Identification and mechanistic exploration of VEGFA and CALML3 as calcium channel-related prognostic genes in cervical cancer

Abstract

Cervical cancer is the fourth most common cancer among women worldwide. Calcium ion channel-related genes (CICRGs) play an important role in the proliferation, differentiation, migration and angiogenesis of cervical cancer (CC). However, the diagnostic potential of calcium channel-related genes in cervical cancer remains underexplored. This study combined the TCGA-CC, GSE9750, and GSE63514 datasets with gene set map04020 to identify differentially expressed genes (DEGs). Weighted gene co-expression network analysis (WGCNA) was used to filter key module genes, followed by enrichment analysis and protein-protein interaction (PPI) network construction. Candidate prognostic genes were identified through three machine learning methods, with expression verified and survival differences assessed via Kaplan–Meier (KM) curves. Gene Set Enrichment Analysis (GSEA) correlated these genes with tumor-related processes. Additionally, immunological analysis and a ceRNA network were constructed. A total of 30 candidate genes were identified by intersecting 5089 DEGs, 3417 key module genes, and 240 CICRGs, and a PPI network was constructed. Expression verification and survival analysis identified two prognostic genes, VEGFA and CALML3. GSEA revealed VEGFA was positively correlated with tumor-related processes, while CALML3 was negatively correlated. Immunological analysis confirmed CALML3 as an effective indicator of immune cell infiltration. Finally, the drug-disease network revealed three disease processes associated with CALML3 and VEGFA, and the drug–gene network highlighted bisphenol A, tetra chlorodibenzo dioxin (TCDD), and Estradiol as co-acting agents of the prognostic genes. Immunohistochemistry showed higher expression levels of CALML3 and VEGFA in CC tissues, suggesting their potential as prognostic biomarkers. This study not only identifies key prognostic genes but also provides valuable insights into the molecular mechanisms of cervical cancer, offering potential therapeutic targets and biomarkers for prognosis and immune infiltration, thus contributing to the advancement of personalized treatment strategies for CC. The novelty of this study lies in the systematic integration of the KEGG calcium signaling gene set (map04020) with multi-cohort transcriptome data using a prior-constrained, multi-algorithm consensus workflow, which enhances robustness and reduces false positives. Moreover, we demonstrate the immune-infiltration relevance of CALML3 and develop a high-performance two-gene nomogram together with regulatory and drug-association networks, offering new clues for calcium-channel-related prognostic assessment and therapeutic exploration in cervical cancer.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-026-04683-0.

Introduction

Cervical cancer (CC) is the fourth most common cancer among women worldwide [1]. According to the data released by China’s National Cancer Center on the JNCC in February 2024, China reported roughly 150,700 new CC cases and approximately 55,700 deaths from the disease in 2022. A primary “culprit” of CC is persistent HPV infection [2]. Furthermore, CC is also linked to other factors, including sexual activity at an early age, having multiple sexual partners, prolonged contraceptive use, smoking, persistent bacterial or viral infections, and a weakened immune system [3, 4]. Cervical cytology is the primary means of initial screening for CC.

Further assessment of CC risk is conducted based on medical history and HPV DNA testing. For individuals with abnormal cervical cytology and HPV test results, doctors will recommend a colposcopy and biopsy of suspicious areas for pathological examination to confirm the presence of cancer cells, which is the gold standard for diagnosing CC. Imaging studies such as transvaginal ultrasound, CT, and MRI can be used to determine whether distant metastasis has occurred, providing a basis for clinical staging and treatment planning [5, 6]. While current screening techniques are capable of identifying precancerous cervical lesions and early CC to some degree, misdiagnoses remain unavoidable, particularly for specific histological types like undifferentiated and clear cell carcinomas. Although cervical biopsy is the gold standard for diagnosing CC, it may still result in missed or misdiagnosed cases due to limitations in equipment, technology, and physician experience. Current treatments primarily consist of surgery, radiotherapy, and chemotherapy. In recent years, novel treatment approaches like targeted therapy and immunotherapy have increasingly been adopted in clinical settings, they are still in the exploratory stage and face issues such as limited indications and high costs [6–8]. Researching calcium ion channel genes related to CC holds far-reaching significance for the treatment of CC. It helps uncover the pathogenesis of CC but provides new therapeutic targets and optimizes treatment regimens. It also promotes the development of CC research, improving patients’ survival rates and quality of life.

Calcium ion channels are glycoproteins that span the cell membrane and are found ubiquitously in various tissue cells throughout our bodies. They are primarily responsible for regulating the flow of calcium ions within the body. This diverse family of channels includes voltage-gated calcium channel, ligand-gated calcium channels, transient receptor potential channels (TRPs), store-operated calcium channels, and arachidonate-regulated calcium entry channels, among others [9]. The concentration of intracellular Ca²⁺ is related to proliferation, invasion, and metastasis of cancerous tissues. The homeostasis and signaling of intracellular Ca²⁺ are regulated by an array of Ca²⁺ channels, pumps, exchangers, sensors, and effectors activated by Ca²⁺. Modifications in the expression, structure, and functionality of these proteins can elicit detrimental consequences [10]. Studies indicate a strong correlation between Ca²⁺ channels and the onset, progression, as well as the prognosis of CC [10]. Plasma membrane Ca²⁺ channels include intermediate-conductance Ca²⁺-activated K⁺ channels (IKCa1), TRPs, and Ca²⁺ release-activated Ca²⁺ channels (CRACs). Research results indicate that the higher the malignancy of CC tissues, the higher the expression of IKCa1. IKCa1 contributes to the dedifferentiation and proliferation of malignant CC cells [11]. TRPs are one of the extracellular Ca²⁺ influx channels, encompassing seven subfamilies: TRPV (vanilloid), TRPM (melastatin), TRPC (canonical), TRPP (polycystin), TRPML (mucolipin), TRPA (ankyrin), and TRPN (NOMPC-like) [11]. Research has indicated that TRPV1 serves as a key prognostic indicator for the survival of CC patients, exhibiting significantly elevated expression levels in CC tissues when compared to both cervical intraepithelial neoplasia and normal epithelial tissues [12]. TRPM4 is highly expressed in CC samples compared to normal cervical epithelial tissues [13]. TRPV6 emerges as a potential new prognostic indicator for early-stage CC, given its markedly decreased mRNA and protein levels in tissues and cell lines of early-stage cervical squamous cell carcinoma. Additionally, TRPM7 is targeted by miR-543 in CC, and reintroducing TRPM7 expression can partly counteract the tumor-inhibiting actions of miR-543 on CC cells [14]. CRAC channels are made up of the calcium release-activated calcium modulator 1 and stromal interaction molecules (STIM) [15]. STIM1 is crucial for the proliferation, migration, and angiogenesis of CC cells. When STIM1 is silenced in CC cells, it markedly suppresses cancer cell growth. Conversely, overexpressing STIM1 promotes the invasion of CC cells, whereas knocking out STIM1 diminishes their migratory capacity [16]. Cells contain endoplasmic reticulum Ca²⁺ release channels, primarily comprising inositol 1,4,5-triphosphate receptor (IP3R3) and ryanodine receptor channels. Research has indicated that genetic variations in IP3R3 are linked to the risk of developing CC [17]. Furthermore, Inositol 1,4,5-trisphosphate 3-kinase C inactivates IP3R3, leading to the suppression of the IP3 pathway and a reduction in the Ca²⁺ signaling cascade [18]. Investigating genes related to Ca²⁺ channels may provide new biomarkers for the diagnosis of CC and open up new targeted therapeutic approaches in treatment, which is of crucial significance.

This study employed bioinformatics methods to mine CC diagnostic genes related to calcium ion channel genes based on transcriptome data, and constructed a prediction model based on these diagnostic genes, providing a new method for clinical diagnosis of cervical cancer. Additionally, functional enrichment analysis, immunological analysis, and drug prediction analysis were conducted on these diagnostic genes, aiming to provide theoretical basis for further exploring the molecular mechanisms and treatment strategies of cervical cancer. The innovation of this study lies in developing a new bioinformatics method by combining the relationship between calcium ion channel genes and cervical cancer to screen diagnostic genes, and successfully constructing a prediction model with practical application value. Meanwhile, through systematic functional analysis and drug prediction, this study provided new insights into the molecular mechanisms and therapeutic targets of cervical cancer, promoting the application of precision medicine in cervical cancer.

Bioinformatics approaches have been widely applied in cervical cancer research. Previous studies have developed pyroptosis-related prognostic models using algorithms such as LASSO regression [19], and multiple investigations have focused on the immune microenvironment of cervical cancer and performed machine-learning–based analyses [20, 21]. However, systematic studies of calcium channel–related genes in cervical cancer remain limited, and reliance on a single algorithm has inherent constraints for gene selection. Therefore, we integrated weighted gene co-expression network analysis (WGCNA) with three machine-learning algorithms (SVM-RFE, LASSO, and Boruta) to systematically identify calcium channel–related prognostic genes from transcriptomic data and to construct a predictive model, providing a novel strategy for the clinical diagnosis of cervical cancer. We further investigated the molecular mechanisms and therapeutic potential of these prognostic genes through functional enrichment analyses, immune infiltration profiling, and drug response prediction, thereby offering new theoretical evidence for mechanistic studies and targeted therapy in cervical cancer.

Materials and methods

Data collection

Gene expression sequencing data, as well as clinical information in TCGA-cervical cancer, were sourced from the Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/), which covered eight peritumoral tissues and 296 primary cervical tumor tissues. The TCGA cervical cancer data, due to its comprehensive and high-quality data, large sample size, inclusion of tumor and normal tissue comparisons, integration with clinical data, and public accessibility, provides significant support for research on the molecular mechanisms of cervical cancer, prognostic assessments, and the discovery of therapeutic targets. The microarray sequencing data for datasets GSE9750 and GSE63514 originated from the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/). Detailedly, the GSE9750 dataset was a training set containing 33 CC epithelial tissue samples and 24 cervical epithelial tissue (control) samples [22]. The GSE63514 dataset contained 28 CC and 25 control epithelial tissue samples to act as a validation set [23]. These two datasets (GSE9750 and GSE63514) contained a sufficient number of cervical cancer patient and normal control samples, were publicly available, and had reliable data quality. They provided an adequate sample size and diversity for the analysis, ensuring the stability and reproducibility of the results. The dataset selection process is shown in Supplementary Fig. 1. The calcium ion channel gene set (map04020) was sourced from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (https://www.genome.jp/kegg/).

Differential expression and weight gene co-expression network analysis (WGCNA)

Differentially Expressed Genes (DEGs) between control and CC samples from the GSE9750 database were filtrated (|log2FC|≥0.5 and p < 0.05) by R package ‘limma’ (version 3.52.4) [24]. The R packages ‘DEGsgplot2’ (version 3.3.6) [25] and ‘heatmap3’ (version 1.1.9) [26] were employed to make volcano maps and heat maps to visualize DEGs. In order to obtain genes associated with CC, using the R package ‘WGCNA’ (version 1.71) [27], WGCNA was conducted based on the GSE9750 dataset. Firstly, clustering analysis was employed to “remove sample outliers.” Subsequently, the most suitable soft threshold power was chosen to ensure high generalization and cohesion among modular genes. Then, WGCNA was created through dynamic tree cutting. With the aim of finding crucial module genes strongly associated with CC, an analysis was conducted to investigate the correlation between modular structures and CC traits, and the genes within the significant modules related to CC were determined as essential module genes.

Screening and enrichment analysis of candidate genes

The candidate genes were screened by intersecting DEGs, key module genes, and CICRGs. To gain insights into the biological roles and signaling pathways of the identified candidate genes, we employed the R package ‘clusterProfiler’ (version 4.7.1) to perform Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses [27], setting a significance level at p < 0.05. Furthermore, we constructed a PPI network using the STRING database (https://cn.string-db.org/) to investigate the molecular interactions among the candidate genes, considering only interactions with a confidence score exceeding 0.4. To avoid target generalization inherent to purely data-driven pipelines, we implemented a prior-constrained design in which candidate genes were first restricted to the calcium signaling gene set (KEGG map04020). This constraint ensures that downstream screening and validation focus on genes with direct functional relevance to calcium-channel signaling, rather than producing broadly cancer-associated markers without pathway specificity.

Identification and verification of candidate prognostic genes for CC

Three types of machine learning approaches were utilized to identify prognostic genes that could precisely predict the progression of CC. We implemented a support vector machine-recursive feature elimination (SVM-RFE) framework using the R package ‘e1071’ (version 1.7–11) [28], which allowed for the iterative removal of genes from a pool of candidates, thereby refining our selection of characteristic genes. Samples with missing values and genes with zero expression across all samples were removed. Feature importance was evaluated using 10-fold stratified cross-validation, where the lowest 50% of features were eliminated in each iteration. The model with the minimum error was selected, and its ranking was output using the WriteFeatures function. Meanwhile, the dimensionality reduction of the candidate genes was achieved by applying the Least Absolute Shrinkage and Selection Operator (LASSO) algorithm via the ‘glmnet’ package in R software (version 4.1.8) [29]. Subsequently, 10-fold cross-validation was used to train the model. The optimal alpha value was selected at the minimum standard error (lambda.min), which resulted in the model having the maximum generalization performance. The important features (genes with non-zero weights) in the model at this point were selected as the feature genes identified by LASSO. Characteristic genes related to CC were subsequently pinpointed utilizing the LASSO methodology. In addition, the Boruta algorithm was also performed by constructing shadow features and random forest voting, respectively, with the ‘Boruta’ package in R (version 8.0.0) [30] to acquire characteristic genes. Afterward, the candidate prognostic genes for CC were determined through the intersection of characteristic genes obtained from the three distinct algorithms. Last but not least, to ensure that candidate prognostic genes acquired above influenced the occurrence and development of CC, the expression levels of the identified candidate prognostic genes were independently validated in both the GSE9750 and GSE63514 datasets.

Acquisition of prognostic genes and construction of nomogram model

Survival analysis was performed on the candidate prognostic genes in the TCGA-CC dataset. First, the FPKM expression data was preprocessed, including gene ID conversion, averaging the expression of duplicate genes, and filtering out normal samples. Clinical data was integrated with the expression data by matching sample IDs. The expression levels of each gene were divided into high and low expression groups based on the median value. The Kaplan–Meier (KM) curve was used to analyze the overall survival differences between the two groups. Survival curves were assessed for significance using the log-rank test (p < 0.05 considered significant), and survival probabilities over a 15-year follow-up period were calculated. Prognostic genes with obvious differences in survival rates between the groups with high and low expression levels were chosen for subsequent analysis. Subsequently, the expression data of prognostic genes was extracted from the expression matrix. After dividing the samples into Case and Control groups, a design matrix was constructed and merged into the final analysis dataset. A binary logistic regression model was built using the lrm function from the rms package, with the maximum number of iterations set to 1000 to ensure model convergence. The model used the group status (group) as the dependent variable and the expression levels of the prognostic genes as the independent variables. Based on the results of the logistic regression model, a nomogram model has been created to predict the prognostic efficacy of prognostic genes for CC through the R package ‘glam’ (version 8.0.0) [31]. The predictive performance of the nomogram model was assessed by employing a calibration curve, a decision curve analysis (DCA), and a receiver operating characteristic (ROC) curve. ROC analysis was performed using the pROC package (version 1.18.0, https://xrobin.github.io/pROC/) to evaluate the diagnostic performance of the nomogram. The roc function was used to calculate the ROC curve of the nomogram, and the area under the curve (AUC) along with its 95% confidence interval was computed. An AUC value greater than 0.6 indicated good diagnostic performance of the model.

Gene set enrichment analysis (GSEA)

GSEA was performed to investigate further biological functions and differences in signaling pathways related to prognostic genes through the ‘GSEA’ function in R (p < 0.05). The association between prognostic genes and other genes in GSE9750 was first computed and ranked. Afterward, the C2: KEGG signaling pathway set (homo sapiens) obtained through R package ‘msigdbr’ (version 7.5.1) [32] was utilized as the background set for conducting GSEA on the ranked genes. The Benjamini-Hochberg (BH) method was used for multiple testing correction during the analysis. Pathways with a corrected p-value (adj.p) < 0.05 were selected as significantly enriched pathways.

Immunological analysis of CC

In order to understand the immune infiltrating landscape of CC, CIBERSORTx was utilized to compute the proportions of 22 distinct immune cell types in each cervical epithelial tissue sample from the GSE9750 dataset, following immune cell enrichment, in order to quantify their abundance and the samples exhibiting significant infiltration (p < 0.05) were selected for further analyze. The Wilcoxon test was utilized to compare the infiltration levels of 22 distinct immune cell types between the CC and the control groups. In addition, the correlation between prognostic genes and immune cell enrichment scores was analyzed by Pearson analysis to determine whether prognostic genes could indicate immune cell infiltration in CC (cor ≥ 0.3, P < 0.05).

Construction of competing endogenous RNAs (ceRNA) network

A ceRNA (lncRNA-miRNA-mRNA) network was synthesized to profoundly explore prognostic genes’ expression regulation patterns during the occurrence and progression of CC. Primarily, the pivotal miRNAs were identified by intersecting miRNAs targeting prognostic genes (miRWalk database, http://129.206.7.150/, and Targetscan database https://www.targetscan.org/vert_80/) and miRNAs related with CC (miRNet database, https://www.mirnet.ca/). The Starbase database (https://rnasysu.com/encori/) was employed to forecast lncRNAs targeting crucial miRNAs. Last, Cytoscape software (specifically version 3.9.1) [33] was utilized to visualize a ceRNA network, highlighting the top five lncRNAs associated with each miRNA.

Disease association analysis and drug prediction based on prognostic genes

Examining the disease progression and medications linked to the prognostic genes facilitated a deeper comprehension of their function in CC. The CTD database (http://ctdbase.org/) was employed to forecast diseases related to drugs to prognostic genes, and drug-disease and drug–gene networks were visualized by Cytoscape software.

10 immunohistochemistry analysis (IHC)

An IHC analysis was performed to assess the expression levels of CALML3 and VEGFA genes in CC tissues. The TMA sections underwent initial deparaffinization and rehydration processes. Heat-induced antigen retrieval for CALML3 and VEGFA was carried out in citrate acid buffer (pH 6.0) for 15 min. The sections were allowed to incubate with a CALML3 Polyclonal antibody (1:50 dilution, 117275-1-AP, Proteintech) and VEGFA polyclonal antibody (1:50 dilution, 19003-1-AP, Proteintech) at 4 °C overnight. Afterward, the slides underwent incubation with Enhanced Enzyme-Labeled Goat Anti-Mouse/Rabbit IgG Polymer for 20 min at a temperature of 37 °C. Following this, 3,3′-diaminobenzidine chromogenic staining and nuclear counterstaining with hematoxylin were carried out. Human liver cancer tissue was the positive control for CALML3, while human lung cancer tissue was the positive control for VEGFA. PBS-substituted primary antibodies were considered as the negative control. The slice was entirely scanned with a digital scanner, and then the software Image-Pro Plus 6.0 (Media Cybernetics) was leveraged to measure the regions stained positively for CALML3 and VEGFA in the slice.

Statistical analysis

Statistical analysis was performed using R software (https://www.r-project.org/). The Wilcox test was used to compare the two groups, and a P-value below 0.05 was considered indicative of a significant difference.

Results

Identification of DEGs and key module genes

Among the 5,089 DEGs between CC and normal samples, 1,212 were up-regulated and 3,877 were down-regulated (Fig. 1A and B). Clustering analysis manifested that there were no outliers or abnormal samples in the GSE9750 dataset (Fig. 1C). The optimal soft threshold power (β) was determined to be 12 when R2 reached 0.8, with the mean connectivity approaching zero indefinitely (Fig. 1D). The co-expression network revealed the identification of a total of five modules (Fig. 1E). The correlation analysis between five modules and CC indicated that MEyellow, MEturquoise and MEblue were markedly correlated with CC (|Cor|>0.30 and P < 0.05). In addition, a further screening process identified 3417 genes within key modules that exhibited significant associations with CC under the conditions of |GS|≥0.30 and p ≤ 0.05. (Fig. 1F).

Fig. 1.

Identification of differentially expressed genes (DEGs) and module genes. (A) The volcano plot displays DEGs (P < 0.05, |log2FC| ≥ 0.5): red for up-regulation, blue for down-regulation, and grey for non-significance. (B) The heatmap showcases the 20 most notably up- and down-regulated DEGs from mRNA microarrays. The upper panel displays gene distribution, while the lower panel depicts gene expression. The x-axis represents samples, and the y-axis represents DEGs. The color intensity reflects normalized gene expression levels, with red signifying high expression and blue signifying low expression. “Sample” denotes the categorization of samples. (C) Dendrogram of dissimilarity between samples. (D) The graph on the left depicts the scale-free fitting index β analysis across different soft threshold powers. The right graph represents the average connectivity analysis for various soft threshold powers. (E) A dendrogram of genes was displayed and generated through clustering analysis based on (1-TOM) dissimilarity. Each branch within the dendrogram corresponds to a distinct gene. (F) The correlation between gene modules and sample phenotypes is illustrated, with each line providing the respective correlation coefficient and associated p-value

Screening and function analyses of 30 candidate genes

The number of candidate genes was 30 according to the Venn plot (Fig. 2A). According to the GO analysis, the candidate genes were found to be enriched in 846 categories, among which 54 belonged to cellular components such as pore complex, neuronal dense core vesicle and plasma membrane raft and the like, 78 molecular functions such as G protein-coupled amine receptor activity, phospholipase C activity and monoatomic ion-gated channel activity and so forth, and 714 biological processes including calcium ion transmembrane transport, calcium ion transmembrane import into the cytosol and positive regulation of cytosolic calcium ion concentration and the like (Fig. 2B). Besides, candidate genes were enriched into 88 KEGG pathways, including calcium signaling pathway, which was associated with a calcium ion channel, as well as cGMP − PKG signaling pathway and so on (Fig. 2C). To investigate the protein-level interactions among candidate genes, we built a PPI network comprising 28 nodes and 76 edges. Besides, the paired interactions covered CALML3-CAMK2A, CAMK2A-NOS1, and CALML3-PLCD1, etc.( Fig. 2D).

Fig. 2.

Enrichment analysis of candidate genes. (A) The intersected genes of DEGs, key module genes, and calcium ion channel-related genes (CICRGs). (B) Gene Ontology (GO) analysis of calcium ion channel proteins related to cervical cancer (C) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis of calcium ion channel proteins related to cervical cancer. BP is the biological process; CC is the cellular component; MF is the molecular function. (D) Visualization of the protein-protein interaction (PPI) network

A total of eight candidate prognostic genes for CC were screened

Three machine learning methods were used to filter candidate prognostic genes based on candidate genes. The SVM-RFE algorithm has been filtrated the nine characteristic genes (PLCB3, BDKRB2, ITPR2, ERBB2, VEGFA, ADRA1A, CALML3, ITPR3, and TBXA2R) (Fig. 3A). Several 11 characteristic genes (ADRA1A, VEGFA, ERBB2, TBXA2R, ITPR2, ITPR3, PLCD1, BDKRB2, PLCB3, CALML3 and AVPR1A) were screened by LASSO logistic regression algorithm (Fig. 3B, C). Besides, there were 17 characteristic genes (ADORA2A, ADRB2, BDKRB1, BDKRB2, CALML3, ERBB2, FGF22, ITPR2, ITPR3, PLCB3, PLCD1, PPIF, STIM1, TACR1, TBXA2R, VDAC2, and VEGFA) after screening through Boruta algorithm (Fig. 3D). Subsequently, several eight candidate prognostic genes for CC were identified by intersecting characteristic genes obtained from the three algorithms (Fig. 3E). Besides, the expression of eight candidate prognostic genes displayed that seven genes (VEGFA, CALML3, ITPR3, ITPR2, ERBB2, BDKRB2, and TBXA2R) were differentially expressed in GSE9750 and GSE63514. Their expression trends were consistent (Fig. 3F, G).

Fig. 3.

Feature gene screening using machine learning. (A) The relationship between SVM generalization error and feature count. Feature selection used the LASSO regression model through 10-fold cross-validation and lambda 1se. (B) Generate coefficient distribution plots for log(lambda) sequence. (C) LASSO non-zero coefficient 11 significant genes in TNBC. (D) Box plot of scores in all iterations of the Boruta algorithm. The scores of 17 essential genes are higher than the shadowmax score. (E) The intersected genes of these machine learning were selected. (F) Verification by the GSE9750 training set. (G) Verification by the GSE63514 Validation set

A nomogram model has been created on the basis of two prognostic genes

The KM curves revealed that only two genes (VEGFA and CALML3) of the eight candidate prognostic genes caused significant differences in patients’ survival time in TCGA-CC. Besides, poorer survival in CC patients was associated with higher expression of VEGFA and lower expression of CALML3 (Fig. 4A, B). Therefore, VEGFA and CALML3 were defined as prognostic genes. To construct a model for predicting the occurrence of CC, a nomogram model was set up with the support of two prognostic genes (Fig. 4C). The results evaluated the nomogram model via calibration curve, DCA curve, and ROC curve. The calibration curve showed that the slope of the calibration curve was close to 1 (Fig. 4D). The net benefit of the DCA curve was higher. The model demonstrated a greater net benefit compared to a single gene (Fig. 4E). Besides, the AUC value of the ROC curve was 0.961 (Fig. 4F). The results above demonstrated that the nomogram model possessed high prediction accuracy.

Fig. 4.

Survival analysis of transcriptome sequencing data and clinical data of cervical cancer samples in the TCGA database. (A) The KM curves of the VEGFA. (B) The KM curves of the CALML3. (C) Nomogram. On the left are the diagnostic factors, each corresponding to a score. The total score is obtained by summing up the scores of all factors (Total Points), and then the survival rate is predicted based on the total score. (D) Nomogram calibration curve. The abscissa represents the predicted survival rate by the nomogram, and the ordinate represents the actual survival rate. (E) DCA curve. The horizontal axis represents the High-Risk Threshold range, and the vertical axis represents the Net Benefit. (F) ROC curve of the diagnostic model

Prognostic genes were closely linked to a variety of tumor-related processes

To delve deeper into the potential role of prognostic genes in differentiating CC samples from normal ones, we performed a GSEA-KEGG pathway analysis for each of the two prognostic genes individually. The TOP5 pathways enriched for each prognostic gene are illustrated in Fig. 5A and B. VEGFA had a significant positive correlation with cell cycle, DNA replication, pyrimidine metabolism, etc., tumor-related processes, while CALML3 was negatively correlated with DNA replication, EMC receptor interaction, etc., tumor-related processes.

Fig. 5.

GSEA enrichment trend plot. (A) GSEA enrichment of VEGFA. Present only the five most enriched pathways, ordered by their enrichment score (ES), in a figure comprising three sections. The upper section illustrates the computation of the ES value. For each gene from left to right, an ES value is calculated and connected to form a line. On the leftmost side, a particularly prominent peak represents the ES value for the gene set phenotype. The middle part of the figure features each line representing a gene in the gene set and its ranking position in the gene list. The bottom part displays the matrix of gene-phenotype associations. (B) GSEA enrichment of CALML3

Prognostic genes had indicative effect on immune cell infiltration in CC

A comprehensive characterization of the immune cell infiltration landscape within the tumor microenvironment of cervical carcinoma epithelial tissue was conducted. The CIBERSORTx analysis demonstrated notable disparities in the proportions of immune cell infiltration between the disease and control groups. (Fig. 6A). The Wilcoxon test indicated that CD 4 naive T cells, M0 Macrophages, M1 Macrophages, and plasma cells were significantly enriched in CC, while CD8 T cells, CD4 memory resting T cells, resting dendritic cells, and resting M cells were reduced in CC (Fig. 6B). Moreover, to find out whether prognostic genes can indicate immune cell infiltration in CC, the spearman correlation analysis demonstrated that CALML3 was an almost significant positive correlation with Dendritic cells resting, T cells CD4 memory resting and Monocytes, while Macrophages M0, and T cells CD4 naive were significant negative correlated with CALML3, the correlation between VEGFA and immune cells was relatively low (cor ≥ 0.30, P < 0.05). This revealed that VEGFA had a weak indicative effect on immune cell infiltration in CC, but CALML3 was deemed an effective indicator of immune cell infiltration in CC (Fig. 6C).

Fig. 6.

Analysis of immune cell infiltration. (A) The composition of the immune cell infiltrate. (B) The relative percentage of 22 immune cells in each training set samples (GSE9750). The box color represents the sample grouping; the top indicates significance. * P ≤ 0.05, ** P ≤ 0.01, *** P ≤ 0.001,**** P ≤ 0.0001,NS: no significance. (C) Pearson correlation analysis between key diagnostic genes and the enrichment scores of immune cells

A CeRNA network was synthesized on the basis of CALML3 and VEGFA

A total of seven pivotal miRNAs were identified by intersecting 2013 miRNAs targeting prognostic genes (CALML3 and VEGFA) and 120 miRNAs related with CC (Fig. 7A). In addition, the Starbase database was used to obtain 1640 lncRNA. There are 29 nodes and 30 edges in the lncRNA-miRNA-mRNA regulatory network. The nodes included two prognostic genes (CALML3 and VEGFA), five miRNAs (hsa-miR-107, hsa-miR-326, hsa-miR-133b, hsa-miR-429 and hsa-miR-218-5p) and 22 lncRNAs (MIATNB, AC022400.6 and APTR, etc.). The ceRNA network composed of them included CALML3-hsa-miR-326-TMEM147-AS1, VEGFA-hsa-miR-107-MIATNB, and other relationship pairs, besides, a sum of three miRNAs (hsa-miR-107, hsa-miR-326 and hsa-miR-133b) could regulate two prognostic genes at the same time (Fig. 7B).

Fig. 7.

Construction of lnRNA-miRNA-mRNA network in key diagnostic genes. (A) A Venn diagram illustrates the mRNAs participating in the ceRNA network, where the red region denotes the count of miRNAs derived from the overlap of miRWalk and Targetscan databases, the blue region depicts the quantity of cervical cancer-associated miRNAs, and the central purple region signifies the number of miRNAs incorporated into the ceRNA network. (B) lnRNA-miRNA-mRNA ceRNA network in key diagnostic genes. The red ones are key diagnostic genes, the purple ones are miRNAs, the blue ones are lncRNAs, and the lines indicate their interactions

Three major diseases process and drugs were linked to CALML3 and VEGFA

The TOP10 diseases associated with two prognostic genes are shown in Fig. 8A. The drug-disease network showed that three kinds of the disease process (Neoplasms, Inflammation, Chemical and Induced Liver Injury) co-associated with CALML3 and VEGFA. Based on the CTD database, the number of drugs related to CALML3 and VEGFA, respectively, were 23 and 838. The drug–gene network displayed drugs with the reference value (reference > 2) (Fig. 8B); thereinto, bisphenol A, tetra chloro dibenzo dioxin (TCDD) and estradiol were co-acting agents of prognostic genes.

Fig. 8.

Disease and drug prediction of key diagnostic genes based on the CTD database. (A) Prediction of disease correlation. The graph depicts genes as red nodes and drugs as blue nodes, with the thickness of the connecting lines indicating the reference count and the numerals on these lines denoting the specific number of references. (B) Gene-drug relationship network. The graph shows that the red ones are key diagnostic genes, and the yellow ones are related drugs

Expression of CALML3 and VEGFA genes in cervical cancer tissues

Five CC patients and normal cases were selected for immunohistochemical analysis. The expression levels of CALML3 and VEGFA genes were found to be elevated in CC tissues compared to normal cervical tissues (Fig. 9A). Using Imge-Pro Plus, we conducted a quantitative analysis of CALML3-positive and VEGFA-positive areas in whole-slice preparations and observed that the average density of CALML3 in CC was higher compared to the normal control. (t-test p-value = 0.034) (Fig. 9B) Moreover, the mean density of VEGFA in CC was larger than normal control (t-test p-value = 0.0306) (Fig. 9C).

Fig. 9.

Validation of CAML3 and VEGFA gene expression by IHC. (A) Immunohistochemistry analysis (IHC) staining results for normal controls and CC tissues are presented, with the top image originating from the cervical tissue of normal control and the bottom image derived from a CC patient. (B) CAML3 IHC staining reveals an increased mean density in CC whole slices compared to the normal control. (C) The IHC staining of VEGFA indicates that the average density in whole CC slices is elevated relative to the normal control. The top indicates significance. * P ≤ 0.05

Discussion

While our analysis builds upon established bioinformatics modules, the key advance of this work is the customized design combining “calcium-channel gene-set constraint + multi-cohort integration + multi-dimension validation + mechanism extension.” This design enabled us to identify two calcium-channel–related prognostic genes and to further connect them with immune infiltration patterns, predictive modeling, and regulatory/drug networks, thereby addressing gaps left by narrower studies focusing on single calcium-channel subfamilies.

Cervical cancer is a malignant tumor caused by persistent infection with high-risk types of HPV, posing a severe threat to women’s health and potentially leading to uterine damage, infertility, and various complications. Although current treatments have specific effects, issues of over-treatment or insufficient treatment still exist [2, 34, 35].

The role of calcium ion channel-related proteins in CC cannot be overlooked, as a specific set of proteins maintains cellular Ca²⁺ homeostasis and signal transduction. Abnormal expression of these proteins is often closely associated with the malignant progression of CC. We will delve into the molecular mechanisms by which Ca²⁺ channel proteins play a role in cancer, focusing on aspects such as Ca²⁺ channels, pumps, exchangers, sensors, and Ca²⁺-activated effectors [36]. The IKCa1 channel, alternatively termed IK1, SK4, hSK4, hKCa4, or KCa3.1, belongs to the KCa1 channel family and plays a crucial role in regulating the proliferation and migration of cancer cells [37]. Liu Ling and her team discovered that CC tissues exhibit elevated levels of IKCa1 expression. Furthermore, they demonstrated that IKCa1 channel blockers, such as clotrimazole and IKCa1-specific siRNA, effectively suppress the growth of HeLa CC cells [11]. ITPR3 is an intracellular Ca2 + release channel on the endoplasmic reticulum membrane that responds to the binding of the second messenger inositol IP3 [38]. Upon exposure to various extracellular stimuli, the phospholipase C pathway is activated, generating IP3, which then binds to IP3 receptors and triggers the release of Ca2 + from the endoplasmic reticulum (ER). According to Yuh-Cheng Yang (2017), there is a significant association between the AT haplotype in the ITPR3 gene and cervical squamous cell carcinoma risk in Taiwanese women [17]. SERCA, a pump transporting Ca2 + from the cytosol to the ER lumen, disrupts ER folding by depleting ER Ca2 + and elevating cytosolic Ca2+, increasing unfolded/misfolded proteins, activating ER stress, initiating the unfolded protein response pathway, and ultimately causing ER stress-related cell death [39]. W Li and others found that when HeLa cells are exposed to SBF-1, SERCA activity is inhibited [40]. S100 proteins, a large subgroup of the EF-hand protein family, are small calcium-binding proteins. Upon binding to Ca2+, S100 calcium-binding proteins undergo structural and functional alterations, functioning as Ca2 + sensors that translate intracellular Ca2 + fluctuations into cellular responses [41]. In their study, Meng Man and colleagues [37] observed that the overexpression of S100A11 enhances the proliferation, migration, invasion, and epithelial- mesenchymal transition (EMT) of CC cells.

Additionally, they identified the activation of Wnt/β-catenin signaling as a key mechanism in this process, which is known to play a pivotal role in regulating cancer cell growth, survival, invasion, and migration [38]. Wang X and co-authors [39] discovered a significant correlation between the overexpression of S100A14 and both the FIGO stage and lymph node metastasis in CC patients. In their study, Tian et al. [40] demonstrated that S100A7 triggers EMT and enhances CC cells’ migration, invasion, and metastatic potential. Overexpression of S100A7 may be a useful marker for estimating the risk of cervical dysplasia progressing to malignancy. Studying the regulatory mechanisms of the expression of CICRGs not only contributes to a deeper understanding of the pathogenesis of CC but also provides new reference data and potential therapeutic targets for the clinical treatment of CC. For example, CICRGs expression can increase CC cells’ proliferative and migratory capacities, thereby achieving therapeutic goals. Additionally, the expression levels of CICRGs may serve as biomarkers for early diagnosis, prognosis assessment, and therapeutic response monitoring of CC, furnishing a more accurate foundation for clinical diagnosis and management of CC.

Based on bioinformatics analysis, we have identified diagnostic genes related to calcium channel genes in CC, constructed a ceRNA network, and conducted drug prediction analysis for these diagnostic genes. This provides theoretical data for studying the molecular mechanisms underlying CC development.

Through differential expression analysis and WGCNA, we screened 30 candidate genes related to Ca2 + channel genes in CC. An enrichment analysis of these genes showed that they were notably concentrated in signaling pathways and biological processes related to the transport of calcium ions across membranes, the regulation of cytosolic calcium ion concentration towards increase, and calcium-related signaling pathways. The signaling pathways and biological processes involved in the positive regulation of cytosolic calcium ion concentration and calcium-mediated signaling have multifaceted effects on the development of CC, involving gene transcription, cell proliferation, DNA repair, cell apoptosis, survival, cell invasion, metastasis, and other aspects.

Three machine learning algorithms and seven candidate diagnostic genes (VEGFA, CALML3, ITPR3, ITPR2, ERBB2, BDKRB2, and TBXA2R) were screened from 30 candidate genes. Further screening these eight genes ultimately identified two prognostic genes: VEGFA and CALML3 Based on the GSEA results of the prognostic genes, it was found that VEGFA was significantly positively correlated with tumor-related processes such as cell cycle, DNA replication, and pyrimidine metabolism. In contrast, CALML3 was significantly negatively correlated with tumor-related processes such as DNA replication and ECM-receptor interaction.

VEGFA, also known as the vascular permeability factor (VPF), serves as an endothelial growth factor and a modulator of VPF. This agent exerts its effects specifically on endothelial cells and exhibits diverse functions, such as facilitating enhanced vascular permeability, stimulating angiogenesis and vasculogenesis, promoting endothelial cell proliferation and migration, and inhibiting cell apoptosis [41] Many tumor cells secrete VEGFA to promote the formation of new blood vessels, providing nutrition and oxygen for tumor growth and metastasis. VEGFA plays a crucial role in the development of tumors such as breast cancer, gastric cancer, and colorectal cancer. Therefore, VEGFA is also a target for many anticancer therapeutic drugs [42, 43]. Research by Hua Guo and colleagues [44] indicates that LINC00707 and miR-382-5p can serve as novel biomarkers for CC and potential therapeutic targets for its treatment through modulation of VEGFA. Studies by Jiali Peng and colleagues [45] demonstrate that JAM3 promotes migration and invasion of CC cells by activating the HIF-1α/VEGFA pathway. CALML3 is a specific light chain for MYO10, a non-conventional myosin crucial for cytoskeleton dynamics and cellular functions. The calcium-dependent interaction between CALML3 and MYO10 is essential for MYO10’s role in filopodial extension, a critical cellular morphological process involving cell migration, invasion, and other physiological and pathological activities [46]. CALML3 is implicated in the progression of various tumors and diseases. In tongue and esophageal squamous cell carcinomas and gastric cancer, CALML3 acts as a tumor suppressor [47]. CALML3 is lowly expressed in tumor tissues in liver cancer and exhibits tumor-suppressing activity by inhibiting lung metastasis [48] Research has indicated a positive association between decreased CALML3 expression and patient survival rates [49]; conversely, other studies propose that CALML3 is overexpressed in lung cancer and acts as an oncogene promoting tumor growth. However, there are limited reports on CALML3 in CC. The research conducted by C-N Liu and colleagues [50] revealed that CALML3-AS1 is an overexpressed lncRNA in CC. High levels of CALML3-AS1 predict a poor prognosis for CC patients, and CALML3-AS1 acts as an anti-metastatic lncRNA by regulating the Wnt/β-catenin pathway in CC. The opposing roles of VEGFA and CALML3 in tumor-related processes may reflect their significant importance in maintaining a balance between tumor cell proliferation and inhibition. This balance has a crucial impact on tumorigenesis, development, and metastasis. Due to the pivotal roles of VEGFA and CALML3 in tumor-related processes, they may serve as potential therapeutic targets. Inhibitors targeting VEGFA have already been widely used in tumor treatment, and intervention strategies targeting CALML3 may also provide new ideas and methods for tumor therapy.

Immunological analysis reveals that CALML3 indicates immune infiltration in CC tissues. Reduced CD8 + and CD4 + T cell levels impair immune function, worsening the prognosis of CIN and CC patients. NK and B cells are crucial for immune protection as they eliminate HPV and infected cells. Regulatory T cells promote tumor progression by inhibiting the anti-tumor abilities of cytotoxic T cells and NK cells [51]. CALML3 can be used as an immunogen to prepare antibodies with high specificity for endogenous CALML3 proteins, further proving its importance in immune responses.

To forecast the regulatory interplay and potential patterns between CC and two pivotal diagnostic genes, we constructed a ceRNA regulatory network encompassing CALML3-hsa-miR-326-TMEM147-AS1 and VEGFA-sa-miR-107-MIATNB. Furthermore, three miRNAs (hsa-miR-107, hsa-miR-326, and hsa-miR-33b) concurrently regulate two prognostic genes. MicroRNAs are pivotal in modulating proteins crucial for ER Ca2 + homeostasis and maintaining luminal Ca2 + levels, including Ca2 + transporters and channels (e.g., IP3R and SERCA), as well as ER residents and Ca2 + binding partners like calreticulin, immunoglobin binding protein, and protein disulfide isomerases. By finely tuning ER Ca2 + homeostasis, microRNAs serve as key regulators within the protein homeostasis network [52]. Long non-coding RNAs (lncRNAs), defined as non-coding RNAs exceeding 200 nucleotides in length, play significant roles in inflammatory responses, cardiovascular diseases, and various types of cancers [53] LncRNAs regulate the expression of genes related to calcium binding by sponging microRNAs [54]. Consequently, microRNAs and lncRNAs can influence the initiation and progression of CC at the molecular level by regulating calcium-related pathways.

The results of predictive analysis of disease drugs indicate that two genes are jointly associated with liver injury caused by tumors, inflammation, chemicals, and drugs. BPA, TCDD, and estradiol may be drugs that act on these two key diagnostic genes. CC development is marked by the persistence of high-risk HPV infection alongside localized chronic inflammation, ultimately driving the oncogenic process. Significant expression of IL-18 has been detected in CC cells, and its level of expression has been found to correlate positively with the aggressiveness of CC [55]. Research conducted by Seoyoung C. Kim and colleague [56] demonstrated that individuals afflicted with systemic inflammatory diseases (SID), such as rheumatoid arthritis and systemic lupus erythematosus, have a 1.5-fold increased risk of developing high-grade cervical dysplasia and CC compared to women without SID. Therefore, these two genes may influence the development of CC by inducing inflammatory responses. While the detrimental health impacts of estrogenic endocrine disruptors (EEDs), like BPA and phthalates, have been acknowledged, their involvement in the progression of CC remains uncertain. Estrogen controls vital cellular functions such as growth, specialization, and programmed cell death in tissues that respond to hormones. Therefore, disturbances in estrogen signaling by EEDs may lead to the occurrence and development of cancer [57]. BPA exhibits both estrogenic and anti-androgenic properties due to its ability to bind to steroid receptors and activate the ER signaling pathway in a manner similar to estrogen [57]. TCDD (2,3,7,8-Tetrachlorodibenzo-p-dioxin) is a polychlorinated dibenzodioxin with both estrogen agonist and antagonist activities [58]. Estrogen acts as an effective mitogen in ER-positive cells, and continuous exposure to estrogen stimulation may promote DNA instability, cell proliferation, and the tumorigenic transformation of epithelial cells into cancer [59]. In the CaSki cell line, estradiol increases the expression of HPV oncogenes. The administration of indole-3-carbinol, a compound with anti-estrogenic activity, eliminated the expression of HPV oncogenes [60]. If drug design can be optimized or drug synthesis processes improved to inhibit or eliminate the estrogen agonist-related activity of TCDD while retaining its estrogen antagonist activity, it may have potential as a therapeutic drug for cervical cancer.

This study identified two diagnostic genes potentially indicative of CC pathogenesis. Analysis of their functional pathways, immune features, regulatory networks, and related drugs may aid in mechanism exploration and new diagnostic strategies for CC. However, further experimental and clinical research is required. Our team will continue to focus on CC research, exploring pathogenesis, diagnostics, and treatments to enhance patient survival and quality of life.

Conclusion

Based on bioinformatic analysis, a sum of two prognostic genes (VEGFA and CALML3) related to calcium ion channels was identified, providing underlying methods for the targeted therapy for cervical cancer.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Quan Yang: Conceptualization, Date Curation, Formal Analysis, Methodology, Software, Visualization, Writing-Original Draft, Writing-Review & Editing; Jiayi Zhao: Methodology, Supervision; Xiaoyan Liu: Supervision, Validation; Ying Chen: Data Curation, Investigation; Jing Liu: Data Curation; Jialiang Mai: Formal Analysis; Menghui Hong (Corresponding Author): Conceptualization, Resources, Supervision, Validation, Writing-Original Draft, Writing-Review & Editing;

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Data availability

The datasets generated during and/or analysed during the current study are available in the Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/), datasets GSE9750 and GSE63514 originated from the Gene Expression Omnibus (GEO) repository (https://www.ncbi.nlm.nih.gov/geo/)，the Kyoto Encyclopedia of Genes and Genomes (KEGG) database ( [https://www.genome.jp/kegg/](https:/www.genome.jp/kegg) ). The complete bioinformatics analysis code for this study has been made publicly available in a GitHub repository: https://github.com/menghuihong22-boop/cervical_cancer.git.

Declarations

Ethical approval and consent to participate

The study was conducted following the Declaration of Helsinki and was approved by the Ethics Committee of Foshan Women and Children Hospital Affiliated to Guangdong Medical University (protocol number FSFY-MEC-2025-031). As this was a retrospective study using anonymized data, the requirement for informed consent was waived by the Ethics Committee. The requirement for informed consent was waived by the Ethics Committee of Foshan Women and Children Hospital Affiliated to Guangdong Medical University (protocol number FSFY-MEC-2025-031) because this was a retrospective study using anonymized data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.