2.1 Clinical specimens

Cervical cytology specimens preserved in PreservCyt^® solution (Hologic, Inc., Marlborough, MA, USA) were procured from multiple commercial biorepositories, including BIOFLUIDS.com (Los Osos, CA), Discovery Life Sciences (DLS, Huntsville, AL), and iSpecimen Inc. (Lexington, MA). All specimens had undergone prior cytological and histopathological assessment to confirm diagnostic status before their inclusion in this study. Each supplier attested that the biospecimens were collected under appropriate institutional protocols with documented informed consent from donors, ensuring ethical compliance for downstream research use. These Pap Smear derived specimens represent a non-invasive sampling method commonly used in routine cervical screening, making them highly suitable for clinical transcriptomic analysis.

All samples used in this study were de-identified Pap smear specimens purchased between April 17, 2024, and April 28, 2025. The research protocol was reviewed and approved by the Salus IRB with a waiver of informed consent, as the study posed no more than minimal risk (Protocol #ABS014 06 01 2025). Data collection and analysis were conducted only after IRB approval on June 27, 2025. At no point during or after the study did the researchers have access to information that could identify individual participants.

2.2 DNA and RNA extraction from clinical specimens

Cellular material from PreservCyt^® preserved specimens (Hologic, Inc., Marlborough, MA, USA) was recovered by pelleting. Approximately 3.5 mL of each specimen was used for nucleic acid extraction, with 1 mL allocated for genomic DNA isolation and the remaining 2 mL for total RNA purification. For DNA extraction, samples were centrifuged at 4,000 × g for 10 minutes at 4°C to pellet the cells. After discarding the supernatant, the resulting cell pellet was resuspended in 200 µL of phosphate-buffered saline (PBS). Genomic DNA was then extracted using the GenFind^TM V3 Blood, Cell 161 and Serum Genomic DNA Isolation Kit (Beckman Coulter, Brea, CA, USA; Cat. No. 162 C34881), following the manufacturer’s protocol. Purified DNA was stored at −20°C until further use.

RNA extraction followed a similar initial centrifugation step (4,000 × g for 10 minutes at 4°C). After removal of the supernatant, the pellet was immediately resuspended in 1 mL of TRIzol^TM Reagent (Thermo Fisher Scientific, Cat. No. 15596018), and total RNA was subsequently isolated using the PureLink^TM RNA Mini Kit (Thermo Fisher Scientific, Cat. No. 12183025), following the manufacturer’s “TRIzol Plus Total Transcriptome Isolation” protocol. RNA yield and purity were determined spectrophotometrically using the NanoDrop^TM Lite Plus (Thermo Fisher Scientific, Madison, WI, USA), while integrity was assessed via the Qubit^TM RNA HS Assay Kit (Invitrogen, Cat. No. Q32855) on a Qubit^TM 3.0 Fluorometer. In samples exhibiting visible red blood cell contamination, additional RNA processing was conducted using the GlobinClear^TM Human Globin mRNA Removal Kit (Thermo Fisher Scientific, Cat. No. AM1980) to eliminate abundant globin mRNA, which could interfere with downstream transcriptome profiling. All purified RNA samples were stored at −80°C until further downstream application.

2.3 HPV DNA genotyping

Genomic DNA from each sample was evaluated for HPV genotype using 20 parallel quantitative PCR (qPCR) reactions, each incorporating a custom-designed primer–probe set targeting a specific HPV genotype. The panel included high-risk types (HPV16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, and 68) and low-risk types (HPV6, 11, 26, 53, 66, 73, and 82). All primer–probe sets were acquired as 20 × stocks from Thermo Fisher Scientific, with the following specific Assay IDs: HPV16 (AIX02E2), HPV18 (AIY90LA), HPV31 (AI0IYRI), HPV33 (AI1RWXQ), HPV35 (AI20U3Y), HPV39 (AI39S96), HPV45 (AI5IRGE), HPV51 (AI6RPMM), HPV52 (AI70NSU), HPV56 (AI89LY2), HPV58 (AIAA04V), HPV59 (AIBJZA3), HPV68 (APZTGM6), HPV6 (AIGJRZZ), HPV11 (AIFATTR), HPV26 (AIHSP57), HPV53 (AII1OCF), HPV66 (AID1VNJ), HPV73 (AIKAMIN), and HPV82 (AILJKOV) (Supplementary S1 Table). Assays were conducted with appropriate positive controls (TrueMark^TM Comprehensive Microbiota Control; Thermo Fisher Scientific, Cat. No. A50382) and internal reference controls (TaqMan^TM Copy Number Reference Assay targeting the RNase P gene; Cat. No. 4403328). Human genomic DNA confirmed to be HPV-negative was included as a negative control for each qPCR run.

Reactions were prepared using the TaqMan^TM Fast Advanced Master Mix (Thermo Fisher Scientific, Cat. No. 4444557) and performed on a QuantStudio^TM 7 Pro Real-Time PCR System. Each 10 µL reaction mixture contained 5 µL of master mix, 0.5 µL of the genotype-specific primer–probe set, 2 µL of DNA template (10 ng/µL), and 2.5 µL of nuclease-free water. Thermal cycling conditions were standardized across all reactions and consisted of an initial incubation at 50°C for 2 minutes, followed by enzyme activation at 95°C for 2 minutes, and then 40 amplification cycles comprising denaturation at 95°C for 5 seconds and annealing/extension at 60°C for 20 seconds.

2.4 Assessment of E6/E7 mRNA expression by RT-qPCR

Quantification of E6 and E7 mRNA expression was performed on RNA samples confirmed to be HPV-positive by prior genotyping. The RT-qPCR assays were conducted using the iTaq^TM Universal SYBR^® Green One-Step Kit (Bio-Rad, Hercules, CA, USA; Cat. No. 1725151). Primer sequences targeting the E6/E7 regions were adopted from the study by Pan et al., which developed and validated a multiplex assay for high-risk HPV mRNA detection [25]. The specific primers used are listed in Supplementary S1 Table. Each 10 µL RT-qPCR reaction consisted of 5 µL of SYBR Green master mix, 0.125 µL of iScript^TM reverse transcriptase, 0.6 µL of a forward/reverse primer mix (0.3µM each final concentration), 2 µL of RNA template (20ng/µL), and 2.3 µL of nuclease-free water. Reactions were run using the following thermal protocol: reverse transcription at 50°C for 10 minutes, polymerase activation at 95°C for 4 minutes, followed by 40 amplification cycles consisting of denaturation at 95°C for 15 seconds and annealing/extension at 60°C for 30 seconds.

2.5 Transcriptome sequencing and gene expression quantification

Transcriptomic profiling was performed on extracted RNA samples following dilution to a concentration ≤ 30ng/µL, with a maximum input of 100 ng per reaction. Library preparation was carried out using the Ion AmpliSeq^TM Transcriptome Human Gene Expression Kit (Thermo Fisher Scientific, USA; Cat. No. A26327), according to the manufacturer’s protocol. Amplified libraries were purified using Agencourt^® AMPure^® XP magnetic beads (Beckman Coulter, USA; Cat. No. A63881) and quantified using the Ion Library TaqMan^TM Quantitation Kit (Thermo Fisher Scientific, USA; Cat. No. 4468802) on a StepOnePlus^TM Real-Time PCR System.

Eight RNA libraries were pooled per sequencing run, generating a 300 pM pooled library, which was loaded onto the Ion Torrent Genexus^TM Integrated Sequencer. Sequencing was performed under the instrument’s “Library-to-Result” automated workflow, enabling integrated template preparation, chip loading, sequencing, and initial data processing. Sequence reads were aligned to the hg19 human reference genome using Torrent Suite^TM Software (Thermo Fisher Scientific; version 6.8.1.1), and raw gene expression was quantified using the AmpliSeq RNA plug-in. Gene counts were exported for downstream differential expression and pathway analysis.

2.6 Bioinformatics and computational analyses

Normalization and differential gene expression (DGE) analyses were performed using the DESeq2 R package (v1.36.0). Raw count data were normalized using the median-of-ratios method to control for library size and composition bias. To address the inherent intra-group heterogeneity of ASCUS lesions, a group-wise differential design was implemented, comparing each cytological category (ASCUS, LSIL, HSIL) directly to the HPV-negative NILM reference (n = 51). This approach mitigated lesion-level variability by aggregating expression trends across individuals rather than relying on pairwise or hierarchical contrasts. Differentially expressed genes (DEGs) were defined by |log₂ fold-change|≥2 and adjusted p-value<0.05 (Benjamini–Hochberg correction). Empirical Bayes shrinkage of |log₂ fold-changes| and Wald tests were used to control for overdispersion and sample imbalance.

2.7 Functional enrichment analyses

Gene Ontology (GO) and pathway enrichment analyses were performed using WebGestalt (2023 release) and the STRING database enrichment tool. GO enrichment analysis encompassed three domains: Biological Process (BP), Molecular Function (MF), and Cellular Component (CC), with an FDR threshold of <0.05. For pathway analysis, KEGG, Reactome, and WikiPathways databases were queried to identify statistically overrepresented pathways (FDR < 0.05), using the full human genome as the background set.

To control for intergroup variability, enrichment scores were aggregated across all DEGs within each cytological class, allowing the identification of consensus pathways rather than individual sample–specific events. This pathway aggregation minimized the influence of outlier expression particularly in heterogeneous ASCUS lesions and highlighted reproducible biological themes such as early oncogenic signaling and immune modulation.

2.8 Hub gene identification via network centrality

The hub genes were identified through integrative network analysis using multiple centrality measures. Protein–protein interaction (PPI) networks were generated in STRING v11.5 with a minimum interaction confidence score of 0.7. All transcriptomic hub genes were mapped to their corresponding proteins to ensure that network topology reflected biologically relevant protein-level associations. Networks were filtered by node degree to retain high-confidence connections and visualized in Cytoscape (v3.10.3) and hub genes were subsequently identified using CytoHubba, a Cytoscape plugin that ranks nodes based on 12 centrality algorithms: Degree, Edge Percolated Component (EPC), Maximum Neighborhood Component (MNC), Density of MNC (DMNC), Maximum Clique Centrality (MCC), Bottleneck, Eccentricity, Closeness, Radiality, Betweenness, Clustering Coefficient, and Stress. Genes ranked in the top 20 across at least two algorithms were considered robust hub genes. To further reduce cytology-related heterogeneity and false positives, only connected network components were analyzed, excluding isolated nodes.

2.9 Driver-passenger analysis of transcriptomic alterations

To distinguish driver from passenger gene expression changes, an integrative framework was employed incorporating temporal expression trends across cytological progression (NILM → HSIL), network centrality metrics (betweenness, degree, closeness) from PPI analysis, functional enrichment context, and cross-referencing with curated cancer driver databases, including COSMIC and IntOGen. A gene was classified as a driver if (i) up-regulated across ≥2 progressive stages, (ii) hub centrality >0.8 in ≥ 1 cytology, and inclusion in FDR < 0.01 pathways. Genes meeting none or only one criterion were classified as passengers.

2.10 Pathway crosstalk and network integration

To evaluate inter-pathway relationships across cytological stages, a pathway crosstalk network was constructed using Cytoscape, where each node represented a pathway and edges reflected gene overlap. Only pathways with FDR < 0.05 and ≥3 constituent genes were retained. Pairwise similarity between pathway gene sets was calculated using both Jaccard Coefficient (JC) and Overlap Coefficient (OC) based on set sizes and overlap. The JC was calculated as JC =(| A ∩ B |)/(| A ∪ B|) and OC as OC = (| A ∩ B |)/(min (|A|,|B|), where A and B depict gene sets of two distinct pathways [26,27]. Both coefficients range from 0 (no similarity) to 1 (complete similarity), but they emphasize different aspects of the relationship between sets. The Jaccard coefficient provides a broader, more balanced measure, while the overlap coefficient highlights the degree of containment relative to the smaller set. Pathway pairs with fewer than two shared genes were excluded. Topological analysis of the crosstalk network was conducted using the CentiScape plugin to compute nodal degree and identify highly connected “hub” pathways.

2.11 PPI network analysis

The STRING-derived PPI network for all hub genes (n = 151 nodes) was refined to include only edges with interaction confidence ≥0.7 and node degree >5. The final network included 2,459 edges and was stratified into functional modules. Cluster analysis using STRING’s functional categorization highlighted three major modules: (i) chemokine response cluster, (ii) mitochondrial translation cluster, and (iii) mitotic spindle and replication cluster, corresponding to early immune evasion, mid-stage viral productivity, and high-grade dysplastic transformation, respectively.

2.12 Statistical analysis

Statistical analysis was performed using SPSS software (version 29; SPSS Inc, Chicago, IL, USA). The association between cytology categories and HPV type and E6/E7 mRNA Ct levels was evaluated using the Chi-square test. The correlation between cytology categories and E6/E7 mRNA Ct value was evaluated using Spearman’s Rank correlation. One-Way analysis of variance (ANOVA) was used to compare mean Ct value across different cytological categories, followed by Tukey’s post-hoc test for pairwise comparison. A p-value <0.05 was considered statistically significant.

3.1 Clinical specimen characterization

A total of 183 Pap Smear specimens were included in this study and analyzed, stratified into four cytological categories and one control category: Negative for Intraepithelial Lesion or Malignancy but HPV infected (HPV-positive NILM, n = 37), Atypical Squamous Cells of Undetermined Significance (ASCUS, n = 44), Low-grade Squamous Intraepithelial Lesion (LSIL, n = 35), High-grade Squamous Intraepithelial Lesion (HSIL, n = 16), and HPV-negative NILM (base control) (n = 51) (Table 1). HPV genotyping identified: high-risk types (HPV 16, 18, 31, 33, 35, 39, 45, 51, 52, 56, 58, 59, and 68) and low-risk types (HPV 6, 11, 26, 53, 66, 73, and 82). Of the 132 specimens, 73 (55.3%) exhibited single HPV strain infections, while 59 (44.7%) showed multiple HPV strains infections. HPV E6 and E7 mRNA expression levels measured by reverse transcription PCR cycle threshold (Ct) values, were categorized as: strong (Ct < 25, n = 74), moderate (25 ≤ Ct < 28, n = 19), weak (28 ≤ Ct < 35, n = 30), and undetectable (Ct ≥ 35, n = 9).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of 132 HPV-Positive Pap Smear Specimens.

https://doi.org/10.1371/journal.pone.0339693.t001

3.2 E6/E7 mRNA Ct with HPV infection and cytology categories

Statistical analyses were conducted to explore the relationship between cytological categories, HPV infection patterns (single vs. multiple), and E6/E7 mRNA expression measured by cycle threshold (Ct) values. A chi-square test showed a significant association between cytology categories and Ct level categories (X² = 24.407, df = 9, p = 0.003). Spearman Rank Correlation showed a significant negative correlation between cytology categories and Ct value (ρ = −0.321, p < 0.001), suggesting that Ct values (indicating higher viral mRNA expression) tend to decrease as cytological abnormalities more severe. No significant relationship was found between cytology categories and HPV infection pattern (single vs. multiple) (X² = 0.249, df = 3, p = 0.983). One-Way ANOVA indicated a significant difference in mean Ct values across the cytological categories (F = 3.17, df = 3, p = 0.027). Tukey’s post-hoc comparison revealed that the mean Ct value for LSIL cases (22.3 ± 4.2) was significantly lower than for NILM cases (25.83 ± 6.3, p = 0.019). No significant differences in mean Ct values were observed among other cytological category comparisons (ASCUS: 24.68 ± 4.7; HSIL: 22.87 ± 4.74).

3.3 Identification of DEGs

Transcriptomic analysis identified 9,194 differentially expressed genes (DEGs) across all cytological categories (HPV-positive NILM, ASCUS, LSIL, HSIL) when compared to HPV-negative NILM controls (|log₂FC| ≥ 1, p < 0.05). Of these, 891 genes (9.7%) met the stringent criteria of |log₂FC| ≥ 2 and p < 0.05 across the cytology groups when compared to normal controls. The distribution of significant DEGs per cytological category was as follows: NILM (498 upregulated, 12 downregulated); ASCUS (85 upregulated, 3 downregulated); LSIL (165 upregulated, 6 downregulated); and HSIL (70 upregulated, 52 downregulated). In total, 818 genes were upregulated and 73 were downregulated across all categories. DEG identification was performed using DESeq2 with false discovery rate correction for multiple testing.

3.4 Stage-specific hub gene signatures reveal molecular progression

Hub gene analysis identified 308 highly connected genes across the four cytological categories, revealing distinct functional signatures at each stage of HPV-driven progression. These hub genes represent the most influential molecular players driving stage-specific biological processes. Hub genes were subsequently identified using CytoHubba, a plugin for cytoscape.

3.5 Summary of progression

Transcriptomic analysis reveals a stepwise molecular progression from HPV infection to precancerous lesions. In NILM, despite normal cytology, we observed a coordinated upregulation of protein synthesis machinery (RPL5, RPS6) and growth signaling (EGFR, MYC) alongside strategic downregulation of inflammatory pathways (IL1B, IL8, CXCL1/2), establishing viral persistence while maintaining epithelial homeostasis. These dual strategies-upregulating host translational and proliferative machinery while simultaneously suppressing immune signaling-represent HPV’s initial foothold in establishing persistent infection despite normal cytology.

ASCUS maintains this translational emphasis while introducing early oncogenic drivers (CCND1, SHH) and continued immune suppression. These oncogenic factors signal the beginning of cellular transformation without full dysplastic commitment. As lesions progress to LSIL, hub genes shift toward epithelial differentiation alteration (AGR2, DAG1) and intensified viral productivity, with suppression of antimicrobial defenses (MPO, DEFA4). This stage represents active viral replication and assembly, reflected in the robust translational signature and emerging viral escape from epithelial defense mechanisms.

The transition to HSIL marks the most dramatic molecular reprogramming, characterized by intense cell cycle hyperactivation (CDK1, PLK1), DNA replication licensing (MCM3–6, RFC3/4), and chromosomal instability (KIF family, CENPF), concurrent with significant downregulation of epithelial differentiation markers (SPRR1A, KRT family) and barrier function genes. This represents HPV’s transition from productive infection to transformative integration, with viral oncoproteins driving uncontrolled proliferation, genomic instability, and loss of differentiation.

Preliminary trends, further supported by enrichment analysis in subsequent sections, revealed a clear progression from homeostatic maintenance in NILM to increasingly dysregulated cell cycle and suppressed differentiation in HSIL (p < 0.01). These findings demonstrate HPV’s molecular evolution from immune evasion to host genome subversion, with distinct hub gene signatures marking critical transitions in dysplastic severity.

3.6 GO enrichment analysis of hub genes

Gene ontology (GO) analysis was performed across cytological groups to identify the functional roles of hub genes associated with cervical lesion progression. This analysis included three primary aspects: Biological Process (BP), Cellular Component (CC), and Molecular Function (MF). The functional characterization of genes associated with four cytological stages was performed using WebGestalt. An over-representation analysis was performed, selecting “Biological Process” as the GO functional category, with gene symbols as identifiers (ID type) and entire genome as the enrichment background.

3.6.1 Biological Process (BP) analysis reveals functional reprogramming.

Gene Ontology analysis revealed a systematic shift in cellular priorities across disease progression. Early-stage lesions (NILM/ASCUS) prioritized metabolic processes (51 and 42 genes respectively), reflecting the virus’s hijacking of host biosynthetic machinery. However, HSIL demonstrated a fundamental reprogramming toward regulatory and stress-response functions, with significant enrichment in biological regulation (37 genes, p < 0.001), stimulus response (32 genes), and cellular reorganization (35 genes) (Fig 1A). This transition from metabolic support to dysregulated control mechanisms marks the shift from viral persistence to malignant transformation. Downregulated processes showed consistent immune suppression across all stages, with response to stimuli (34 genes) and cell communication (33 genes) prominently suppressed in NILM, establishing the immunotolerant foundation for progression (Fig 1B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. GO Biological Process Enrichment of Differentially Expressed Hub Genes Across Cervical Lesion Stages.

Bar plots illustrating Gene Ontology (GO) biological process (BP) enrichment of upregulated (Fig 1A), downregulated (Fig 1B) and hub genes across four cervical cytological stages: NILM (Negative for Intraepithelial Lesion or Malignancy), ASCUS (Atypical Squamous Cells of Undetermined Significance), LSIL (Low-grade Squamous Intraepithelial Lesion), and HSIL (High-grade Squamous Intraepithelial Lesion). Enrichment analyses were performed using WebGestalt (2023 release), applying a false discovery rate (FDR) threshold of <0.05. Cytological groups are color-coded as follows: blue (NILM), green (ASCUS), red (LSIL), and orange (HSIL). Bar length indicates the number of genes associated with each enriched term, and the x-axis shows –log10 transformed FDR values indicating statistical significance.

https://doi.org/10.1371/journal.pone.0339693.g001

3.7 Functional progression across cytological categories

Gene ontology analysis demonstrates distinct functional profiles from NILM to HSIL across multiple GO domains. In Biological Process terms, upregulated genes shift from metabolic activity to regulatory functions. Metabolic processes are enriched in earlier stages: NILM (24.4%), ASCUS (24.1%), and LSIL (25.6%) but decrease in HSIL (15.4%). Conversely, genes involved in biological regulation increase from NILM (11.96%) to HSIL (15.81%), becoming the most enriched category in HSIL. Additionally, we observed stress response elevation and tissue architecture disruption in HSIL: response to stimulus peaks in HSIL (13.7% vs ≤ 9.8% in other stages), while cellular component organization doubles in HSIL(15% vs ≤ 8.6% in other stages). Meanwhile, cell communication is significantly reduced in ASCUS (6.3%) and LSIL (7.5%) compared to NILM (8.6%), showing mid-grade signaling suppression. Downregulated genes also exhibit distinct stage-specific patterns: HSIL shows severe suppression of cell communication (4% vs 13–15% in other stages) and cell population proliferation (2% vs 5% in NILM/LSIL), indicating progressive loss of growth control mechanisms. Furthermore, HSIL doubled disruption of developmental processes (18.2%) compared to early stages (≤8.7%).

In the Cellular Component terms, upregulated genes in HSIL reflect marked architectural changes. Nuclear components are significantly elevated in HSIL (16.9%) compared to NILM (8.6%), and chromosomal elements rise dramatically to 10.5%, representing a five-fold increase over other stages (≤2.8%). In contrast, components related to mitochondria and protein synthesis machinery, such as ribosomes and envelopes are notably reduced or absent in HSIL (2.5% vs. 7.9–9.1% in earlier stages). Cytoskeletal reorganization is distinctly enriched in HSIL (8.0% vs. ≤ 2.2% in other grades), while membrane and protein complexes remain relatively stable across all stages. Among downregulated genes, membrane components remain consistent across stages. However, HSIL shows pronounced suppression of cytosolic (15.1%) and nuclear (11.8%) components. LSIL, on the other hand, is marked by disruption in the extracellular space (17.7%) and vesicular transport systems, including the endomembrane system and vesicles (16.1% each). Notably, protein complexes appear to be selectively retained in HSIL downregulation (2.2%), contrasting with higher suppression rates in earlier stages (8–10%).

In Molecular Function terms, upregulated genes in HSIL reveal a distinct pattern of functional reprogramming. Ion binding becomes the dominant activity in HSIL (16.6% vs. ≤ 12.9% in earlier stages), while structural molecular activity is nearly eliminated (0.6% compared to 12–15% in other stages). Nucleic acid binding is most prominent in normal cytology (22.9%) but decreases to 14.7% in HSIL. Catalytic activity shows stage-dependent changes, with hydrolase activity rising fourfold from NILM (2.5%) to HSIL (9.2%), and chromatin binding increasing during dysplastic progression. Protein binding remains stable across all stages (26–28%), suggesting preservation of core molecular interactions. Downregulated genes show pronounced suppression of structural molecule activity (16.1%), and a near-complete loss of molecular transducer activity (1.8% vs. 5–13% in other stages) in HSIL. Protein binding downregulation increases progressively, peaking in LSIL (50%), while catalytic hydrolase activity is entirely absent at this stage. ASCUS displays a specific decline in molecular adaptor functions (7.1% downregulation), which are completely lost in HSIL.

This GO term progression provides a functional framework for understanding the systematic reprogramming of cellular machinery during HPV-driven neoplastic transformation.

3.8 Pathway enrichment analysis of hub genes

Pathway enrichment analysis was performed to highlight the key molecular mechanisms and their implications for cervical lesion progression, focus on statistical significance (FDR), biological relevance, and gene involvement. STRING enrichment tools, Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome, and WikiPathways databases were used. To ensure that the enrichment analysis carried out across these databases provides comprehensive pathway coverage, we considered pathways with a false discovery rate (FDR), p < 0.05. The analysis revealed a distinct stage-specific pathway progression across cytological categories. In NILM group, the upregulated hub genes were most significantly enriched in mitochondrial translation (elongation, initiation, termination), metabolism of proteins, ribosome/translation (Supplementary S6.1 Table), while downregulated genes were enriched in immune system, innate immunity, NOD-like receptor signaling, chemokine signaling, and IL-17 signaling pathways(Supplementary S6.2 Table), reflecting a balance between active proliferation and suppressed inflammation that maintains epithelial homeostasis.

ASCUS showed a shift towards mitochondrial translation and cancer-related pathways (e.g.,pathways in cancer, proteoglycans in cancer, breast cancer, prostate cancer, bladder cancer) in upregulated genes, with notable EGFR tyrosine kinase inhibitor resistance (Supplementary S7.1 Table), while downregulated genes remained enriched in inflammatory signaling (IL-17, TNF, NF-κ B) and cytokine-cytokine receptor interactions, indicating cells under proliferative stress with emerging immune evasion (Supplementary S7.2 Table).

LSIL maintained upregulation in mitochondrial translation and protein metabolism while adding hormonal and receptor related pathways (mammary gland development and nuclear receptors) (Supplementary S8.1 Table), with downregulation in cytokines/chemokines pathways representing a transition toward dysplasia with metabolic reprogramming in a tolerogenic microenvironment (Supplementary S8.2 Table).

HSIL demonstrated significant upregulation in cell cycle/DNA replication, DNA repair/replication stress response, and oncogenic signaling/transcriptional dysregulation pathways (Supplementary S9.1 Table), with downregulation in cornified envelope formation, oxidative stress response, and mineral homeostasis pathways, reflecting uncontrolled proliferation, genomic instability, and epithelial barrier disruption (Supplementary S9.2 Table). These findings illustrate the progression from metabolic equilibrium in NILM to immune evasion in ASCUS, metabolic/hormonal rewiring in LSIL, and ultimately cell-cycle hyperactivation, replication stress, and chromosomal instability characteristic of high-grade dysplasia and epithelial disintegration in HSIL, providing a comprehensive molecular framework for understanding cervical carcinogenesis.

3.9 HPV-related pathways by cytology

After excluding pathways unrelated to HPV, the most biologically relevant pathways associated with HPV infection and cervical lesion progression were presented as follows (Supplementary S10 Table):

3.10 Pathway crosstalk analysis

Pathway crosstalk analysis was performed to identify shared molecular signatures and interaction networks between the cytological categories. We constructed a network graph where nodes represent unique pathways and edges represent gene overlap between pathways. Pathways with FDR adjusted p-value >0.05 or those represented by fewer than 3 genes were excluded to maintain analytical rigor. Furthermore, to calculate the overlap between pathways, the Jaccard Coefficient (JC) and Overlap Coefficient (OC) was calculated. Specifically, the JC was calculated as JC =(| A ∩ B |)/(| A ∪ B|) and OC as OC = (| A ∩ B |)/(min (|A|,|B|), where A and B depict gene sets of two distinct pathways. Cytoscape was then used to visualize the interrelationships among these pathways. Network analysis showed a fair interconnection among the cytological categories representing 81 pathways shared across >2 cytological categories (1513 edges) and none of them are isolated (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Pathway Crosstalk Network Across Cervical Cytological Categories (Full Network).

This network diagram illustrates the inter-pathway crosstalk among statistically enriched biological pathways identified across four cervical cytological categories: NILM (blue), ASCUS (green), LSIL (red), and HSIL (orange). Each node represents a unique pathway enriched in one or more cytological stages (FDR < 0.05, ≥ 3 constituent genes) and edges represent gene overlap between pathways. The overlap between pathways is quantified using the Jaccard Coefficient (JC) and Overlap Coefficient (OC), with a minimum of two overlapping genes required for inclusion. Pathways exhibiting the highest centrality are visualized as hub nodes and reflect biologically critical signaling axes during neoplastic progression. Notably, NILM pathways clustered toward immune regulation and epithelial adhesion, ASCUS and LSIL retained significant overlap with NILM (OC > 0.9), while HSIL pathways formed a distinct cluster dominated by cell cycle, replication, and chromatin remodeling processes. Network topology was visualized in Cytoscape and pathway identities are annotated according to KEGG, Reactome, and WikiPathways databases.

https://doi.org/10.1371/journal.pone.0339693.g002

To identify the most interconnected and influential pathways, we performed nodal degree analysis using the CentiScape plugin to identify key pathway nodes with nodal degree >5, which resulted in 668 edges depicted in Fig 3. NILM showed the largest distinct enriched pathways (n = 234), followed by ASCUS (n = 46), LSIL(n = 41), and HSIL(n = 89). JC and OC demonstrated high overlap between ASCUS and LSIL with NILM (OC = 0.978, 0.927, respectively), indicating that ASCUS shared most of its pathway profile with NILM (45 out of 46 ASCUS pathways were found in NILM), and LSIL sharing a large portion of its pathways with NILM (38 out of 41 LSIL pathways were found in NILM). Additionally, a moderate similarity (OC = 0.512, 21 out of 41 LSIL pathways were found in ASCUS) was observed between ASCUS and LSIL. In contrast, HSIL showed minimal overlaps with other cytological categories with NILM (OC = 0.067), ASCUS (OC = 0.043), and LSIL (OC = 0), indicating the set of pathways in HSIL was largely unique to the high-grade lesion category.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Pathway Crosstalk Network Across Cervical Cytological Categories (Node degree ≥5).

This network visualization depicts the subnetwork with nodal degree ≥ 5 representing crosstalk among significantly enriched molecular pathways (FDR < 0.05; ≥ 3 genes) across four cytological stages of HPV-positive cervical lesions: NILM (blue), ASCUS (green), LSIL (red), and HSIL (orange). Each node represents a unique biological pathway identified through enrichment analysis using KEGG, Reactome, and WikiPathways databases. Node color indicates the cytological stage(s) in which the pathway was enriched. Edges between nodes reflect shared gene content, with a minimum threshold of two overlapping genes for edge retention. The network illustrates how pathway architecture evolves with disease progression: early-stage NILM and ASCUS pathways cluster around immune signaling and homeostatic regulation, while LSIL pathways show transitional overlap with metabolic and growth factor signaling. In contrast, HSIL forms a distinct, tightly connected module centered on cell cycle regulation, DNA replication, mitotic spindle assembly, and chromatin remodeling. High-connectivity nodes such as “Cell Cycle,” “DNA Replication,” and “ATR Activation in Response to Replication Stress” emerge as central hubs in HSIL, reflecting oncogenic pathway consolidation. The increasing modularity and divergence of HSIL-associated pathways highlight a molecular shift from viral persistence to host genome destabilization and dysplastic transformation.

https://doi.org/10.1371/journal.pone.0339693.g003

ASCUS shares the immuno suppression and hyper-translational state with NILM and LSIL, indicating these are fundamental to HPV infection. What differentiates ASCUS is the incipient activation of oncogenic pathways: the first signs of the host cell responding to viral oncoproteins with changes characteristic of cancer development (growth signals, cell cycle dysregulation). These changes signal a molecular infection point, where immune evasion gives way to nascent oncogenic transformation.

The LSIL molecular signature shares the two main themes of early HPV infection (immune evasion and enhanced protein synthesis) with NILM and ASCUS. What becomes most pronounced in LSIL is the protein synthesis aspect, corresponding to actual virion production. LSIL can thus be distinguished by extremely high levels of ribosomal and mitochondrial protein genes. In contrast, LSIL does not yet share the aggressive cell-cycle signature of HSIL. LSIL is considered as a state where the virus is reproducing efficiently, but the host cell genome is largely intact (no integration in most cases) and the cell’s malignant transformation is not fully realized. Thus, LSIL is often regarded as a transient infection state and can regress if the immune system eventually responds. The HSIL signature is quite specific and distinct from NILM/ASCUS/LSIL. The massive upregulation of cell cycle and replication pathways is unique to high-grade lesions. Early-stage signatures (immune down, translation up) are less conspicuous in HSIL, largely because these aspects are already established (immune pathways long suppressed, and protein synthesis is generally high in all dividing cells). In practice, markers like MCMs, TOP2A, KI-67, and p16 are used to identify HSIL cells, all reflecting the uncontrolled cell cycle that our pathway analysis also captures [65]. HSIL does not share many of the unique “productive infection” features seen in LSIL. For example, HSIL exhibited diminished viral late gene expression (the virus often integrates and stops making L1/L2). Therefore, LSIL and HSIL have almost opposite dominant signatures; LSIL is defined by viral gene expression, HSIL by host cell gene dysregulation.

3.11 PPI network analysis

The PPI was constructed using STRING database and visualized in Cytoscape containing 151 nodes and 2459 interconnected edges. The nodal degree >5 was maintained for PPI network analysis depicted in Fig 4. The STRING analysis categorized a list of genes into 3 distinct categories which comprise cellular response to chemokine, mitochondrial cluster, and mitotic cell cycle processes (Fig 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Global Protein–Protein Interaction (PPI) Network of Hub Genes Across Cytological Categories.

This figure illustrates the comprehensive PPI network of all hub genes identified across the four cytological categories of HPV-positive cervical specimens: NILM, ASCUS, LSIL, and HSIL. The network was constructed using the STRING database (v11.5), with a high-confidence interaction score threshold of ≥ 0.7 to ensure biologically meaningful associations. Only nodes (genes) with a degree > 5 were retained for visualization, resulting in a network of 151 nodes and 2,459 edges. Each node represents a unique hub gene, and edges indicate protein–protein interactions based on experimental data, curated databases, and predictive algorithms. The spatial organization of the network was generated using force-directed layout in Cytoscape (v3.10.3), allowing clusters and highly interconnected regions to emerge visually without imposed functional categorization. This unlabeled network provides a global overview of hub gene connectivity and highlights the dense interaction patterns that underpin the transcriptomic reprogramming during HPV-driven cervical lesion progression. The visualization serves as a reference for understanding network topology prior to functional modular classification.

https://doi.org/10.1371/journal.pone.0339693.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Protein–Protein Interaction (PPI) Network of Hub Genes with Functional Cluster Labeling.

This figure displays the high-confidence PPI network constructed from all identified hub genes across the four cytological stages of HPV-positive cervical lesions. Network construction was performed using the STRING database (v11.5), with a minimum interaction confidence score threshold of 0.7 to retain high-confidence associations. The resulting network includes 151 nodes (hub genes) and 2,459 edges, visualized in Cytoscape (v3.10.3). Only genes with a node degree >5 were retained to minimize noise and highlight core interactors. Nodes represent individual hub genes, and edges denote experimentally validated or predicted protein–protein associations. Functional modules were identified through STRING functional enrichment clustering and labeled based on biological relevance. Three major functional clusters are highlighted: 1. Mitochondrial Translation Cluster: Encompassing genes involved in ribosomal biogenesis and protein synthesis, reflecting enhanced metabolic and translational activity characteristic of early lesion stages (e.g., NILM and ASCUS). 2. Chemokine Response Cluster: Including immune modulators and signaling components involved in cytokine–cytokine receptor interactions and inflammatory signaling pathways, largely downregulated in early lesions but indicative of host immune surveillance. 3. Mitotic and DNA Replication Cluster: Comprising cell cycle regulators, DNA replication licensing factors, and mitotic spindle components, predominantly enriched and upregulated in high-grade lesions (HSIL), representing proliferative dysregulation and genomic instability. This cluster-labeled network underscores the temporal transition from immune evasion and metabolic support (early lesions) to proliferative transformation (HSIL), providing insight into molecular reprogramming along the cervical neoplastic continuum.

https://doi.org/10.1371/journal.pone.0339693.g005

4.1 Molecular prodrome in HPV-positive NILM: gateway to early prevention

Our transcriptomic analysis of HPV-positive cervical samples reveals that even cytologically normal (NILM) specimens harbor significant molecular alterations that precede morphological abnormalities. This ”molecular prodrome” offers a unique opportunity for early risk prediction and preventive intervention.

In HPV-positive NILM samples, we identified significant molecular alterations. A persistent upregulation of growth factor receptor signaling (EGFR, ERBB2, PTK2, PRKCA; log₂FC > 1.5, p < 0.001) with high betweenness centrality (>0.85) was observed. These genes promote cell survival and proliferation and may be influenced by the activity of HPV oncoproteins. Although our study did not directly assess E5 expression, prior studies have shown that HPV16 E5 enhances EGFR signaling by promoting receptor recycling and sustained activation [66]. In addition, HPV E6 and E7 broadly disrupt multiple growth signaling cascades and reduce the expression of cell-surface adhesion molecules to evade immune surveillance [66].

In parallel, the upregulation of ribosomal and protein translation pathways (RPL5, RPL8, RPS6, MRPL14, MRPS11, MRPS2, MRPL48, GADD45G1P1) indicate a hypermetabolic state, consistent with reports of HPV-driven metabolic reprogramming via E6-mediated activation of the mTOR signaling pathway [66,67].

Concurrently, strategic downregulation of immune mediators (IL1B, IL8, CXCL1/2, CCL20, NFKB1, IKBKG) contribute to an immune suppressive microenvironment conducive to viral persistence. Prior studies have shown that HPV E6/E7 can inhibit NF-kB signaling, suppressing antiviral cytokine expression [65], and downregulate CCL20 through NF-κB pathway suppression [68].

These findings reveal a “stealth infection” profile of the cervical epithelium during HPV persistence, characterized by proliferative signaling, metabolic activation, and immune evasion.

4.2 Transcriptome-defined stage-specific signatures: beyond current paradigms

ASCUS represents a crucial early stage in the development of cervical cancer. At this point, ASCUS maintains the HPV-driven immune evasion and metabolic activation seen in NILM. Simultaneously, it introduces oncogenic signaling pathways, such as “Pathways in Cancer” and “Proteoglycan in Cancer”. Key genes involved include CCND1, SHH, IGF1, CDH1, and EGFR. Early overexpression of CCND1 is a known event in cervical dysplasia, contributing to unscheduled cell cycle entry [65]. This, alongside persistent immune suppression and enhanced protein synthesis establishes a molecular environment primed for progression. Notably, ESR1 appears in the ASCUS “Pathway in Cancer” gene list, suggesting that estrogen signaling may be active in ASCUS lesions, potentially promoting growth of infected cells.

LSIL shares two key features of early HPV infection: immune evasion and enhanced protein synthesis with NILM and ASCUS, but it also exhibits a distinct “viral productivity” signature. This is characterized by heightened translational activity and downregulation of antimicrobial defenses (MPO, DEFA4, CEACAM8), creating an environment optimized for viral replication and assembly. The robust translational signature (MRPS9, MRPL19, RPL8, RPS6, GADD45GIP1) reflects the virus’s metabolic hijacking of host machinery for virion production, though LSIL has not yet acquired the aggressive cell-cycle dysregulation seen in HSIL. At this stage, the virus reproduces efficiently, but the host genome remains largely intact, and malignant transformation is not yet fully realized. This may explain why LSIL is often a transient infection state that can regress if the immune system mounts an effective response. This stage-specific signature offers potential targets for antiviral therapies aimed at disrupting viral assembly before genomic integration occurs.

HSIL, by contrast, displays a distinct molecular profile from NILM, ASCUS and LSIL. It is characterized by dramatic molecular reprogramming dominated by cell cycle hyperactivation (CDK1, cyclin B1, CENPF, KIF2C, PLK1, FOXM1) and DNA replication licensing (MCM family, CDC45, PCNA, POLA2, FEN1, RFC3/4), along with epithelial dedifferentiation (SPRR1A, KRT family). The upregulation of cell cycle genes indicates HSIL cells are rapidly proliferating without normal regulatory control. In infections with HPV16, E7 was shown to be primarily responsible for this, as it binds and inactivates Rb, freezing E2F to transactivate S-phase genes and override cell cycle checkpoints [65].

This molecular transformation reflects HPV’s transition from productive infection to transformative integration, with viral oncoproteins driving uncontrolled proliferation and genomic instability. Notably, HSIL and LSIL do not share any significant pathways (OC = 0). In fact, they exhibit almost opposite dominant signatures: LSIL is defined by viral gene expression, whereas HSIL is characterized by host cell gene dysregulation. This underscores a fundamental molecular discontinuity between these stages, marking a critical threshold in disease progression.

Our transcriptome analysis reveals dynamic interplay between molecular drivers and passengers across lesion progression: from foundational driver activation (EGFR/ERBB2) in early stages, to mitotic deregulation (CDK1/PLK1) in HSIL, culminating in replication stress and licensing deregulation (MCMs) that characterize high-grade lesions. This progression suggests distinct therapeutic windows for intervention at each stage.

While genomic mutation analysis can identify structural alterations, only transcriptomic profiling can capture the dynamic functional consequences of these mutations and viral integration events. Our approach reveals not just which genes are altered, but how their expression patterns change across disease progression, providing a functional readout of disease activity that surpasses the capabilities of DNA-based or cytology testing alone.

4.3 Candidate biomarkers

We identified a novel class of transcriptome-based biomarkers repressed in a stage-specific manner in response to HPV E6/E7 oncoproteins. Unlike genotyping or morphology, these biomarkers report the actual dynamic biological state of the cervical epithelial cells. Here, we propose specific gene sets aligned to key stages of cervical lesion progression as candidate biomarkers: For NILM, indicative of viral persistence with immune evasion, candidates include RPL5, RPS6, and EGFR, with downregulated IL1β and CXCL1. The ASCUS stage, marking oncogenic initiation with suppressed inflammation, candidates include CCND1, SHH, PARP1, and IGF1, with downregulated IL8. LSIL, characterized by active viral replication and antimicrobial suppression, candidates include DROSHA and RPL8, plus downregulated MPO and DEFA4. Finally, HSIL, associated with genomic instability and loss of epithelial identity, candidates include CDK1, PLK1, MCM4−6, and EZH2, alongside downregulated KRT1 and SPRR1A. Critically, these signatures not only delineate the current lesion stage but also hold potential for predicting future trajectory (Supplementary S11 Table). For instance, IL1β is one central key player in the immune surveillance interactome by driving inflammation and connecting innate and adaptive immunity. In HPV-infected NILM samples, HPV E6/E7 oncoproteins interfere with the normal processing of pro-IL-1, resulting in reducing the secretion of IL-1β in virus-infected keratinocyte, a key mechanism that helps HPV evade immune evasion and initiate malignancy [69]. The SHH ligand enhances the proliferative and migratory capacity of cervical cancer cells and contributes to the development of radioresistance during cancer treatment [70,71]. Increased expression of EZH2 has been linked with the carcinogenesis, proliferation, biology behavior, and the clinical outcome of cervical cancer. High EZH2 level was correlated with more advanced clinical stage, histologic differentiation, infiltration depth, and lymph node metastasis. Patients exhibiting EZH2-positive expression tend to have a decreased overall survival compared with those with EZH2-negative expression [72].

4.4 E6/E7 mRNA expression: the critical link between viral activity and disease progression

Our analysis of E6/E7 mRNA expression levels (measured by PCR Ct values) provides compelling evidence for the superiority of transcriptomic approaches over conventional HPV DNA testing. We demonstrated a significant association between viral oncogene expression and cytological abnormalities (X² = 24.407, p = 0.003). Spearman Rank Correlation showed a significant negative correlation between cytology categories and Ct value (ρ = −0.321, p < 0.001), suggesting that Ct values (indicating higher viral mRNA expression) tend to decrease as cytological abnormalities more severe. These findings directly validate the central role of active viral oncogene expression rather than mere viral presence in driving disease progression.

The transformative power of transcriptomics is particularly evident when comparing our E6/E7 expression data with conventional HPV genotyping results. While 44.7% of our specimens showed multiple HPV strain infections, we found no significant relationship between infection pattern (single vs. multiple) and cytological categories (X² = 0.249, p = 0.983). This striking contrast reveals the fundamental limitation of DNA-based testing: the presence or number of HPV genotypes tells us little about disease activity or progression risk, whereas oncogene expression levels directly correlate with pathological changes.

The current paradigm of HPV DNA testing fails on a crucial front—it cannot distinguish between transient infections that will resolve spontaneously and transformative infections with high oncogene expression that will progress. Our transcriptomic approach demonstrates that E6/E7 mRNA quantification provides critical information about disease activity that DNA testing fundamentally cannot capture. The presence of HPV DNA without significant E6/E7 expression likely represents a biologically distinct and clinically benign disease state compared to infections with active viral oncogene transcription, explaining why approximately 90% of HPV infections resolve without progression while a minority advance to cancer.

Our comprehensive transcriptomic analysis further revealed that the molecular cascade initiated by E6/E7 expression—including growth factor receptor activation, immune evasion, and metabolic reprogramming—provides far more clinically relevant information than simply detecting viral DNA. This functional readout of viral oncogene activity and host response offers the precision needed for truly personalized risk assessment and intervention strategies that current DNA-based testing methods cannot achieve.

4.5 Viral DNA integration

The integration of high-risk human papillomavirus (HPV) DNA into the host genome is a pivotal event in the oncogenic transformation of cervical epithelial cells, marking the transition from persistent infection to precancerous and ultimately invasive carcinoma. Integration typically disrupts the viral E2 regulatory gene, leading to deregulated, persistent overexpression of E6 and E7, inactivation of p53 and Rb, genomic instability, and unchecked proliferation. In our study, although integration events were not directly sequenced, we experimentally verified active viral transcription by RT-qPCR quantification of E6/E7 mRNA in HPV-positive samples, supporting the functional consequence of E2 dysregulation and aligning with the integration-associated transcriptomic changes observed in HSIL. Consistent with E6/E7-driven effects, HSIL showed hyperactivation of cell-cycle/replication programs (e.g., CDK1, PLK1, MCM3–6, RFC3/4; log2FC>2, p < 0.001) together with induction of DNA damage response pathways, including ATR/ATM signaling (hsa176187, FDR = 2.22e-9), G2/M checkpoint control (hsa69473), homologous recombination, and mismatch repair, reflecting a cellular environment attempting to manage oncogene-mediated replication stress. These alterations indicate chronic DNA damage and stalled forks that HPV exploits to sustain replication, ultimately fostering host genomic rearrangements (oncogene amplifications, tumor-suppressor disruptions). This integration-linked HSIL profile contrasts with the largely productive, episomal viral state in LSIL, where minimal pathway overlap (OC = 0) suggests limited integration and instability. HSIL enrichment in DNA repair/replication stress responses (WP707, FDR = 0.0021) and mitotic dysregulation underscores integration as a key trigger of malignant progression, potentially facilitated by inflammation-derived reactive oxygen species and chromatin state changes. Our findings, in line with prior reports [23,24,62–64] highlight integration as a hallmark of transformative infection in HSIL and point to biomarker opportunities targeting integration sites or downstream effects (e.g., oncogenic fusions) for risk stratification and early intervention in HPV-driven cervical pathogenesis.

4.6 Transcriptomics: revolutionary advancement beyond current ”stone age” diagnostic tools

Current cervical cancer screening relies primarily on cytology and HPV DNA testing—approaches that detect abnormal cell morphology and viral presence but fail to capture the complex molecular dynamics driving disease progression. These methods represent ”stone age” tools in the era of precision medicine, unable to distinguish between transient infections and those on trajectory toward malignancy.

Our transcriptomic analysis represents a quantum leap in diagnostic capability by revealing the functional alterations that precede and drive morphological changes. Unlike conventional methods, transcriptomics captures the dynamic interplay between viral oncogene expression, host immune response, and cellular transformation pathways.

This approach provides not just diagnostic information but offers prognostic insights by identifying active driver events that promote progression.

The remarkable molecular distinctiveness of HSIL, with minimal pathway overlap with earlier stages (OC = 0.067 with NILM, OC = 0.043 with ASCUS, OC = 0 with LSIL), demonstrates how transcriptomics can delineate disease states with unprecedented precision. Such molecular stratification could revolutionize clinical management by identifying patients requiring aggressive intervention versus those who can be safely monitored.

4.7 Clinical implications: new era of molecular risk stratification

The identification of stage-specific molecular signatures has profound implications for cervical cancer prevention and management. Current approaches treat all HPV-positive patients within a cytological category as having equivalent risk, leading to both over-treatment and under-treatment.

Our transcriptomic profiling enables molecular risk stratification within each cytological category. For example, HPV-positive NILM patients with activated EGFR/ERBB2 signaling and downregulated immune mediators might warrant more aggressive surveillance than those without these signatures. Similarly, ASCUS cases with oncogenic driver activation (CCND1, SHH) might benefit from early intervention rather than surveillance.

The early driver events we identified represent potential biomarkers for progression risk assessment and targets for preventive intervention. Small molecule inhibitors targeting EGFR/ERBB2 signaling, immunomodulatory agents that restore immune detection, or metabolic modulators that target enhanced protein synthesis could potentially disrupt the carcinogenic cascade before irreversible genomic destabilization occurs.

4.8 Future directions: pap smear transcriptomics as the new frontier

Our successful transcriptomic profiling of routine Pap smear specimens demonstrates the feasibility of incorporating molecular analysis into standard clinical practice. Future developments should focus on translating these comprehensive signatures into clinically applicable biomarker panels that can be readily implemented in screening programs.

Prospective studies are needed to validate the predictive value of these molecular signatures for progression across the spectrum of HPV-induced lesions. Integration of transcriptomic data with other molecular features (methylation, microRNA profiles) could further refine risk prediction and therapeutic targeting. The dramatic molecular discontinuity between LSIL and HSIL warrants further investigation to identify the critical molecular switches that mediate this transition. Understanding these pivotal events could reveal novel targets for preventing progression to high-grade lesions.

Our findings call for a paradigm shift in how we conceptualize HPV-induced carcinogenesis, not as a continuous progression but as a series of distinct molecular states with critical thresholds and driver events that can be targeted for prevention and early intervention.

4.9 Study limitation

Although variations in group sizes and the absence of experimental validation may introduce certain constraint, the integrated transcriptomic approach applied in this study provides valuable insights into the molecular mechanisms underlying cervical lesion progression. Future work will extend these findings through immunohistochemical, proteomic, and in vivo validations of representative hub proteins to substantiate their biological significance and clarify their roles in disease pathogenesis. While our network analyses integrated transcript-to-protein associations inferred from high-confidence STRING interaction scores (≥0.7), future work will include experimental protein-level confirmation. Key HSIL hub genes (CDK1, PLK1, EZH2, MCMs) and early-lesion immune markers will be validated through immunohistochemistry (IHC) or multiplex immunoassays in a prospective cohort. These studies will establish RNA-to-protein concordance and strengthen the translational relevance of the identified molecular networks. Future integration of targeted proteomic validation will further corroborate the transcriptome-derived network signatures of cervical lesion progression.