Integrative Analyses of Individual Patient Genomic-Data to Discover Novel Biomarkers: Application to Cervical Pre-Cancer DNA-Methylation Datasets

Abstract

Objective:

The growing availability of public genomic repositories led to increased use of integrative and meta-analytic approaches to combine multi-omics datasets for biomarker discovery. Building on this framework, our study applies a molecular discovery approach to identify diagnostic biomarkers for cervical pre-cancer in high-risk HPV-positive women using aggregated genomic evidence across populations.

Methods:

This study performed a secondary analysis of publicly available DNA-methylation datasets identified through a systematic search of the Gene Expression Omnibus database to enable integrative analysis. After employing standard preprocessing methods across the selected DNA-methylation datasets, statistical analyses (specifically limma and meta-analysis with a random-effects model) were used to calculate the cumulative standardized effect sizes for all 9570 genes across the 4 selected studies.

Results:

This cross-dataset analysis led to the identification of 4 promising, non-population-specific genes, that show potential for detecting precancerous cervical lesions in high-risk HPV-positive women. Collectively, these genes reflect a molecular profile associated with Cervical Intraepithelial Neoplasia lesions undergoing malignant progression.

Conclusions:

By leveraging aggregated data, our findings enhance the potential of generalization of biomarker signals and illustrate the strength of multi-data sets integration in advancing cervical pre-cancer triage screening strategies. Given their diagnostic promise, further experimental validation in large and diverse cohorts is warranted to evaluate their clinical applicability.

Introduction

With the increasing availability of public datasets in online repositories (for instance: Gene Expression Omnibus (GEO), ¹ ArrayExpress ² and The Cancer Genome Atlas (TCGA), ³ efforts to integrate and aggregate genomic data have gained attention over the past few decades. These efforts often leverage meta-analysis techniques to facilitate data integration and knowledge discovery, ⁴ particularly for identifying novel diagnostic and prognostic biomarkers. Existing approaches range from combining single-omics data types ⁵ to integrating multiple omics layers.^6,7 Moreover, recent studies have sought to couple these aggregative, meta-analytic strategies with predictive modeling frameworks to enhance biomarker discovery and disease classification performance. This study adds another perspective on how to integrate available data sets to release its full potential in finding the biomarker that is not population-specific.

To clarify the implementation framework of this study, we employed a molecular discovery approach aimed at identifying diagnostic biomarkers for cervical precancer in women positive for high-risk HPV, in which we believe the findings remain applicable for future implementation in this domain. Cervical cancer (CC) remains a major global public health challenge, despite the implementation of successful preventative strategies. ⁸ Nearly all cases of CC are caused by persistent infection with high-risk Human Papillomavirus (hrHPV). ⁹ Consequently, primary screening has increasingly shifted toward hrHPV DNA testing due to its high sensitivity for detecting cervical disease. ¹⁰ However, the majority of hrHPV infections are transient and will spontaneously clear. ¹¹ This biological reality means that current HPV testing, while highly sensitive, often lacks the specificity required to distinguish harmless, temporary infections from those lesions that are likely to progress to high-grade cervical intraepithelial neoplasia (CIN) two-thirds or invasive cancer. ¹² This diagnostic ambiguity leads to significant clinical issues, including patient anxiety, high rates of unnecessary colposcopy referral, and the potential for overtreatment of self-resolving lesions. ¹³ To improve the efficacy of cervical cancer prevention, there is a need for objective, molecular triage biomarkers that can accurately identify hrHPV positive women who are genuinely at risk for progressive disease, ¹⁴ and is not limited to a specific population.

Methods

The overall research workflow comprises 4 stages, namely: developing robust search strategy, extracting and preparing data sets, conducting statistical analyses, and validating the results by using Bioinformatics tools (Figure 1). The reporting of this study conforms to the STROBE ¹⁵ for observational studies and adheres to the principles of MIAME ¹⁶ for microarray data reporting, as recommended by the EQUATOR Network. The completed STROBE checklist is provided as a supplementary file (Supplemental Table S1).

Figure 1.

The overall research workflow.

Search Strategy

We systematically searched potential previously published studies in the Gene Expression Omnibus (GEO) database to identify relevant publicly available datasets pertaining to cervical cancer. The search was performed using “cervical cancer” as the primary keyword, followed by the application of predefined inclusion criteria.

• Organism : homo sapiens

• Study type : Expression profiling by array, Expression profiling by genome tiling array, Genome binding/occupancy profiling by array, Genome binding/occupancy profiling by genome tiling array, Genome variation profiling by array, Genome variation profiling by genome tiling array, Methylation profiling by SNP array, Methylation profiling by genome tiling array.

• Attribute : tissue, swab, exfoliated cervical cells, strain, lesion, biopsy, cervical smear.

From the GEO search results, we next systematically selected studies that employed DNA methylation assays. To reduce technical variability and ensure comparability across datasets, we further restricted our inclusion to studies conducted using the Illumina array platform. Illumina Infinium arrays (especially, HumanMethylation450K BeadChip and Methylation EPIC BeadChip) are among the most widely adopted technologies for genome-wide DNA methylation profiling, ¹⁷ offering high coverage across CpG sites, reproducibility, and standardized preprocessing pipelines. ¹⁸

Data Extraction and Preparation

A systematic search of the GEOs database was conducted from database inception to June 2025 to identify eligible DNA methylation datasets using predefined search terms. the inclusion criteria for further analyses were those that contained hrHPV–positive samples with clearly annotated lesion status and available raw methylation data generated using Illumina methylation array platforms. Given our objective was to identify global biomarkers capable of distinguishing hrHPV-positive women without lesions or with low-grade squamous intraepithelial lesions (LSIL: CIN0 and CIN1) from those with high-grade squamous intraepithelial lesions (HSIL: CIN2 and CIN3) or cervical cancer, we included only samples from the selected previously published studies that met these criteria. Cases were defined as hrHPV-positive samples with high-grade squamous intraepithelial lesions (HSIL) or cervical cancer, whereas controls consisted of hrHPV-positive samples without lesions or with low-grade lesions (CIN0–CIN1).

 Raw Illumina IDAT files, containing paired red and green channel intensities, were prioritized along with manifest files for probe annotation and metadata. Raw data were preferred to enable uniform preprocessing, background correction, normalization, and quality control, reducing technical variability across studies. When IDAT files were unavailable, preprocessed data (eg, CSV tables with β-values or intensities) were used to maximize dataset inclusion. While less flexible, these datasets still supported comparability for downstream meta-analysis.

Raw IDAT files were preprocessed and normalized using the minfi package ¹⁹ in R opensource software. ²⁰ Illumina preprocessing functions (ie, preprocess Illumina) were applied to generate β-values, representing the proportion of methylated signal at each CpG site. Annotation files from the corresponding Illumina platform (ie, HumanMethylation450 and Methylation EPIC BeadChip)^18,21 were used to map CpG sites to genomic features. Probes associated with promoter regions and CpG islands were retained, while unclassified probes were excluded. To stabilize variance and enable downstream statistical analyses, CpG-level β-values were then transformed into M-values using the formula.

$$M=lo{g}_2\left(\frac{\beta +ϵ}{1-\beta +ϵ}\right),$$

where $ϵ=1x{10}^{-6}$ prevents division by zero. The M-values were then aggregated to the gene level by averaging across all CpGs mapping to the same gene. All genes present across the final selected datasets were retained for further analyses.

Statistical Analyses

To identify differentially methylated genes, we restricted the dataset to genes common across all studies. The M-values were analyzed using the limma framework. ²² A linear model of the form

$$M_{ij}={\beta }_{0i}+{\beta }_{1i}{G}_j+{ϵ}_{ij},$$

was fitted for each gene $i$ , where $M_{ij}$ is the M-value for sample $j$ , $G_j$ is the group indicator (0 = control, 1 = case), ${\beta }_{0i}$ is the baseline methylation level for the control group, and ${\beta }_{1i}$ estimates the differential methylation between case and control groups. Empirical Bayes moderation was applied to stabilize variance estimate across genes. For each gene $i$ , the group-specific means were calculated as

$${\overline{x}}_{i,control}=\frac{1}{n_{control}}{\displaystyle \sum }_{j=1}^{n_{control}}{M}_{ij},{\overline{x}}_{i,case}=\frac{1}{n_{case}}{\displaystyle \sum }_{j=1}^{n_{case}}{M}_{ij}.$$

The unstandardized mean difference was calculated as

$${\Delta }_i={\overline{x}}_{i,case}-{\overline{x}}_{i,control}.$$

The standardized effect size (Hedge’s g) was ²³

$$g_i=J.\frac{{\Delta }_i}{S_i^2},$$

where $S_i^2$ is the moderated within-study variance estimated from limma model, and $J$ is the Hedge’s correction factor:

$$J=1-\frac{3}{4N-9},whereN=n_{case}+n_{control}.$$

We followed the approach of Novianti et al ²⁴ to aggregate effect sizes for each gene across datasets (see Section: Meta-analysis for gene selection), thereby strengthening the cumulative evidence for differential methylated-genes. In this study, we employed cumulative effect sizes to identify differentially methylated genes (DMGs), rather than the traditional fold-change methods typically used for this purpose. The underlying reasoning for this approach is elaborated in the Supplementary Material (Text 2).

Identification of Cervical Dysplasia–Associated Differentially Methylated Genes

From DMGs, cut-off criteria were set as FDR < 0.01. Genes showing significant cumulative effect sizes differentiating hrHPV-positive controls from HSIL case. To identify cervical dysplasia–relevant genes, an intersection analysis was performed between the list of DMGs and cervical dysplasia–associated genes retrieved from the Open Targets Platform. The overlap between these datasets was visualized and determined using Venn diagram analysis (http://bioinfogp.cnb.csic.es/tools/venny/), yielding a subset of genes considered as cervical dysplasia–associated DMGs were selected for further analysis of cervical dysplasia–associated DMGs. The Open Targets software enables us to analyze the pathways and diseases associated with these genes.^25,26 A focused filtering strategy was applied, specifically selecting only those GO terms and annotations directly correlated with cervical dysplasia (the foundational stage of carcinogenesis, encompassing transitions from normal epithelium through LSILto HSIL. This step ensured that downstream analysis was restricted to genes with documented functional relevance to the early stages of cervical pathology.^{27 -29}

Module Analysis of Protein-Protein Interaction (PPI) Network

The list of intersected genes between DMGs and cervical dysplasia–associated genes was subsequently used for protein–protein interaction (PPI) analysis. The PPI network was constructed using the STRING database (https://string-db.org/cgi/input.pl) to evaluate potential functional interactions and identify key hub genes within the network.

Pathway Analysis of Aberrantly Methylated DEGs

Genes derived from the PPI network, particularly those involved in key interactions (edges), were subsequently subjected to Gene Ontology (GO) functional enrichment analysis was performed on the selected genes derived from the protein–protein interaction (PPI) network was performed using the Enrichr platform (https://maayanlab.cloud/Enrichr/enrich). Statistically significant terms were identified based on an adjusted p-value (false discovery rate, FDR) < .05. The analysis encompassed the 3 principal GO categories: biological process (BP), cellular component (CC), and molecular function (MF). P-value < .05 was considered as statistically significant.

Classification of Oncogenes and Tumor Suppressor Genes

Genes derived from the PPI network, particularly those involved in key interactions (edges) was subsequently annotated and categorized into oncogenes and tumor suppressor genes (TSGs) using curated databases were obtained from the (http://ongene.bioinfo-minzhao.org/) and (https://bioinfo.uth.edu/TSGene/index.html) respectively. Through this integrative approach, upregulated hypomethylated oncogenes and downregulated hypermethylated TSGs on cervical dysplasia genes were identified.

Results

The overall systematic search strategy identified 4 studies (GEO accession ID: GSE99511, GSE287994, GSE186835, and GSE143752) that fulfilled all predefined inclusion criteria, comprising a total of 439 samples (171 controls and 268 cases). The basic characteristics of the 4 selected studies are summarized in Table 1 and are elaborated in Supplemental Table S1 and S2 (Supplementary Materials). An overview of the number of features retained at each stage of the integrative analysis is presented in Figure 2, with methodological codes (B.2c, B.sd, B.4, C.2a, D.1, D.2) corresponding to the workflow steps shown in Figure 1.

Table 1.

The Selected Studies’ Basic Information.

GEO ID	Year	Country	Array platform	Tissue type	Sample size
					Control	Case
GSE99511	2008	The Netherlands	HumanMethylation450	Self-sampled cervico- vaginal specimens	28	40
GSE143752	2020	Canada	Infinium MethylationEPIC	Cervical tissues	104	82
GSE186835	2021	USA	HumanMethylation450	Cervical tissues	10	18
GSE287994	2025	UK and Greece	Infinium MethylationEPIC	Exfoliated cervical cells	29	116

Note.

• Year: Year of publication of the corresponding study.

• Country: Geographic location from which the samples originated.

• Control: a group of HPV-positive samples without lesions, or with low-grade abnormalities (CIN0 or CIN1).

• Case: a group of HPV-positive samples with lesions, including high-grade abnormalities (CIN2 or CIN3) or cervical cancer.

• A detailed distribution of samples across the original disease states is provided in Supplemental Table S2.

Figure 2.

Summary of the number of features (probes/genes) retained at each stage of the integrative analysis. The codes shown along the arrows (ie, B.2c, B.sd, B.4, C.2a, D.1, D.2) correspond to the methodological steps described in Figure 1 and indicate the procedures used to derive the gene sets presented in the subsequent boxes.

Meta-analysis of gene-level effect sizes, adjusted with Benjamini-Hochberg correction, identified 443 significantly differentially methylated genes at FDR = 1% (Figure 3). Among these, 153 genes were hypomethylated, while 290 genes were hypermethylated. The full list is provided in the project repository (https://github.com/sienaclinical/INSIGHT-hrHPV), while summary statistics for the cumulative effect size for the DMGs are presented in Table 2. Low heterogeneity index (τ²) across the DMGs suggests that the selected datasets are sufficiently comparable for synthesis (Table 2). This consistency confirms that an integrative analysis via a meta-analytic approach is a methodologically sound and feasible strategy (Figure 4).

Figure 3.

Volcano plot of 443 differentially methylated genes (DMGs) derived from an integrative analysis of 4 studies. Red points represent high-effect DMGs defined by |θ| > 0.8 and adjusted p-value < .01. Gray points represent non-significant genes that did not meet these combined effect size and statistical thresholds (|θ| ⩽ 0.8 or adjusted p-value > .01).

Table 2.

Summary Statistics of Cumulative Effect Size and Heterogeneity Across the 443 Differentially Methylated Genes.

Random effect parameters	Mean	Median	Range	Standard deviation
Cumulative effect size (θ)	0.2065	0.4932	[−2.0952; 1.8617]	0.810
Between-study variances (τ2)	0.0037	0.0000	[0.0000; 0.5911]	0.036

Figure 4.

GO Functional Annotation: (a) biological process, (b) cellular component and (c) molecular function. The color represents signification, where the darkest color represents more significant genes. (Enrichr Platform: https://maayanlab.cloud/Enrichr/enrich).

For comparison, we also identified DMGs within each individual study. These analyses utilized a methodology identical to our integrative approach to ensure direct comparability between the single-study findings and the meta-analysis results. Across the 4 studies, 2 datasets shared a common set of DMGs, with 11 genes overlapping between them. (Supplemental Table S4)

Differentially Methylated Genes and its Association with Cervical Dysplasia

To further refine biologically relevant targets, an intersection analysis was performed between the identified DMGs and cervical dysplasia associated genes retrieved from the Open Targets Platform. Forty-five genes were identified, comprising 32 upregulated hypomethylated genes and 13 downregulated hypermethylated genes.

PPI Network Construction

STRING online data base was employed to construct PPI networks. Based on the cervical dysplasia–associated DMGs, a PPI network containing 45 nodes and 16 edges was constructed, with a PPI enrichment P-value of .00845.

Pathway Analysis of Aberrantly Methylated DEGs

Gene Ontology (GO) enrichment analysis was conducted across the 3 principal domains: biological process (BP), cellular component (CC), and molecular function (MF). The results indicated that, within the BP category (Figure 4a), negative regulation of programed cell death, regulation of apoptotic process, and negative regulation of apoptotic were highest significantly enriched among the differentially methylated genes (DMGs). In the CC category (Figure 4b), significant enrichment was observed in the Golgi stack, intracellular membrane-bounded organelle, and intracellular membraneless organelle. Furthermore, within the MF category (Figure 4c), DNA polymerase binding was identified as a significantly enriched function.

Identification of Oncogenes and Tumor Suppressor Genes

From a PPI network containing 45 nodes and 16 edges, further classification revealed 2 upregulated hypomethylated oncogenes, namely CSF1 and BCL2L1 (AUC = 72.32%) and 2 downregulated hypermethylated tumor suppressor genes, namely DNMT1 and WT1 (AUC = 31.45%) suggesting a potential epigenetic mechanism contributing to cervical dysplasia progression (Figure 5).

Figure 5.

Overlap of cervical dysplasia–associated genes across DMGs, STRING PPI network, and oncogene or TSG databases. (a) 2 hypomethylation upregulated oncogenes were identified and (b) 2 hypermethylation downregulated tumor suppressor genes were identified (diagram venn on https://bioinfogp.cnb.csic.es/tools/venny/).

Discussion

Although public datasets constitute a valuable resource, greater efforts are needed to optimize and to cumulatively integrate knowledge from those datasets for the discovery and creation of knowledge, which might cover its quality and reliability of data, as well as insufficient bioinformatics knowledge to mine the data.^30,31 Building on this need, this study presents a systematic framework that leverages public datasets to identify global biomarkers with potential diagnostic utility for cervical pre-cancer in hrHPV–positive women. While identifying relevant, previously published data is a significant challenge, systematic approaches enable the efficient extraction of insights. This cost-effective strategy provides an alternative to de novo data generation, especially in well-studied cancer diseases.

Our initial search strategy unintentionally yielded 4 studies conducted exclusively in developed countries. This selection likely reflects the high cost and infrastructure challenges that often limit such genomic profiling studies in developing nations. ³² To address this, our integrative framework provides a method for generalizing these results, offering potential application to low- and middle-income countries where discovery studies remain prohibitively costly. Importantly, cross-population validation is essential to distinguish universal biomarkers from those driven by population-specific genetic or environmental factors, thereby strengthening the robustness and translational potential of the identified candidates.

Our study focused on DNA methylation rather than RNA expression or other molecular assays because epigenetic alterations, particularly aberrant CpG methylation, occur early in cervical carcinogenesis and are stably preserved in clinical specimens such as cervical scrapes.^33,34 In contrast, RNA expression profiles are more dynamic, tissue-specific, and influenced by short-term environmental or physiological changes, which can reduce their robustness as diagnostic biomarkers. ³⁵ Moreover, methylation markers are more suitable for integration across multiple cohorts, as they are less sensitive to technical and biological variability than transcriptomic data.^36,37 By targeting DNA methylation, our approach aims to identify biomarkers that are not only biologically relevant but also technically reliable translation into clinical practice that is based on cross-populations sample sets. Although our study applies the integrative approaches for DNA methylation datasets in cervical cancer, we strongly believe the presented framework is readily extended to other omics data types and to other disease domains.

The identification of a robust molecular signature that distinguishes HSIL from transient hrHPV infection remains a critical frontier in cervical cancer prevention. ³⁸ While the introduction of hrHPV testing has significantly enhanced screening sensitivity, its limited specificity often leads to clinical over-management. ³⁹ and unnecessary psychological burden for women with infections that may never progress to malignancy. ⁴⁰ By integrating data across 4 independent GEO datasets, this study identifies a 45-gene differentially methylated signature that specifically characterizes the transition from hrHPV-positive status to advanced dysplasia. This “epigenetic switch” suggests that the progression of cervical pathology is not merely a consequence of viral presence, but is deeply rooted in the systematic remodeling of the host cell’s DNA methylation landscape.^41,42

The functional connectivity observed in the protein-protein interaction (PPI) network further reinforces the non-random nature of these molecular alterations. With a PPI enrichment p-value of .00845, the network demonstrates that the 45 identified genes are part of an interconnected biological module rather than isolated incidents of genomic noise. The identification of these hub genes provides specific targets for future diagnostic development. ⁴³

Functionally, the results of the Gene Ontology (GO) enrichment analysis show a complex strategic shift in cellular behavior, pivoting from homeostasis toward a state of heightened survival and resistance to physiological stress. The high statistical significance of terms within the Biological Process (BP) category, specifically those related to the negative regulation of programed cell death and the global regulation of apoptotic processes, suggests that the primary biological consequence of these methylation changes is to bypass the host cell’s innate “suicide” signals. ⁴⁴ In the specific context of hrHPV infection, where the integration of viral DNA typically triggers intense cellular stress and DNA damage responses, the epigenetic suppression of apoptosis acts as a critical survival mechanism.^45,46 By silencing pro-apoptotic signals or activating anti-apoptotic mediators through aberrant methylation, the cell ensures the continued existence and replication of a population harboring significant genomic instability. ⁴⁷ This evasion of apoptosis effectively provides a “safe harbor” for the virus to maintain its episomal or integrated presence while the host cell accumulates the secondary genetic hits necessary for full malignant transformation. ⁴⁸

The complexity of this transition is further highlighted by the enrichment observed in Cellular Component (CC) and Molecular Function (MF) domains. Significant enrichment in the Golgi stack and various intracellular membrane-bounded organelles suggests that the DMGs may also be involved in remodeling the cell’s secretory pathways and internal trafficking, potentially altering the tumor microenvironment or the presentation of surface antigens to the immune system.^{49 -51} Crucially, the identification of “DNA polymerase binding” as a significantly enriched molecular function implies that these epigenetic alterations are not merely passive markers of disease, but active participants in the dysregulation of the replication machinery. ⁵² By directly influencing the proteins that govern DNA synthesis and repair, these methylation changes facilitate the rapid, uncontrolled proliferative capacity that defines HSIL. ⁵³ This combination(where apoptosis is inhibited and the replication apparatus is accelerated)forms the functional cornerstone of cervical dysplasia, providing a clear molecular rationale for why these specific genes serve as robust biomarkers for clinical progression.^52,54

WT1 promoter methylation is strongly associated with cervical carcinogenesis, with hypermethylation increasing from normal tissue to cervical intraepithelial neoplasia (CIN) and peaking in cervical cancer, correlating with reduced gene expression and suggesting epigenetic silencing. ⁵⁵ This aligns with our findings of concordant changes in WT1 methylation and expression, and its association with HPV16/18 infection further supports a role for viral-driven epigenetic regulation. ⁵⁵ More broadly, aberrant DNA methylation is a key hallmark of cervical cancer and serves as a promising biomarker for identifying high-grade CIN and predicting progression risk.^41,56 Genome-wide studies reveal both hyper- and hypomethylation events affecting carcinogenic pathways during CIN progression. ⁵⁷ Integrating multiple methylation markers, including WT1, may therefore improve diagnostic and prognostic accuracy.^41,55,56 It should be noted that the relationship between DNA methylation and gene expression is not always strictly inverse, as gene regulation in cancer is influenced by additional mechanisms such as alternative promoter usage, copy number variation, and transcriptional regulation.

DNMT1 plays a central role in maintaining aberrant DNA methylation and tumor suppressor gene silencing in cervical carcinogenesis. Its expression increases from early CIN and correlates with disease severity.^58,59 HPV infection, particularly via E6-mediated p53 suppression, upregulates DNMT1, promoting epigenetic silencing and malignant transformation. ⁶⁰ DNMT1-driven hypermethylation also suppresses tumor-suppressive microRNAs (eg, miR-484) and genes such as UTF1, enhancing proliferation and metastasis.^61,62 Conversely, DNMT1 inhibition reduces promoter methylation, restores gene expression, and induces apoptosis, supporting its therapeutic potential. ⁶³ Folate deficiency may further exacerbate DNMT1 dysregulation and aberrant methylation. ⁶⁴

Macrophage colony-stimulating factor-1 (CSF-1) and its receptor CSF1R are key mediators in HPV-associated cervical disease progression. Elevated serum CSF-1 levels in hrHPV-positive women with cervical intraepithelial neoplasia (CIN) indicate early pathway activation. ⁶⁵ In invasive cancer, CSF1R and CSF-1 are overexpressed, and their inhibition promotes apoptosis and reduces motility, supporting a role in tumor survival and invasion. ⁶⁶ High-grade lesions further exhibit immune suppression, M2 macrophage enrichment, and disrupted IL34–CSF1R signaling, implicating this axis in tumor microenvironment reprograming. ⁶⁷ Systemic changes, including elevated G-CSF and immunosuppressive myeloid cells, suggest coordinated immune evasion. ⁶⁸ Consistently, CSF1R blockade in HPV-driven models reduces tumor-associated macrophages, enhances CD8⁺ T-cell infiltration, and delays tumor progression. ⁶⁹

BCL2L1 encodes the anti-apoptotic protein Bcl-xL, a key BCL-2 family member that promotes resistance to programed cell death. ⁷⁰ In cervical cancer, Bcl-xL is frequently overexpressed, with tumor cells showing strong dependence on anti-apoptotic proteins such as Bcl-xL and MCL-1 for survival. ⁷⁰ Altered co-expression of apoptosis-related genes, including Bcl-xL, p53, Bax, Bcl-2, and Mdm2, in cervical intraepithelial neoplasia (CIN) and cancer reflects disruption of apoptotic balance during malignant transformation. ⁷¹ These findings suggest that Bcl-xL upregulation facilitates the survival and expansion of dysplastic cervical epithelial cells during carcinogenesis. ⁷⁰

Despite evidence supporting the involvement of CSF1 and BCL2L1 in cervical dysplasia, the epigenetic regulation of this gene, particularly through hypomethylation, has not been well characterized. In our study, hypomethylation of CSF1 and BCL2L1, identified among differentially methylated genes (DMGs), was able to distinguish between normal cervical tissue and CIN grades 1 to 3. Furthermore, CSF1 and BCL2L1 were identified as a significant gene in Gene Ontology enrichment analysis, highlighting its potential biological relevance in cervical dysplasia progression and suggesting its utility as a potential epigenetic biomarker.

These findings should be considered a foundation for prospective clinical validation. Although the 2 hypomethylation oncogene and 2 hypermethylation TSG panels show strong theoretical potential, their diagnostic performance requires confirmation using quantitative methylation-specific PCR (qMSP) in independent cohorts. ⁷² Among the DMGs identified in the individual datasets, only WT1 was significantly differentially methylated in GSE287994. This suggests that while individual studies may lack the statistical power to identify these markers in isolation, their significance is robustly captured through our integrative meta-analysis.

From a clinical perspective, a validated methylation-based panel could serve as a triage tool for hrHPV-positive women, improving risk stratification beyond viral detection alone.^42,54,73 Importantly, DNA methylation markers can be reliably assessed in cervical exfoliated or scraped samples collected during routine screening, supporting the feasibility of implementing qMSP-based assays in clinical practice. ⁷⁴

Several limitations should be acknowledged. This study relies on publicly available datasets with heterogeneous sampling protocols and clinical annotations, and despite standardized preprocessing, potential batch effects may influence methylation measurements. The predominance of datasets from developed countries may also reduce generalizability. Furthermore, the absence of independent experimental validation underscores the need for prospective, multi-population cohorts and functional studies to confirm the biological relevance and clinical utility of these candidate biomarkers.

Conclusion

Our integrative analysis identified 4 promising biomarkers for detecting precancerous lesions in hrHPV positive women. These biomarkers were derived from high-dimensional datasets aggregated across diverse populations, enhancing their potential generalizability. Given their diagnostic relevance and translational promise, experimental validation in a large and diverse cohort should be pursued to confirm their utility in clinical settings.