Identification and validation of CNTNAP2/PAX1 hypermethylation as an epigenetic biomarker panel for detection of cervical cancer and precancerous lesions in hrHPV-positive women

Abstract

Background

Cervical cancer remains a significant global health burden, and effective triage strategies are urgently needed for high-risk human papillomavirus (hrHPV)-positive women in screening programs. This study aimed to identify and validate DNA methylation markers for detecting cervical precancerous lesions in hrHPV-positive women.

Methods

Reduced representation bisulfite sequencing (RRBS) was performed on a discovery cohort (n = 17) to screen for differentially methylated candidate genes. Quantitative methylation-specific PCR (qMSP) was used to evaluate the performance of selected markers in an independent cohort of 347 hrHPV-positive women stratified into training (n = 154) and validation (n = 193) sets. Diagnostic accuracy was assessed using receiver operating characteristic (ROC) curves, sensitivity, specificity, and area under the curve (AUC).

Results

Among the candidates identified, CNTNAP2 and PAX1 methylation levels increased progressively with cervical disease severity (both P < 0.0001). In the independent validation set, the combined CNTNAP2/PAX1 panel achieved an AUC of 0.904 (95% CI: 0.853–0.942) for CIN2 + detection and 0.932 (95% CI: 0.887–0.963) for CIN3 + detection. The panel yielded a sensitivity of 73.8% and specificity of 95.5% for CIN2+, and a sensitivity of 89.2% and specificity of 87.8% for CIN3+. In comparison, HPV16/18 genotyping showed sensitivities of 54.1% for CIN2 + and 56.8% for CIN3+, with specificities of 81.1% and 76.3%, respectively. Cytology (≥ ASCUS) had higher sensitivity (CIN2+: 93.1%; CIN3+: 91.2%) but substantially lower specificity (CIN2+: 16.9%; CIN3+: 14.9%); notably, cytology specificity in this cohort was likely underestimated due to verification bias. The CNTNAP2/PAX1 panel outperformed both comparator methods in overall diagnostic accuracy.

Conclusions

The CNTNAP2/PAX1 methylation panel demonstrates robust diagnostic performance for detecting cervical precancerous lesions in hrHPV-positive women, supporting its potential utility as an objective and effective triage tool in hrHPV-based cervical cancer screening.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-16208-7.

Background

Cervical cancer (CC) represents the fourth most prevalent malignancy among women globally. Furthermore, it constitutes the second most common cancer diagnosis and the third leading cause of cancer-related mortality among women in developing countries, accounting for approximately 90% of global CC deaths [1]. As a nation bearing one of the world’s highest CC burdens, China exhibits persistently rising incidence and mortality rates [2]. Critically, the age-standardized incidence rate in China substantially exceeds the World Health Organization (WHO) threshold of 4 cases per 100,000 women-years [3]. Consequently, early screening remains a critical preventive strategy against this disease.

The protracted progression from precancerous lesions to invasive CC, typically exceeding a decade, creates a substantial window for early diagnosis and intervention. Consequently, multiple diagnostic methods are employed to differentiate normal cervical tissue, cervical intraepithelial neoplasia (CIN), and invasive carcinoma, or to identify carcinogenic risk factors. Human papillomavirus (HPV) infection is a well-established etiological factor for CC. The high-risk HPV (hrHPV) test, a cornerstone of CC screening, offers high sensitivity for detecting cervical abnormalities. However, its clinical utility is limited by lower specificity compared to cytology, primarily due to the transient nature of many HPV infections, which are often cleared spontaneously by the host immune system [4, 5]. Although cytological techniques can effectively screen for CIN, their widespread implementation in developing countries remains challenging. This limitation stems from the technique’s inherent reliance on the expertise and subjective interpretation of cytopathologists, necessitating sophisticated infrastructure and trained personnel [6]. Emerging evidence suggests that DNA methylation testing offers a promising molecular triage strategy following primary HPV screening [7, 8]. This approach not only demonstrates high performance and potential for automation but also reduces unnecessary colposcopy referrals, thereby minimizing overtreatment and associated patient anxiety. These attributes make methylation testing particularly suitable for resource-limited settings where complex infrastructure is unavailable [9].

Current methodologies for detecting DNA methylation include targeted next-generation bisulfite sequencing, microarray analysis, and quantitative methylation-specific PCR (qMSP) [10–12]. However, whole-genome bisulfite sequencing (WGBS) and microarrays are cost-prohibitive for large-scale studies, whereas qMSP is inherently limited to interrogating predefined gene targets. In contrast, reduced representation bisulfite sequencing (RRBS) offers a cost-effective and efficient alternative for genome-wide methylation profiling. RRBS employs restriction enzyme digestion to enrich for CpG-dense genomic regions, including CpG islands, promoter regions, and enhancer elements. This targeted enrichment significantly reduces sequencing costs and volume compared to WGBS while simultaneously achieving higher-resolution methylation data within these functionally critical areas [13]. Compared to genome-wide methylation sequencing, the sequencing volume will be greatly reduced and higher precision resolution will be achieved in the CpG islands, promoter regions, and enhancer element regions. Consequently, RRBS is particularly well-suited for large-scale clinical studies, enabling cost-effective comparative epigenomic analysis across diverse cohorts.

This study aimed to develop and validate a DNA methylation-based biomarker panel for triaging hrHPV-positive women in cervical cancer screening. Using a discovery-to-validation framework, we first screened candidate methylation markers via RRBS in a small discovery cohort, then rigorously evaluated the diagnostic performance of the selected markers in independent training and validation cohorts comprising hrHPV-positive women with histologically confirmed outcomes.

Methods

Sample collection

To screen for differentially methylated genes, we utilized a cohort of 17 cervical scraping specimens obtained from women with definite histopathological results. Specimens were collected in PreservCyt Solution (ACS01; Liferiver, Shanghai, China) and stored at -80 °C. All specimens were distributed as follows: cervical cancer (n = 5), CIN1 (n = 3), CIN2 (n = 3), CIN3 (n = 3), and controls (cervicitis, n = 3). Detailed information is shown in Supplementary Table S1.

To evaluate the performance of candidate methylation markers, we enrolled a cohort of 1,439 hrHPV-positive women registered at the Third Affiliated Hospital of Shenzhen University between July 2023 and January 2025. Residual cervical scrapings were collected following hrHPV testing. Within this cohort, 700 women had definitive histopathological diagnoses: control (normal and cervicitis, n = 293), CIN1 (n = 231), CIN2 (n = 72), CIN3 (n = 54), CC (n = 37; comprising squamous cell carcinoma [SCC] and adenocarcinoma [AC]), and vaginal/endometrial lesions (n = 13). For methylation analysis, we included all CIN2 or worse (CIN2+) cases (n = 163) and randomly selected subsets of control (n = 100) and CIN1 (n = 138) samples. After excluding samples with insufficient residual material (n = 31), failed methylation assays (n = 21), or non-cervical malignancies (breast cancer, n = 1; thyroid cancer, n = 1), a final set of 347 hrHPV-positive samples underwent methylation analysis. Samples were stratified into training and validation sets based on collection date. All specimens were stored at -80 °C prior to processing. The study flowchart is detailed in Fig. 1B.

Fig. 1.

Study protocol. A Flowchart of candidate gene screening. B Flow diagram of validation. Control group comprises hrHPV-positive women with histologically confirmed normal or cervicitis. RRBS, reduced representation bisulfite sequencing; hrHPV, high-risk human papillomavirus; CIN, cervical intraepithelial neoplasia; SCC, squamous cell carcinoma; AC, adenocarcinoma

Cytological results were retrieved from medical records and classified according to the Bethesda System: negative for intraepithelial lesion or malignancy (NILM), atypical squamous cells of undetermined significance (ASCUS), low-grade squamous intraepithelial lesion (LSIL), atypical squamous cells-cannot exclude high-grade squamous intraepithelial lesion (ASC-H), atypical glandular cells (AGC), high-grade squamous intraepithelial lesion (HSIL), and SCC. All samples underwent independent review by two experienced pathologists, with discordant cases resolved by consensus.

The study protocol received approval from the Institutional Review Board (IRB) of the Medical Laboratory, Third Affiliated Hospital of Shenzhen University. A waiver of informed consent was granted for the use of residual, de-identified clinical specimens.

RRBS assay

Genomic DNA was isolated from cervical scrapings using the QIAamp Fast DNA Tissue Kit (Qiagen, Düsseldorf, Germany) according to the manufacturer’s protocol. DNA concentration and purity were assessed spectrophotometrically by measuring A260/A280 ratios. Samples exhibiting ratios between 1.8 and 2.0 were deemed suitable for downstream analysis. Qualified DNA samples were enzymatically digested with MspI (New England Biolabs, Ipswich, MA, USA) for reduced representation bisulfite sequencing (RRBS). Subsequently, digested DNA underwent bisulfite conversion using the EZ DNA Methylation-Gold™ Kit (Zymo Research, Irvine, CA, USA). Libraries were prepared from bisulfite-converted DNA using the Accel-NGS Methyl-Seq DNA Library Kit (Swift Biosciences, Ann Arbor, MI, USA), which ligates adapters to single-stranded DNA fragments. Finally, paired-end sequencing (2 × 150 bp) was performed on an Illumina HiSeq 4000 platform (LC Sciences, Houston, TX, USA) by LC-BIO Co., Ltd (HangZhou, China).

HrHPV DNA assay

HrHPV infection status was determined using the hrHPV Genotyping Real-Time PCR Kit (Liferiver, Shanghai, China) according to the manufacturer’s instructions. This assay detects 15 hrHPV genotypes: HPV16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, 66, 68, and 82.

Methylation test

One milliliter of PreservCyt-preserved sample was centrifuged at 12,000 × g for 2 min. The resulting pellet was resuspended in 300 µL of phosphate-buffered saline (PBS). Genomic DNA was isolated using the Nucleic Acid Extraction Kit (Yilifang, Shenzhen, China) according to the manufacturer’s protocol. DNA concentration was measured using a Thermo Scientific NanoDrop One spectrophotometer. For bisulfite conversion, 300 ng of DNA was processed using the EZ DNA Methylation-Lightning Kit (Zymo Research, Irvine, CA, USA), adhering to the manufacturer’s specifications. This reaction deaminates unmethylated cytosines to uracils. Bisulfite-converted DNA (1 µL, representing approximately 15 ng of initial DNA input) was analyzed by qMSP using the MethyLight qPCR Mix (GenStar, Beijing, China) to assess methylation levels in candidate genes, including PAX1 – a previously validated biomarker [14]. Primer and probe sequences are detailed in Supplementary Table S2. ACTB (β-actin) [7] served as the internal reference gene for normalization and quality control. Samples exhibiting a crossing point (Cp) value > 36 for ACTB were excluded from analysis, ensuring adequate DNA quality and conversion efficiency.

Relative methylation levels were calculated using the ΔCp method: ΔCp = Cp(target gene) - Cp(ACTB). Lower ΔCp values indicate higher methylation level in the target region.

Statistical analysis

Differentially methylated regions (DMRs) were identified using the R package MethylKit (version 3.2.3) with standard sliding window parameters (1000 bp windows, 500 bp overlap, P-value < 0.05). Genome-wide methylation profiles were subsequently generated for all samples within each diagnostic group. Sequencing depth across samples ranged from approximately 35 to 90 million high-quality reads, ensuring robust coverage for methylation calling. The valid data was obtained by removing the reads that containing adaptor, the proportion of it containing N (which means that base information cannot be determined) was greater than 5% and low-quality (base numbers with mass value Q ≤ 10 account for more than 20% of the entire read). After quality filtering, the alignment rates ranged from 52.1% to 74.5%, with an average of 65.6% of reads successfully aligned to the reference genomes using Bismark (version 0.22.1) (Supplementary Table S3).

To comprehensively characterize the epigenetic landscape, we conducted three integrated analyses: (1) Chromosomal distribution of CpG methylation was visualized using circos plots depicting region-specific methylation patterns; (2) Functional element analysis examined methylation alterations in regulatory regions including promoters, enhancers, and CpG islands; (3) The average methylation level of the whole genome range of each group was calculated and plotted as a violin plot. Differential methylation between cancer and control groups was further analyzed through volcano plots of DMRs. The above content was accomplished using the R package (version 3.2.3).

Differences in DNA methylation levels between normal and in order of increasing severity of disease (CIN1, CIN2, CIN3, and CC), were visualized using box plots. The Kruskal-Wallis test was used for comparisons among multiple groups, and Wilcoxon tests were performed for pairwise comparisons between each group and the control group. Receiver operating characteristic (ROC) curves were generated to evaluate the diagnostic accuracy of candidate genes in both training and validation sets. Area under the curve (AUC) values with 95% confidence intervals (95% CI) were calculated using DeLong’s method. The optimal cutoff value (ΔCp) was generated in training set using the Youden index and evaluated in the validation set. Sensitivities, specificities, positive predictive values (PPV), and negative predictive values (NPV) of CNTNAP2/PAX1, HPV16/18 genotyping, cytology test were calculated to evaluate their clinical performance for detecting CIN2 + and CIN3 or worse (CIN3+). To estimate the real-world clinical utility of the CNTNAP2/PAX1 panel in a general hrHPV-positive screening population, we calculated adjusted PPV and NPV using the sensitivity and specificity from our independent validation set and the true disease prevalence observed in the original consecutive cohort of 1,439 hrHPV-positive women. The adjusted PPV and NPV were calculated using the following formula:

Confidence intervals for sensitivity and specificity were calculated using the Wilson method via an online calculator (https://metricgate.com/). Confidence intervals for the predictive values were the standard logit confidence intervals given by Mercaldo et al. 2007 [15]; except when the predictive value was 0% or 100%, in which case a Clopper-Pearson confidence interval was reported. All other statistical analyses were performed using MedCalc (version 23.3.3), and graphs were generated using GraphPad Prism (version 10.4.1). Reported P-values are two-sided, and P < 0.05 was used as the significance threshold.

Results

DNA methylation profiles of RRBS

Exploratory RRBS analysis revealed generally high CpG methylation levels across all chromosomes and disease groups (Supplementary Figure S1). Methylation was most pronounced in promoter regions, first exons, and last exons (Supplementary Figure S2A). Genome-wide methylation levels were high across all groups, with the cancer group showing modestly lower overall methylation (Supplementary Figure S2B).

Differentially Methylated Regions (DMRs) analysis

In the cancer versus control comparison, hypermethylated DMRs predominated in promoters, exons, and CpG islands, whereas hypomethylated DMRs were more frequent in intronic and intergenic regions (Supplementary Figure S3A). The association between fold change of methylation level and P-value of DMRs was analyzed. There are more hyper-methylated DMRs than hypo-methylated DMRs in cancer vs. control, and the hypermethylated DMRs have more significant hits, as shown in Supplementary Figure S3B.

Differentially methylated gene identification

Functional enrichment analysis suggested that DMR-associated genes were involved in developmental processes, transcriptional regulation, and cancer-related pathways, including Wnt and Hippo signaling (Fig. 2). Given the limited sample size of the discovery cohort (n = 17), these pathway-level observations should be interpreted as hypothesis-generating rather than definitive.

Fig. 2.

GO Enrichment and KEGG Enrichment of Cancer vs. Control group. A The percentage of genes per GO Term. B The results of GO enrichment analysis which were displayed in scatter graphics. C Statistics were made according to the percentage of genes enriched on the pathway, and different KEGG main classes were marked with different colors. D The results of KEGG enrichment analysis which were displayed in scatter graphics. Rich Factor (ratio of the number of different genes to the total; the greater the Rich Factor value, the greater enrichment degree). GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes

To prioritize candidate genes, we ranked all DMRs by methylation fold-change (cancer vs. control) and selected the top 25 DMR-associated genes for downstream analysis. Selection criteria required an absolute FDR-adjusted p < 0.05 (Supplementary Table S4). The methylation patterns of these candidates were evaluated in the TCGA (accessed through UALCAN), where 10/25 genes exhibited statistically significant alterations (Supplementary Figure S4). Next, we analyzed the methylation patterns of these 10 genes across different disease stages (control, CIN1, CIN2, CIN3, and cancer), and identified that the DMRs of 6 of the 10 genes showed an increasing trend in mean methylation during cervical carcinogenesis (Supplementary Figure S5). From this analysis, we identified six novel genes (BICC1, CNTNAP2, PABPC4L, KCNK17, KCNK16, TCF4) with potential as biomarkers for cervical cancer detection. The detailed work flow chart is shown in Fig. 1A. Among the six genes, CNTNAP2 and PABPC4L combined smaller P-values in the TCGA database with more pronounced stepwise differences in DMR methylation levels across groups. Notably, this selection was based solely on the RRBS discovery findings and TCGA validation, and was completed prior to the training/validation cohort split, thereby avoiding feature selection bias. We prioritized CNTNAP2, a gene previously shown to be hypermethylated in other cancers [16, 17], and PABPC4L, a gene with no prior cancer-associated reports, for qMSP testing. Although we initially anticipated that PABPC4L might represent a major discovery, subsequent qMSP testing in the training set did not yield satisfactory discriminatory performance, and thus it was not evaluated in the validation set.

The characteristics of the samples

A total of 347 hrHPV-positive women (median age: 39 years; IQR: 31–48 years) were included in the methylation analysis. Clinical characteristics are summarized in Table 1. Among the participants, 118 (34.0%) tested positive for HPV16/18, and 293 (84.4%) had abnormal cytology results (≥ ASCUS), with 7 cases lacking cytological data. Histopathological analysis revealed the following distribution: 90 (25.9%) cervicitis or normal (classified as control), 116 (33.4%) CIN1, 61 (17.6%) CIN2, 45 (13.0%) CIN3, 30 (8.6%) SCC, and 5 (1.4%) AC.

Table 1.

Characteristics of participants

Characteristic	Training set (n = 154)	Validation set (n = 193)	All (n = 347)
Age
Median (IQR)	41 (32–50)	38 (30–46)	39 (31–48)
hrHPV detection, n(%)
HPV16/18	60 (39.0)	58 (30.1)	118 (34.0)
Non−16/18 HPV	94 (61.0)	135 (69.9)	229 (66.0)
Single HPV	110 (71.4)	135 (69.9)	245 (70.6)
Multiplex HPV	44 (28.6)	58 (30.1)	102 (29.4)
Cytology#, n(%)
NILM	21 (13.6)	26 (13.5)	47 (13.5)
ASCUS	95 (61.7)	112 (58.0)	207 (59.7)
LSIL	21 (13.6)	36 (18.7)	57 (16.4)
ASC-H	4 (2.6)	9 (4.7)	13 (3.7)
AGC	0 (0)	2 (1.0)	2 (0.6)
HSIL	9 (5.8)	3 (1.6)	12 (3.5)
SCC	2 (1.3)	0 (0)	2 (0.6)
Diagnosis, n(%)
Control	41 (26.6)	49 (25.4)	90 (25.9)
CIN1	33 (21.4)	83 (43.0)	116 (33.4)
CIN2	37 (24.0)	24 (12.4)	61 (17.6)
CIN3	24 (15.6)	21 (10.9)	45 (13.0)
SCC	16 (10.4)	14 (7.3)	30 (8.6)
AC	3 (1.9)	2 (1.0)	5 (1.4)

AGC atypical glandular cells, ASCUS atypical squamous cells of undetermined significance, ASC-H atypical squamous cells, cannot exclude high-grade squamous intraepithelial lesion, hrHPV high-risk human papillomavirus, HSIL high-grade squamous intraepithelial lesion, LSIL low-grade squamous intraepithelial lesion, NILM no intraepithelial lesion or malignancy

#There were 2 cases with no cytological results in the training set, 5 cases in validation set. Control group comprises hrHPV-positive women with histologically confirmed normal or cervicitis

Single HPV infection predominated in this study, accounting for 70.8% of cases (Fig. 3A). Among these singly infected women, HPV16 (25.3%), HPV52 (20.4%), and HPV58 (19.6%) exhibited the highest prevalence (Fig. 3B). Overall, the distribution of HPV genotypes was as follows (Fig. 3C): HPV16 (29.7%), HPV58 (25.1%), HPV52 (24.2%), HPV51 (10.4%), HPV33 (7.8%), HPV39 (6.9%), HPV56 (6.9%), HPV66 (6.6%), HPV18 (5.8%), HPV68 (5.2%), HPV31 (4.0%), HPV59 (3.2%), HPV35 (2.6%), HPV82 (2.6%), HPV45 (1.2%). ASCUS was the most frequent cytological diagnosis. Among these cases, 38.6% (80/207) progressed to CIN2+, 6.3% (13/207) to CC. Most normal or CIN1 (≤ CIN1) cases were found in NILM, ASCUS, and LSIL groups (Fig. 3D). Notably, 4 CC cases were missed by cytology (NILM) but detected due to hypermethylation of CNTNAP2 or PAX1.

Fig. 3.

The genotypes of hrHPV and classification of cytology in this study. A The proportion of single hrHPV infection and multiple hrHPV infection. B The proportion of different hrHPV genotypes. C The prevalence of hrHPV genotypes in participants. D The sample numbers of histological outcomes in each cytological classification. Purple, blue, green, orange, and red represent the control, CIN1, CIN2, CIN3, and cancer groups, respectively. Control group comprises hrHPV-positive women with histologically confirmed normal or cervicitis. Cancer contains squamous cell carcinoma and adenocarcinoma. AGC, atypical glandular cells; ASCUS, atypical squamous cells of undetermined significance; ASC-H, atypical squamous cells, cannot exclude high-grade squamous intraepithelial lesion; CIN, cervical intraepithelial neoplasia; HSIL, high-grade squamous intraepithelial lesion; LSIL, low-grade squamous intraepithelial lesion; NILM, no intraepithelial lesion or malignancy

The clinical performance of markers

We evaluated the methylation levels of CNTNAP2, PABPC4L, and PAX1 in cervical lesions using qMSP in hrHPV-positive women. The methylation levels of both CNTNAP2 (Fig. 4A) and PAX1 (Fig. 4B) increased with the severity of the lesion, and the differences between groups were statistically significant (P < 0.0001). CIN2, CIN3, and CC exhibited significantly higher methylation than control group (P < 0.0001). In contrast, PABPC4L did not show notable variation (data not shown). In training set, the performance for CIN2 + detection: CNTNAP2 (AUC = 0.862, 95% CI = 0.797–0.912), PAX1 (AUC = 0.826, 95% CI = 0.757–0.882), PABPC4L (AUC = 0.548, 95% CI = 0.466–0.629) (Fig. 4C); for CIN3 + detection: CNTNAP2 (AUC = 0.902, 95% CI = 0.844–0.944), PAX1 (AUC = 0.903, 95% CI = 0.845–0.945), PABPC4L (AUC = 0.610, 95% CI = 0.529–0.688) (Fig. 4D). While CNTNAP2 and PAX1 demonstrated high diagnostic accuracy, PABPC4L performed poorly. Thus, only CNTNAP2 and PAX1 were selected for further evaluation. Validation set, the performance for CIN2 + detection: CNTNAP2 (AUC = 0.841, 95% CI = 0.782–0.890), PAX1 (AUC = 0.902, 95% CI = 0.851–0.940); for CIN3+: CNTNAP2 (AUC = 0.874, 95% CI = 0.819–0.917), PAX1 (AUC = 0.931, 95% CI = 0.885–0.962) (Fig. 4E and F). The consistent high performance in both training and validation sets highlights the robustness of CNTNAP2 and PAX1 methylation in CIN2+/CIN3 + detection. When analyzed together, the methylation of CNTNAP2/PAX1 (CNTNAP2/PAX1^m) panel achieved: AUC = 0.904 (95% CI = 0.853–0.942) for CIN2+, AUC = 0.932 (95% CI = 0.887–0.963) for CIN3 + in validation sets (Fig. 4G).

Fig. 4.

The methylation and performance of candidate genes among hrHPV-positive women. Correlation between DNA methylation levels of CNTNAP2 (A) and PAX1(B) (relative to the reference gene ACTB) and lesion grades in hrHPV-positive cervical scrapes are shown (n = 347). Box plots show medians with lower and upper quartile and range whiskers. Purple, blue, green, orange, and red dots represent the DNA methylation levels of individual cases in the control, CIN1, CIN2, CIN3, and cancer groups, respectively. Control group comprises hrHPV-positive women with histologically confirmed normal or cervicitis. Performance of CNTNAP2, PABPC4L, PAX1 for CIN2+ (C) and CIN3+ (D) detection in training set. E - G Performance of CNTNAP2, PAX1 and marker panel CNTNAP2/PAX1 in validation set for CIN2 + and CIN3 + detection. The AUC of ROC curves was calculated. Green, red, blue, light purple, and pink lines represent CNTNAP2, PAX1, PABPC4L, CIN2+, and CIN3+, respectively. AUC, area under the curve; CIN, cervical intraepithelial neoplasia; ROC, receiver-operating characteristic; ns, not significant. ****P < 0.0001 (Wilcoxon tests)

To assess the clinical utility of CNTNAP2/PAX1^m in cervical cancer screening, we compared its performance to HPV16/18 genotyping and cytology. In the validation, while CNTNAP2/PAX1^m exhibited lower sensitivity than cytology (73.8% vs. 93.1%) in distinguishing CIN2 + from ≤CIN1 cases, its specificity was significantly higher (95.5% vs. 16.9%). It is important to note that cytology specificity in this cohort is likely underestimated due to verification bias, as women with normal cytology are less likely to undergo confirmatory colposcopy/biopsy per screening guidelines. Both CNTNAP2/PAX1^m metrics were superior to those of HPV16/18 (sensitivity: 54.1%; specificity: 81.1%), as shown in Table 2. For detecting CIN2+, the positive predictive value (PPV) and negative predictive value (NPV) of CNTNAP2/PAX1^m (88.2% and 88.7%, respectively) were superior to those of HPV16/18 (56.9% and 79.3%) and cytology (33.3% and 84.6%). For the detection of CIN3+, both CNTNAP2/PAX1^m and HPV16/18 exhibited higher sensitivity and NPV, accompanied by lower specificity and PPV, compared to their performance for CIN2 + detection. These trends were similarly observed in the training set. Overall, CNTNAP2/PAX1^m demonstrated superior sensitivity, specificity, PPV, and NPV compared to HPV16/18 and outperformed cytology in all metrics except sensitivity. Adjusted PPV and NPV values based on the prevalence of CIN2+/CIN3 + in the hrHPV-positive women in this study are detailed in Supplementary Table S5. For several triage methods, the adjusted PPV significantly decreased, while the adjusted NPV significantly increased.

Table 2.

Performance of HPV16/18, cytology, CNTNAP2/PAX1^m for the detection of CIN2 + or CIN3 + among 347 hrHPV-positive women

Characteristic	Sensitivity		Specificity		PPV [% (95% CI)]	NPV [% (95% CI)]
	n/N	% (95% CI)	n/N	% (95% CI)
detection of CIN2+
Training set (n = 154)
HPV16/18	42/80	52.5 (41.7–63.1)	56/74	75.7 (64.8–84.0)	70.0 (59.7–78.6)	59.6 (53.1–65.7)
Cytology≥ASCUS	71/78	91.0 (82.6–95.6)	14/74	18.9 (11.6–29.3)	54.2 (51.0−57.4)	66.7 (46.1–82.4)
CNTNAP2/PAX1 m	63/80	78.8 (68.6–86.3)	64/74	86.5 (76.9–92.5)	86.3 (77.8–91.9)	79.0 (71.0−85.3)
Validation set (n = 193)
HPV16/18	33/61	54.1 (41.7–66.0)	107/132	81.1 (73.5–86.8)	56.9 (46.4–66.8)	79.3 (74.2–83.6)
Cytology≥ASCUS	54/58	93.1 (83.6–97.3)	22/130	16.9 (11.4–24.3)	33.3 (31.1–35.7)	84.6 (66.5–93.8)
CNTNAP2/PAX1 m	45/61	73.8 (61.6–83.2)	126/132	95.5 (90.4–97.9)	88.2 (77.2–94.3)	88.7 (83.8–92.3)
detection of CIN3+
Training set (n = 154)
HPV16/18	29/43	67.4 (52.5–79.5)	80/111	72.1 (63.1–79.6)	48.3 (39.4–57.4)	85.1 (78.5–89.9)
Cytology≥ASCUS	37/42	88.1 (75.0−94.8)	16/110	14.5 (9.2–22.3)	28.2 (25.6–31.1)	76.2 (55.6–89.1)
CNTNAP2/PAX1 m	39/43	90.7 (78.4–96.3)	89/111	80.2 (71.8–86.5)	63.9 (54.6–72.3)	95.7 (89.7–98.2)
Validation set (n = 193)
HPV16/18	21/37	56.8 (40.9–71.3)	119/156	76.3 (69.0−82.3)	36.2 (27.6–45.8)	88.1 (83.6–91.6)
Cytology≥ASCUS	31/34	91.2 (77.0–97.0)	23/154	14.9 (10.2–21.4)	19.1 (17.3–21.1)	88.5 (70.9–96.0)
CNTNAP2/PAX1 m	33/37	89.2 (75.3–95.7)	137/156	87.8 (81.8–92.1)	63.5 (52.9–72.9)	97.2 (93.1–98.9)

Caution: The PPV and NPV values shown above were calculated directly from the observed data in the training and validation cohorts, which were enriched for CIN2+/CIN3 + cases. These values are presented only for relative comparison between methods within this study. They should not be interpreted as estimates of real-world clinical performance. CNTNAP2/PAX1^m, CNTNAP2/PAX1 methylation, the combined DNA methylation level of the CNTNAP2 and PAX1 gene panel; 95% CI 95% confidence interval, ASCUS atypical squamous cells of undetermined significance, CIN cervical intraepithelial neoplasia, hrHPV high-risk human papillomavirus, NPV negative predictive values, PPV positive predictive values

Discussion

Several studies have employed RRBS to investigate DNA methylation patterns in various cancers, including gastric cancer [18], breast cancer [19], and non-small cell lung cancer [20]. However, only one previous study has comprehensively examined genome-wide DNA methylation in CC tissues compared to paracancerous tissues [21]. In the present study, we applied RRBS technology to analyze DNA methylation patterns in cervical scrapings, utilizing the R package MethylKit for data analysis, which is recognized as the preferred analytical tool for RRBS datasets [22]. Our findings demonstrate that DNA methylation sites were widely distributed across all chromosomes, with significant differences observed between cancerous and control samples. Consistent with previous findings by Yuan et al. [23], hypermethylation events significantly outnumbered hypomethylation events. Previous research has reported that most methylation-driven genes likely regulate the transcription and expression of oncogenes or tumor suppressor genes in CC [24]. Similarly, our analysis revealed that differentially methylated genes in CC were primarily enriched in biological processes such as cell development, cellular communication, transcriptional regulation, and signal transduction. Additionally, these genes were associated with multiple cancer types, suggesting potential shared regulatory mechanisms in tumorigenesis.

In the current study, we identified host gene DNA methylation markers associated with cervical precancer and cancer using RRBS and validated selected candidates in an independent cohort of hrHPV-positive specimens using qMSP. Through our discovery analysis, we identified six novel candidate gene markers—BICC1, CNTNAP2, PABPC4L, KCNK17, KCNK16, and TCF4—that exhibited significant hypermethylation in CC compared to control samples. No prior reports associate these genes methylation with cervical cancer based on our systematic review. Independent validation in hrHPV-positive women demonstrated that CNTNAP2 methylation levels progressively increased with lesion severity, distinguishing CIN2 + from ≤CIN1 with high accuracy in both training and validation cohorts. These findings suggest that CNTNAP2 holds promise as a diagnostic biomarker for cervical precancerous lesions. Although CNTNAP2 has not been previously implicated in CC, aberrant methylation of this gene has been documented in other malignancies, including pancreatic cancer [16, 25], malignant rhabdoid tumor [17]. CNTNAP2 was also related to oral squamous cell carcinoma [26], breast cancer [27] and oligodendrogliomas [28]. Functionally, CNTNAP2 plays a role in synaptic transmission [29] and interacts with transcription factor 21 [30], potentially contributing to tumor suppression [31] or cancer cell migration [32]. This cross-cancer pattern of epigenetic dysregulation aligns with the marked heterogeneity of epigenetic modifications observed across diverse tumor types in pan-cancer m⁶A methylome analyses [33], underscoring that specific genes may undergo convergent epigenetic silencing through mechanisms that vary by tissue context. For future investigation, the immune subtyping strategy employed to guide mRNA vaccine application in glioblastoma [34] suggests a potential translational direction: whether CNTNAP2 methylation status could similarly inform patient stratification for immunotherapeutic or vaccine-based approaches in cervical cancer.

Consistent with previous reports [14, 35], PAX1 methylation demonstrated high sensitivity and specificity for detecting CIN3 + lesions in our study population. In contrast, CNTNAP2 exhibited lower specificity than PAX1 but higher sensitivity for CIN2 + detection, suggesting its potential superiority as a screening marker for CIN2 + lesions. To optimize diagnostic performance, we combined CNTNAP2 and PAX1 into a methylation panel. Compared with other methylation markers evaluated in hrHPV-positive women, the CNTNAP2/PAX1^m panel demonstrated favorable performance. For CIN2+, POU4F3 methylation showed a sensitivity of 70.1% and specificity of 81.4% [36], while the bi-marker panel ASCL1/LHX8, validated in the Dutch IMPROVE screening cohort, achieved a sensitivity of 59.5% and specificity of 76.1% [37], whereas our panel achieved a sensitivity of 73.8% with higher specificity (95.5%). For CIN3+, the GynTect assay yielded a sensitivity of 67.4–73.2% and specificity of 76.0–84.8% [38, 39], POU4F3 reported 72–74% sensitivity and 77–89% specificity [36, 40], and ASCL1/LHX8 showed 76.9% sensitivity and 74.5% specificity [37]. In comparison, the CNTNAP2/PAX1^m panel achieved a sensitivity of 89.2% and specificity of 87.8% in our validation cohort. While cross-study comparisons are limited by population differences, these results support the potential of CNTNAP2/PAX1^m as an effective triage tool.

When extrapolated to a screening population using the true disease prevalence from our source cohort (CIN2+: 11.3%; CIN3+: 6.3%), the CNTNAP2/PAX1^m panel yielded adjusted PPVs of 67.4% for CIN2 + and 33.0% for CIN3+, substantially higher than those of HPV16/18 (26.7% and 13.9%, respectively) and cytology (12.5% and 6.7%, respectively). The adjusted NPV of 99.2% for CIN3 + indicates that a negative result provides strong reassurance against high-grade disease. Notably, our analysis identified 4 CC cases that were missed by cytology but detected by either CNTNAP2 or PAX1 methylation. This finding highlights the potential of methylation-based testing as a complementary approach to conventional cytology in cervical cancer screening.

This study has several limitations. First, our discovery phase utilizing RRBS was conducted on a relatively small cohort (n = 17). While this approach successfully identified promising candidate markers like CNTNAP2, the small sample size may limit the generalizability of the genome-wide methylation patterns and pathway enrichment analyses, might contribute to a high false-discovery rate. Second, although splitting our cohort by collection date into training and validation sets is a pragmatic approach to ensure temporal independence, it carries an inherent risk of introducing batch effects. To mitigate this, all samples were processed using identical standard operating procedures and reagent lots where possible. Furthermore, the validation set is relatively small, which may affect the generalizability of the findings. Future analysis of the sensitivity and specificity of the panel in larger, multi-center prospective cohorts will provide the most robust evidence for clinical utility. In addition, according to China’s cervical cancer screening guidelines, only non-16/18 hrHPV-positive women with cytology ≥ ASCUS are referred for colposcopy. Consequently, the number of cytology-negative women with available histopathology results was limited, potentially leading to an underestimation of cytology’s specificity (i.e., pseudo-low specificity). Additionally, while we confirmed the diagnostic potential of CNTNAP2 in detecting cervical lesions, the four remaining candidate genes (BICC1, KCNK17, KCNK16 and TCF4) remain unvalidated. Future studies should expand the cohort size and prospectively evaluate the utility of these genes in CC screening through large-scale observational or diagnostic validation studies. Finally, we acknowledge that the methylation signal in cervical scrapings originates from a heterogeneous cell mixture. Increased CNTNAP2 or PAX1 methylation may reflect true hypermethylation, an increased proportion of methylated cells, or both. Regardless, the association of the composite signal with disease severity remains robust and clinically actionable. Future single-cell studies may provide mechanistic clarity.

Conclusions

In summary, the CNTNAP2/PAX1 methylation panel demonstrated robust diagnostic performance for detecting cervical precancerous lesions among hrHPV-positive women, achieving high sensitivity and specificity for both CIN2 + and CIN3 + endpoints. The panel outperformed HPV16/18 genotyping and cytology across multiple metrics, suggesting its potential as an effective and objective triage tool. These findings support the utility of methylation-based biomarkers as a valuable adjunct in hrHPV-based cervical cancer screening.

Abbreviations

95% CI

95% confidence interval

AC

Adenocarcinoma

AGC

Atypical glandular cells

ASC-H

Atypical squamous cells cannot exclude high-grade squamous intraepithelial lesion

ASCUS

Atypical squamous cells of undetermined significance

AUC

Area under the curve

CC

Cervical cancer

CGI

CG island

CIN

Cervical intraepithelial neoplasia

Cp

Crossing point

CpG

Cytosine phosphate-guanine

DMRs

Differentially methylated regions

GO

Gene Ontology

hrHPV

High-risk human papillomavirus

HSIL

High-grade squamous intraepithelial lesion

KEGG

Kyoto Encyclopedia of Genes and Genomes

LSIL

Low-grade squamous intraepithelial lesion

NILM

No intraepithelial lesions or malignancy

NPV

Negative predictive values

PPV

Positive predictive values

qMSP

Quantitative methylation-specific PCR

ROC

Receiver-operating characteristic

RRBS

Reduced representation bisulfite sequencing

SCC

Squamous cell carcinoma

Authors’ contributions

YH and HY contributed equally to the design and writing of this manuscript and should be considered co-first authors; KK and RJ collected the samples; CZ detected the samples; YW and DL performed literature searches and were responsible for data visualization; DX carefully and rigorously revised the manuscript; all authors have read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (Grant No. 81772921), the Science and Technology Planning Project of Shenzhen Municipality, China (NO.JCYJ20240811001945001, NO.JCYJ20230807142806014), the Shenzhen Key Medical Discipline (No. SZXK054), Guangdong Provincial Clinical Research Center for Laboratory Medicine (No.2023B110008), and the International Collaborative Project of Shenzhen Science and Technology program (GJHZ20220913144213025).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki and was approved by the Research Ethics Committee of Shenzhen Luohu People's Hospital.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.