Fig 1.
Schematic flowchart of the integrated network toxicology and molecular docking study on paraben toxicity in HNSCC.
Fig 2.
Identification of common candidate targets for paraben toxicity in HNSCC.
A three-way Venn diagram illustrates the overlap among three distinct datasets: 582 potential paraben-related targets from Swiss Target Prediction, ChEMBL, and the Similarity Ensemble Approach (SEA); 7,320 known HNSCC-associated genes from GeneCards, Online Mendelian Inheritance in Man (OMIM), and Comparative Toxicogenomics Database (CTD); and 2,250 differentially expressed genes (DEGs) between HNSCC tumors and normal tissues from The Cancer Genome Atlas (TCGA) cohort. The intersection of all three sets reveals 80 common candidate genes, which are hypothesized to be critically involved in the potential toxicological mechanisms of parabens on the development and progression of HNSCC.
Fig 3.
Functional enrichment analysis of differentially expressed genes (DEGs) in the TCGA-HNSCC cohort.
A circular enrichment plot illustrates the top 10 most significantly enriched terms (by p-value) from Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and REACTOME pathway analyses for the HNSCC-DEGs. In these plots: the outermost circle lists the significantly enriched term IDs, with different colors representing different categories. The second circle represents the total number of background genes that enriched in each item. The third circle represents the genes in DEGs list that enriched in the target term. The innermost circle represents the rich factor, which is the ratio of enriched genes to total background genes, reflecting the degree of enrichment of the target pathway.
Fig 4.
Identification of hub genes via MCODE and CytoHubba analysis.
(A) A core protein-protein interaction (PPI) subnetwork was extracted using the MCODE algorithm, with the highest MCODE score (19.905), containing 22 nodes and 418 edges, indicating a highly interconnected protein cluster. (B) An upset plot shows the overlap of the top 20 genes ranked by each of the 12 CyroHubba algorithms. Horizontal bars represent the frequency of each gene across all methods. From this analysis, 12 hub genes (CCNB1, CDK1, CCNA2, CDK2, CDK4, TYMS, AURKA, CCNA1, CHEK1, CCNB2, PLK1, CDC25A) were not only highly ranked but were also all members of the core MCODE subnetwork, confirming their status as central hub genes for further investigation.
Fig 5.
Validation of hub gene expression in HNSCC using independent GEO datasets.
The mRNA expression levels of the identified hub genes were analyzed in three independent HNSCC cohorts from the Gene Expression Omnibus (GEO) repository. (A) GSE83519 (22 paired tumor and adjacent normal tissues), Note: A constant value (C = 3.911890567) was added to the raw expression data prior to log₂ transformation to eliminate negative values. This constant value was calculated as |min (negative value)| + 0.001 to enable transformation of negative values; (B) GSE58911 (15 paired samples); (C) GSE160042 (10 paired samples). Differential expression between tumor and matched adjacent normal tissues is shown for each dataset. Statistical significance was determined by paired t-test (p < 0.05).
Fig 6.
Multivariable Cox regression analysis of the 12 hub genes for prognostic prediction in HNSCC.
A forest plot presents results of multivariable Cox proportional hazards regression model incorporating all 12 candidate hub genes. Each gene is shown with its hazard ratio (HR) and 95% confidence interval (CI). CCNA1 was identified as a significant independent predictor of poor overall survival in months (adjusted HR = 1.097, 95% CI: 1.021–1.179, p = 0.012). No other hub genes showed significant independent prognostic value.
Fig 7.
Functional enrichment analysis of the 12 hub genes.
Circular enrichment plots visualize the top 10 most significantly enriched terms from (A) Gene Ontology (GO) and (B) Kyoto Encyclopedia of Genes and Genomes (KEGG) and REACTOME pathway analyses for the 12 hub genes.
Fig 8.
Association between CCNA1 expression and immune cell infiltration levels in TCGA-HNSC data.
Correlation between CCNA1 expression (log2(TPM + 1)) and infiltration levels of CD8 + T cell and B cell was analyzed across TCGA-HNSC tumors (n = 522) using TIMER2.0. Infiltration levels were estimated with six algorithms (EPIC, MCP-COUNTER, TIMER, CIBERSORT, QUANTISEQ, and XCELL) and adjusted for tumor purity. The association was evaluated using partial Spearman’s correlation analysis. Spearman’s ρ > 0 and p < 0.05 indicates significant positive correlation; ρ < 0 and p < 0.05 indicates significant negative correlation; p > 0.05 indicates no significance. TPM: transcripts per million. The analysis was performed by TIMER2.0 based on TCGA-HNSC data and detailed analysis can be further obtained from TIMER2.0.
Fig 9.
Molecular docking analysis of paraben derivatives binding to hub proteins.
A structure-based blind protein-ligand docking strategy was performed using CB-DOCK2 to predict binding affinities between six paraben derivatives and 12 hub proteins. Results are presented as a heatmap, with the binding energy scores (Vina score, in kcal/mol) for each compound-protein pair indicated numerically. More negative scores represent stronger predicted binding affinity.
Fig 10.
Sankey diagram of paraben-hub gene-pathway interactions.
The diagram summarizes the proposed mechanistic links between paraben exposure (left), hub genes (middle), and enriched biological pathways (right), illustrating how parabens may influence HNSCC pathogenesis through core regulatory genes.