the participants. Populationbased studies in different geographical and ethnic backgrounds are needed to validate these results.
In the present study, we adapted histopathologically diagnosed CIN3 + as the end point because CIN2 is equivocal in nature with a tendency to regress to normal instead of progressing to CIN3 + , where the likelihood of CIN2 progression to invasive cancer is only 5 % [41]. Further, diagnosis of CIN2 is much less reproducible than CIN3 because of the difference in the natural history of CIN2 from that of CIN3 [42]. However, CIN3 has a higher tendency to progress to invasive cancer because it is an immediate precursor with a similar virological profile and has better reproducibility [31]. Therefore, it is more appropriate to adapt CIN3 + as a surrogate end point for early diagnosis of cervical cancer.

Conclusions
POU4F3 methylation testing is a potential molecular biomarker for the triage of hrHPV-positive women for CIN3 + lesions. We envision an era of molecular screening for cervical cancer.

Patients
We conducted a retrospective case-control study using hospital-based patient samples in the Tri-Service General Hospital, Taiwan, from December 2009 to November 2010. Patients aged ≥20 years referred for a colposcopy and cervical biopsy and who were managed with conization or surgery after biopsy revealing CIN3 + were enrolled in this study. Cervical scrapings for laboratory analysis were collected in sterile phosphate-buffered saline before biopsy using a cervical brush and were stored at 4°C until DNA extraction for HPV testing using a HCII hrHPV DNA assay (Digene, Silver Spring, MD, USA) and quantitative DNA methylation analysis of potential candidate genes using QMSP. Healthy women undergoing routine Pap screening were selected as controls, only when their Pap smears showed normal pattern. Women with positive  Fig. 3 Proposed cervical cancer screening strategy using hrHPV assay and POU4F3 methylation analysis as a triage test. In this proposed scenario, HCII hrHPV DNA assay is used as the primary screening test, where women without hrHPV infection undergo follow-up 3 to 5 years later. Samples from women with hrHPV infection undergo POU4F3 methylation analysis, where women having positive POU4F3 methylation are referred for colposcopy. Additionally, women with a positive hrHPV assay but negative POU4F3 methylation may repeat HPV assay and DNA methylation analysis 1 year later or suspicious Pap smears were excluded from control. Before the study, all the subjects were informed about the study and were enrolled after obtaining documented full consent. Final diagnosis regarding different stages of cancer was performed by tissue-proven histopathological examination, except in healthy control women. Exclusion criteria applied in this study were compromised quality of Pap smears, patients previously vaccinated with anti-HPV vaccine, cervical neoplasia or existence of other malignancies, surgery related to the uterine cervix, an immunocompromised state, genital warts, or pregnancy. Further, all specimens were delinked from clinical information after numbering each of them until data analysis. In this study, all the women were tested for HPV infection and only samples from hrHPV-positive women underwent DNA methylation analysis. Cervical scrapings of 67 hrHPV-positive women among 100 recruited women underwent DNA methylation analysis to prioritize candidate genes for further analysis of clinical performance. Cervical scrapings from 200 women recruited for analysis of clinical performance were randomly classified using a random number table as a 1:1 ratio into a training set (n = 111) and a testing set (n = 89). The training set included 46 women with histopathologically confirmed CIN3 + and 65 women with normal/CIN1. Cervical scrapings from 68 hrHPV-positive women of the 111 underwent DNA methylation analysis. Methylation levels in the training set were used to generate optimal M-index cutoff values of candidate genes that can distinguish relevant cancerous cases from control. The clinical performance of candidate genes was validated using the optimal cutoff values in the testing set. The testing set comprised 89 women including 39 women with CIN3 + and 50 women with normal/CIN1. Of the 89 women, cervical scrapings from 55 hrHPV-positive women were analyzed further for quantitative assay of DNA methylation. This study was approved by the Institutional Review Board of the Tri-Service General Hospital, National Defense Medical Center.

Extraction of DNA followed by bisulfite modification
Genomic DNA was extracted as previously described using a DNeasy Blood & Tissue Kit (Qiagen, Hilden, Germany) according to the manufacturer's recommendations [28]. Those samples with a DNA yield as low as 500 ng or more (>500 ng) as measured by NanoDrop ND-1000 (Thermo Scientific, Wilmington, DE, USA) were considered for further analysis in this study. Bisulfite modification of genomic DNA samples was performed using a CpGenome DNA Modification Kit (Millipore, Temecula, CA, USA) according to the manufacturer's recommendations, and the samples were dissolved in 70 μL of nuclease-free water [29]. Bisulfite-converted DNA was stored at −80°C until further use.  Table S1). However, for analysis of clinical performance of candidate genes, QMSP for AJAP1, HS3ST2, and POU4F3 and multiplex QMSP for PAX1 and SOX1 were performed in a TaqMan probe system using a LightCycler 480 Real-Time PCR System (Roche Diagnostics, Roche Applied Science) [29]. Briefly, the total reaction volume of 20 μL contained 2 μL of modified template DNA, 1 μL of 20× Custom TaqMan reagent, and 10 μL LightCycler 480 Probes Master (Roche Diagnostics, Roche Applied Science). A mixture of primers and probes was used for PAX1 and SOX1. The reactions were conducted using an initial incubation at 95°C for 10 min, followed by 50 cycles of 95°C for 10 s, and annealing and extension for 40 s at 60°C (using the thermal cycler protocol in the standard mode). The level of DNA methylation was measured in terms of M-index [30]. Results showing the very high Cp values of COL2A (>36) were defined as detection failures.

Statistical analysis
ROC curves for each of the candidate genes were calculated using the data from the training set. Optimal M-index cutoff values of the candidate genes were generated from ROC curves and were used to further analyze the clinical performance of the candidate methylated genes in the testing set. Sensitivities, specificities, positive predictive values (PPV), and negative predictive values (NPV) of AJAP1, HS3ST2, POU4F3, PAX1, and SOX1 for detecting CIN3 + were calculated. IBM SPSS Statistics for Windows, version 20.0 (Armonk, NY, USA) was used for all statistical analyses.