Immunological characteristics of calmodulin-related genes in hepatocellular carcinoma and their potential for prognostic prediction and diagnostic applications

Introduction

Hepatocellular carcinoma (HCC), the most common subtype of liver cancer in clinical practice, exhibits high incidence and mortality rates worldwide, posing a serious threat to human health and life (1,2). According to relevant research projections, the number of new liver cancer cases worldwide will show a significant upward trend over the twenty-year period from 2020 to 2040, with an estimated increase of 55.0%. By 2040, the number of new liver cancer diagnoses globally is expected to exceed 1.4 million cases. Concurrently, liver cancer mortality will also climb, with an estimated 1.3 million deaths projected for 2040—a 56.4% increase from 2020 (3)—underscoring the critical urgency of liver cancer prevention and control. The theoretical preventability of most HCC cases stems from its well-defined major causes: viral factors (chronic hepatitis B and C), metabolic factors (metabolic dysfunction-associated steatohepatitis), and alcohol-related liver disease (4,5). In recent years, tumor markers have demonstrated increasing clinical value in the auxiliary diagnosis, treatment efficacy monitoring, and prognosis assessment of malignant tumors (6,7). Among these, serum alpha-fetoprotein (AFP) holds significant importance as the most commonly used key indicator for detecting HCC and monitoring treatment outcomes (8,9). However, AFP still has significant limitations in diagnosing HCC: although its diagnostic specificity can reach 80–94%, its sensitivity is only 25–65%, making it difficult to meet the needs of early diagnosis. Furthermore, the specific threshold values for AFP diagnosis and prognostic assessment, as well as its predictive value for HCC patient survival rates and postoperative recurrence rates, remain controversial (10,11). Therefore, identifying and developing more sensitive, convenient, and specific HCC diagnostic biomarkers to enable early detection, diagnosis, and intervention holds significant research value for improving patient prognosis and reducing mortality.

The initiation and progression of tumors involve complex reprogramming of intracellular signaling networks. Calcium ions, as ubiquitous second messengers, play a pivotal role in the proliferation, evasion of apoptosis, invasion, metastasis, and treatment resistance of various tumors through dysregulated signaling pathways (12-14). As a classic calcium ion sensor in eukaryotic cells, calmodulin mediates extensive calcium-dependent signaling and regulates diverse cellular processes (15,16). Calmodulin and its associated genes have been implicated in the pathogenesis of a wide range of cancers, according to recent research (15). For example, CALM1, a ubiquitous calcium receptor protein, shows significant overexpression correlated with clinical staging and poor overall prognosis in tubulo-squamous cell carcinoma (17). Furthermore, CaMKK2 not only promotes tumor progression but also induces immune suppression by impairing the immune function of CD8+ and CD4+ T cells, thereby mediating resistance to chemotherapy and immunotherapy in tumors such as glioma (18). Although calmodulin-related genes (CRGs) have been extensively reported in multiple cancers, systematic research on this gene family in HCC remains scarce, and its potential value as a diagnostic and prognostic biomarker has yet to be fully explored.

Based on this, the present study aims to systematically analyze the expression patterns, clinical associations, and biological functions of CRGs in HCC using bioinformatics methods. Specific objectives include: (I) identifying differentially expressed CRGs in HCC tissues through public databases; (II) evaluating the diagnostic efficacy of these genes for HCC; (III) analyzing the correlation between their expression levels and patients’ clinical-pathological characteristics and prognosis, constructing and validating potential prognostic prediction models; (IV) preliminarily exploring the biological processes and immune characteristics potentially implicated. This study aims to provide novel molecular targets for the early diagnosis and prognostic assessment of HCC. We present this article in accordance with the TRIPOD reporting checklist (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0204/rc).

Methods

Analyze data sources

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. For the training set, RNA-Seq data of HCC [The Cancer Genome Atlas (TCGA)-LIHC] were sourced from the UCSC Xena platform, including 345 tumor and 50 normal samples. Corresponding copy number variation and clinical data were acquired from TCGA. Prior to analysis, cases lacking survival information or with an overall survival of less than 30 days were excluded. For independent validation, three microarray datasets from the Gene Expression Omnibus (GEO) database were included: GSE76472 (115 tumor and 52 control samples), GSE84005 (38 tumor and 38 control samples), and GSE14520 (220 tumor samples and 212 control samples). A total of 255 CRGs (Table S1) were identified by searching the human protein atlas database using the keyword “calmodulin”.

Analysis of differential calmodulin expression in HCC

Using the limma package, we conducted a differential expression analysis on the training set’s tumor versus control profiles, with significant genes defined by |logFC| >1 and adj. P value <0.05. Cross-referenced the identified differentially expressed genes (DEGs) with CRGs to obtain differentially expressed CRGs (DECRGs). The clusterProfiler package was utilized to carry out comprehensive GO enrichment and KEGG pathway analyses on these DECRGs, thereby pinpointing their key biological functions and signaling pathways.

Protein interaction network construction

DECRGs were imported into the STRING database for protein-protein interaction network analysis. An interaction threshold of confidence greater than 0.4 was set to construct the PPI network. This network was visualized using Cytoscape (V3.8.0) software. The Degree algorithm within the Cytohubba plugin was further utilized to identify the top 10 key hub genes with the highest degree values in the network for subsequent analysis.

Machine learning methods for screening diagnostic biomarkers

Three machine learning algorithms were employed for variable selection (19-21). Random forest (RF; randomForest package), least absolute shrinkage and selection operator (LASSO) regression (glmnet), and support vector machine recursive feature elimination (SVM-RFE) (e1071) analyses were sequentially performed on the 10 hub genes to filter key features. The outcomes of these analyses were then integrated, and the intersecting genes were selected as potential diagnostic biomarkers.

Analysis of HCC diagnostic genes

The diagnostic efficacy of candidate biomarkers was evaluated by performing receiver operating characteristic (ROC) curve analysis on the TCGA-LIHC dataset with the pROC package. Candidate genes demonstrating an area under the curve (AUC) value above 0.7 were retained as possessing potential diagnostic value. Subsequently, the screened diagnostic genes were integrated to construct a diagnostic model for HCC, and the model was validated using tumor and control samples from the training set (TCGA-LIHC) and multiple independent validation sets (GSE76427, GSE14520, and GSE84005) as input data. The rmda package was utilized for nomogram construction to visualize the model. Its clinical utility was then quantified via decision curve analysis (DCA), while a calibration curve was used to evaluate predictive accuracy. Additionally, single-gene enrichment analysis was performed for each selected gene to explore potential biological functions in HCC.

Prognosis gene screening and risk model development

A multi-stage strategy was implemented to develop a prognostic risk model for HCC (22-24). Step 1: Gene screening. Univariate Cox regression on DECRGs identified candidate genes (P<0.05). To prevent overfitting, LASSO regression (glmnet package) with 10-fold cross-validation was applied for further selection. Step 2: Model construction. Multivariate Cox regression was performed on the filtered genes to define a final prognostic signature and calculate a risk score per patient (based on expression and regression coefficients). Step 3: Model validation. In the TCGA-LIHC training set, survival analysis was performed to compare survival differences between risk groups, ROC curves were used to assess the predictive performance of the model for 1-, 3-, and 5-year survival rates, and the joint distribution of risk scores and survival status was visualized. To further validate the model’s generalizability, the entire analytical workflow was replicated in the independent external dataset GSE14520. Tumor samples from both datasets served as input for prognostic model validation.

Assessment of the association between clinical characteristics and risk scores

Based on clinical data and risk scores, univariate and multivariate Cox regression analyses were performed on the training set samples. A nomogram was constructed using the rms package to predict 1-, 3-, and 5-year survival rates for patients. DCA was further employed to evaluate the nomogram’s decision-support value in clinical practice. Combined with the prognostic risk score, statistical descriptions and comparisons of clinical characteristics in the TCGA-LIHC cohort were performed.

Molecular mechanism differences between risk groups

GSEA (v4.3.2) was performed on different risk subtypes, with screening criteria set as |NES| >1 and adj. P value <0.05. Additionally, DEGs between high- and low-risk groups were identified through differential expression analysis (adj. P value <0.05, |logFC| >1). To characterize their biological functions, both GO and KEGG pathway analyses were carried out on the upregulated and downregulated DEGs, respectively.

Immunological infiltration analysis across risk groups and prediction of immunotherapy response

To systematically analyze the heterogeneity of the tumor immune microenvironment and its association with treatment response, this study employed a multi-algorithm approach. To comprehensively compare immune infiltration between the two risk groups, we employed three distinct approaches: ssGSEA to quantify immune cell and functional enrichment; the ESTIMATE algorithm to assess stromal/immune components and tumor purity in the microenvironment; and CIBERSORT to analyze differences in infiltrating immune cell subsets. To further validate the association between risk stratification and immune phenotypes, we incorporated immune phenotype score (IPS) data from TCGA-LIHC in the TCIA database, presenting the IPS distribution differences between high- and low-risk groups through visual analysis. Finally, validation was conducted in the independent IMvigor210 cohort. Using data from patients receiving anti-PD-L1 therapy in this cohort, we analyzed the association between risk grouping and objective response rates to immunotherapy.

Reverse transcription quantitative polymerase chain reaction (RT-qPCR)

The HCC cell line Huh7 (catalog No. CL-0120) and the normal human hepatic epithelial immortalized cell line THLE-2 (catalog No. CL-0833) were both purchased from Wuhan Pricella Life Science & Technology Co., Ltd. (Wuhan, China). Huh7 cells were cultured in DMEM containing 10% fetal bovine serum, while THLE-2 cells were cultured in their specialized medium, under conditions of 37 ℃, 5% CO₂, and saturated humidity. Total RNA was extracted from cells in the logarithmic growth phase using TRIzol reagent, and cDNA was synthesized by reverse transcription using PrimeScript™ RT Master Mix. The concentration and purity of the RNA were measured using a micro-ultraviolet spectrophotometer. Subsequently, the reaction mixture (20 µL reaction system containing 10 µL of 2× SYBR Green Mix, 0.5 µL each of forward and reverse primers, 1 µL of cDNA template, and made up to volume with RNase-free ddH₂O) was subjected to detection on a QuantStudio 5 real-time PCR system. GAPDH was used as the internal reference gene, and the primer sequences are shown in Table S2. Three replicate wells were set for each sample, and the relative expression levels of target genes were calculated using the 2^–ΔΔCt method.

Statistical analysis

All statistical analyses in this study were performed using R software (v4.4.1). Intergroup comparisons were assessed for statistical significance using the Wilcoxon signed-rank test. Survival analysis involved plotting survival curves using the Kaplan-Meier method, with the log-rank test employed to calculate P values for intergroup differences. Data visualization was primarily accomplished using the ggplot2 plotting package. Correlations between variables were measured using Pearson and Spearman methods, respectively.

Results

Machine learning methods for screening diagnostic biomarkers

Using the TCGA-LIHC cohort, we compared transcriptomic data from HCC tissues with normal tissues and identified 3,571 DEGs (Figure 1A). After intersecting these DEGs with CRGs, we obtained 51 DECRGs (Figure 1B). The expression

Figure 1 Machine learning methods for screening diagnostic biomarkers. (A) Volcano plot of DEGs. (B) Venn diagram of gene set intersection (DEGs and CRGs). (C) Hub genes identified from the PPI network. (D) Feature selection from the LASSO regression analysis. (E) Feature selection curve from the SVM-RFE analysis. (F) Random forest model error rate. (G) Ranked feature importance (random forest). (H) Intersection of candidate genes from three machine learning algorithms. (I) Final candidate gene expression across groups. ***, P<0.001. CRGs, calmodulin-related genes; CV, cross-validation; DEGs, differentially expressed genes; FC, fold change; LASSO, least absolute shrinkage and selection operator; PPI, protein-protein interaction; SVM-RFE, support vector machine recursive feature elimination.

profiles of these DECRGs in tumor versus normal samples were visualized as a heatmap (Figure S1A). Further GO and KEGG enrichment analyses, the above genes were significantly enriched in molecular function categories such as calmodulin binding, calmodulin dependent protein kinase activity, and channel activity, and were mainly involved in signaling pathways including the Calcium signaling pathway and the Apelin signaling pathway (Figure S1B). We constructed a PPI network, and the results showed extensive connections among the proteins encoded by these genes (Figure S1C). we identified the top 10 hub genes based on degree centrality (Figure 1C). Differential expression analysis revealed that these hub genes all exhibited significant expression differences between the tumor and control groups; specifically, CALML6, GRIN1, KCNN3, and PLCB1 were up regulated in the tumor group, whereas the remaining six genes were down regulated (Figure S1D). To establish potential diagnostic markers for HCC, we employed three machine learning strategies for gene screening: LASSO regression analysis identified 5 genes (Figure 1D), SVM-RFE determined 8 genes (Figure 1E), and the RF method selected 5 genes (MeanDecreaseGini >5) (Figure 1F,1G). Integrating results from all three approaches yielded five common candidate genes (Figure 1H). Among these, KCNN3 was significantly upregulated in tumor tissues, while CAMK2B, CAMK4, PVALB, and TRPC4 all showed downregulation trends (Figure 1I).

Analysis of HCC diagnostic genes

The five candidate genes identified in the preliminary screening demonstrated high discriminatory performance (AUC >0.7) in ROC analysis, suggesting their potential value as HCC diagnostic biomarkers. Consequently, they were designated as HCC‑associated diagnostic genes (Figure 2A). A diagnostic model based on these genes demonstrated robust and stable performance in independent training and validation cohorts (Figure 2B). The corresponding diagnostic nomogram showed certain potential clinical reference value in the training set (Figure 2C-2E). Furthermore, single-gene enrichment analysis was carried out to elucidate the potential biological functions of these diagnostic genes in HCC (Figure 2F-2J). The results indicated that the KCNN3 gene, which was highly expressed in tumor tissues, might be associated with biological processes such as tube development, blood vessel morphogenesis, and vasculature development. Conversely, the CAMK2B, PVALB, and CAMK4 genes, which showed low expression in tumors, appeared to be involved in cell cycle regulation.

Figure 2 Analysis of HCC diagnostic genes. (A) ROC analysis identifies diagnostic genes for HCC. (B) Validation of the diagnostic model’s performance across training and validation cohorts. (C) Diagnostic nomogram model for HCC. (D) Calibration curve and (E) DCA assessing the nomogram’s clinical utility. (F-J) Functional enrichment analysis for each diagnostic gene (CAMK2B, TRPC5, KCNN3, PVALB, and CAMK4). DCA, decision curve analysis; GO, Gene Ontology; HCC, hepatocellular carcinoma; LIHC, liver hepatocellular carcinoma; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

Immunological characteristics of HCC

ssGSEA was applied to the TCGA-LIHC cohort to evaluate immune cell and functional enrichment. Comparative analysis revealed a significant enhancement in normal tissues relative to tumors, encompassing the infiltration of multiple immune cells (e.g., neutrophils, NK cells) and the activity of key immune functions (e.g., APC co-inhibition) (Figure 3A,3B). Further analysis using the CIBERSORT algorithm revealed higher infiltration of plasma cells, monocytes, and M2 macrophages in the normal group (Figure 3C). Notably, monocyte infiltration positively correlated with the expression of TRPC5, CAMK2B, and PVALB—genes that exhibited low expression in tumor samples (Figure 3D).

Figure 3 Immunological characteristics of HCC. (A) Comparative analysis of immune cell enrichment and (B) immune function in normal vs. tumor tissues (ssGSEA). (C) Landscape of immune cell subset distribution in samples (CIBERSORT). (D) Correlation matrix evaluating the relationship between immune cell infiltration and diagnostic gene expression. ns, not significant; *, P<0.05; **, P<0.01; ***, P<0.001. HCC, hepatocellular carcinoma; ssGSEA, single sample gene set enrichment analysis.

Prognosis gene screening and risk model development

To construct a prognostic scoring model for risk stratification in HCC patients, we first identified 11 genes significantly associated with prognosis based on DECRGs, forming a candidate gene set (Figure 4A). Subsequently, LASSO regression further extracted 9 key prognostic genes (Figure 4B,4C). Multivariate Cox regression analysis was performed on these genes to establish a risk-scoring prognostic model (Figure 4D). Risk scores were calculated for each clinical sample using the model, and patients were stratified into high-risk and low-risk groups based on the median score. In the training cohort, the two groups exhibited distinct spatial distribution patterns (Figure 4E). Significant expression differences were observed for these 9 genes between high- and low-risk groups (Figure 4F) as well as between tumor and normal tissues (Figure 4G). Further ROC curves for 1-, 3-, and 5-year survival were generated using the TCGA-LIHC cohort, yielding AUC values above 0.7, indicating robust predictive performance (Figure 4H). Survival analysis revealed significantly lower overall survival rates in high-risk patients compared to low-risk patients within the training cohort (Figure 4I). A joint distribution plot of risk scores and survival status visually highlighted the disparity between the two groups (Figure 4J). The study was validated using an independent external dataset (GSE14520), yielding results consistent with the training set (Figure 4K-4M).

Figure 4 Prognosis gene screening and risk model development. (A) Screening DECRGs with prognostic value. (B) Path map of model coefficient changes in LASSO regression analysis. (C) Determining optimal parameters for the LASSO regression model via cross-validation. (D) Screening prognostic genes for HCC using multivariate Cox regression. (E) Spatial distribution map of patient samples across different risk groups in the training set. (F) Expression profiles of nine prognostic genes in high- and low-risk groups. (G) Expression profiles of nine prognostic genes in tumor versus normal tissues. (H) ROC curves for 1-, 3-, and 5-year diagnostic efficiency of the risk model in the training set. (I) Survival curves for risk-stratified patients in the training set. (J) Risk score distribution and corresponding survival status map for training set samples. (K) ROC curve for diagnostic efficiency prediction by the risk score model in the external validation set (GSE14520). (L) Survival curves for risk-stratified patients in the external validation set. (M) Risk score distribution among external validation samples and corresponding survival status plot. *, P<0.05; ***, P<0.001. AUC, area under the curve; CI, confidence interval; DECRGs, differentially expressed calmodulin-related genes; HCC, hepatocellular carcinoma; LASSO, least absolute shrinkage and selection operator; LIHC, liver hepatocellular carcinoma; PC, principal component; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

Gene enrichment analysis across different risk groups

Functional separation was observed between risk groups via GSEA. High-risk samples were characterized by cell cycle-related activities (e.g., DNA replication, mitotic cell cycle phase transition), while low-risk samples showed active metabolic processes like fatty acid and bile acid metabolism (Figure 5A,5B). Subsequent analysis reinforced this dichotomy: upregulated genes were associated with mitotic regulation and enriched in cell cycle and p53 pathways; downregulated genes participated in steroid and xenobiotic metabolism, with notable enrichment in Retinol metabolism (Figure 5C,5D).

Figure 5 Gene enrichment analysis across different risk groups. (A) Biological processes significantly enriched in the high-risk group primarily involve cell cycle regulation. (B) Biological processes significantly enriched in the low-risk group are mainly concentrated in metabolism-related pathways. (C,D) GO (left) and KEGG (right) enrichment results for (C) up-regulated and (D) down-regulated DEGs in the high- and low-risk groups. DEGs, differentially expressed genes; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Assessment of the association between clinical characteristics and risk scores

By integrating univariate and multivariate Cox regression analyses, this study confirmed that the risk score serves as an independent prognostic factor for patients with HCC (Figure 6A,6B). Building on this, we further incorporated clinical features to construct a prognostic nomogram model, which exhibited potential predictive accuracy for the 1-, 3-, and 5-year survival rates of HCC patients (Figure 6C-6E). Analyses demonstrated that male HCC patients had significantly higher risk scores compared to females, and patients with disease progression to T3+T4 stage or III+IV stage also exhibited significantly elevated risk scores (Figure 6F). Moreover, across various clinical characteristic subgroups, patients in the low-risk group showed significantly longer survival times than those in the high-risk group, indicating that the low-risk group was consistently associated with better survival outcomes across all subtypes (Figure 6G).

Figure 6 Assessment of the association between clinical characteristics and risk scores. (A,B) Forest plot of Cox regression analysis for risk scores and clinical-pathological characteristics. (C) Nomogram model for predicting 1-, 3-, and 5-year survival rates in HCC patients. (D,E) DCA (D) and (E) calibration curve for nomogram model predictive performance. (F) Comparison of risk scores across gender and disease staging groups. (G) Survival curve comparison between high- and low-risk patients in different clinical feature subgroups. CI, confidence interval; DCA, decision curve analysis; HCC, hepatocellular carcinoma; LIHC, liver hepatocellular carcinoma; M, metastasis; N, node; T, tumor; TCGA, The Cancer Genome Atlas.

Immunological infiltration analysis across risk groups and prediction of immunotherapy response

The low-risk group was characterized by a broadly activated immune landscape. This included higher infiltration levels of most immune cells (Figure 7A) and enhanced immune function activity (Figure 7B), as measured by ssGSEA. Detailed deconvolution via CIBERSORT pinpointed elevated levels of specific resting immune subsets (Figure 7C), while ESTIMATE analysis revealed higher stromal and immune scores (Figure 7D). Consistent with an immunogenic phenotype, this group exhibited upregulated immune checkpoint expression (Figure 7E) and demonstrated greater sensitivity to anti-CTLA-4/PD-1 immunotherapy (Figure 7F). Additionally, we found that immune checkpoints highly expressed in the low-risk group (such as CD14, IDO2, and CD48) were negatively correlated with the risk score, whereas immune checkpoint molecules with higher expression levels in the high-risk group (such as CD276, TNFSF15, and VTCN1) were positively correlated with the risk score (Figure S2). These features translated to superior clinical outcomes in the IMvigor210 cohort, including prolonged overall survival (Figure 7G), a higher rate of treatment response (Figure 7H), and an inverse correlation between response and risk score (Figure 7I).

Figure 7 Immunological infiltration analysis across risk groups and prediction of immunotherapy response. (A) Comparison of ssGSEA analysis for immune cell types and (B) immune function infiltration levels across risk groups. (C) Differences in infiltration levels of immune cell subpopulations between groups based on the CIBERSORT algorithm. (D) Comparison of tumor microenvironment scores between risk groups. (E) Comparison of immune checkpoint molecule expression levels between risk groups. (F) Differences in response rates to anti-CTLA4 and anti-PD1 immunotherapy among patients in different risk strata. (G) Kaplan-Meier survival curves based on risk stratification in the IMvigor210 cohort. (H) Proportion of patients achieving R (response, including partial and complete response) and NR (no response, including stable disease and disease progression) in low-risk versus high-risk groups. (I) Comparison of risk scores between R and NR groups. ns, not significant; *, P<0.05; **, P<0.01; ***, P<0.001. ssGSEA, single sample gene set enrichment analysis.

Mutation characteristics and drug screening for HCC risk stratification

This study analyzed the mutation spectrum in high- and low-risk groups of HCC patients, revealing that missense mutations were the predominant mutation type in both groups. Among these, TP53 exhibited the highest mutation rate in the high-risk group, while CTNNB1 was the most mutated gene in the low-risk group (Figure 8A). Survival analysis indicated that patients with low-risk status and lower mutation burden demonstrated better prognosis (Figure 8B). Treatment sensitivity analysis suggested that high-risk patients may exhibit greater sensitivity to gemcitabine therapy, while low-risk patients may be better suited for sorafenib treatment (Figure 8C). Furthermore, high expression of ASPM in tumor tissues was negatively correlated with drug sensitivity to lexibulin (Figure 8D).

Figure 8 Mutation characteristics and drug screening for HCC risk stratification. (A) Comparison of mutation spectrum between high- and low-risk HCC groups. (B) Survival analysis of HCC patients based on risk scores and mutation burden. (C) Sensitivity analysis of HCC high- and low-risk groups to different chemotherapeutic agents. (D) Correlation analysis between prognostic gene expression and potential drug sensitivity. HCC, hepatocellular carcinoma; TMB, tumor mutational burden.

Validation of mRNA expression of genes associated with diagnosis and prognosis using RT-qPCR

To verify the reliability of the bioinformatics analysis results, RT-qPCR was performed to detect the mRNA expression levels of the aforementioned diagnostic and prognostic genes in the human normal liver immortalized epithelial cell line THLE-2 and the HCC cell line Huh7. The results (Figure 9A,9B) showed that, compared with THLE-2 cells, the diagnostic genes CAMK2B, PVALB, and TRPC5 were all lowly expressed in Huh7 cells. Among the prognostic genes, RGS2, CAPN6, FAS, and MYO3A were also lowly expressed in Huh7 cells, whereas KCNN1, KCNQ3, and ASPM exhibited high expression. Furthermore, among the shared genes serving as both diagnostic and prognostic markers, KCNN3 was highly expressed in Huh7 cells, while CAMK4 was lowly expressed. These RT-qPCR results were consistent with the previous analysis findings, further validating the conclusions of this study.

Figure 9 mRNA expression levels of genes associated with diagnosis and prognosis in a normal liver cell line (THLE-2) and a HCC line (Huh7) as determined by RT-qPCR. (A) Comparison of expression levels of diagnostic genes. (B) Comparison of expression levels of prognostic genes. ****, P<0.0001. HCC, hepatocellular carcinoma; RT-qPCR, reverse transcription quantitative polymerase chain reaction.

Discussion

HCC, a malignant tumor with persistently high global incidence and mortality rates, poses significant challenges to clinical management due to its difficult early diagnosis, poor prognosis, and marked interindividual variability in treatment response (25,26). Calmodulin, a core regulatory molecule in calcium signaling pathways, extensively participates in critical biological processes such as cell proliferation, differentiation, apoptosis, and immune regulation through interactions with downstream target genes. Its abnormal expression is closely associated with the initiation and progression of multiple tumors (12-14). This study systematically screened DECRGs in HCC using the TCGA-LIHC cohort and external validation datasets. Machine learning algorithms were employed to identify diagnostic biomarkers and construct prognostic risk models. The association between these biomarkers and the tumor immune microenvironment as well as immunotherapy response was explored, providing novel insights and experimental evidence for precision diagnosis and treatment of HCC.

Early diagnosis is crucial for improving the prognosis of HCC patients, but currently used clinical markers such as AFP have limitations in sensitivity and specificity (8,9,11). Therefore, to identify diagnostic biomarkers for HCC, this study screened 51 DECRGs by comparing HCC tissue with normal tissue transcriptomic data. By integrating three machine learning methods and analyzing ROC curves, five genes with significant diagnostic efficacy were ultimately identified: KCNN3, CAMK2B, CAMK4, PVALB, and TRPC4. The diagnostic model constructed using these genes demonstrated stable and robust discriminatory performance in both the training and external validation datasets. Beyond early diagnosis, prognosis assessment is crucial for clinical treatment planning in HCC patients. Traditional clinical and pathological features face limitations in fully reflecting patient heterogeneity during prognosis evaluation (27,28). Therefore, this study identified 9 key prognostic genes through LASSO regression and multivariate Cox regression analysis based on 11 DECRGs associated with prognosis, and constructed a prognostic risk scoring model. This model demonstrated robust predictive performance in both the TCGA-LIHC training dataset and the external validation dataset GSE14520. Notably, KCNN3 and CAMK4 were simultaneously identified as genes associated with both HCC diagnosis and prognosis, suggesting their potential as dual biomarkers with diagnostic and prognostic evaluation capabilities.

Potassium calcium-activated channel subfamily N member 3 (KCNN3 or KCa2.3) is a member of the Small-conductance Ca2+-activated potassium channels (KCa2.x) family (29). KCa2.x exhibits insensitivity to membrane voltage changes, stemming from the absence of most positively charged residues associated with voltage gating in this channel (30). At low intracellular calcium concentrations, this channel can be activated through a specific calmodulin-gated mechanism, thereby performing its physiological ion channel function (31,32). Existing research has confirmed that KCNN3 functional abnormalities are closely associated with the onset and progression of various diseases (33-35). Gain-of-function mutations in KCNN3 have been definitively linked to idiopathic non-cirrhotic portal hypertension, a disorder characterized by abnormal hepatic histology that ultimately disrupts hepatic microcirculation, impairing normal blood supply and physiological function (36,37). Notably, single-gene enrichment analysis in this study suggests KCNN3 may participate in HCC angiogenesis processes (e.g., tube development, blood vessel morphogenesis, and tube morphogenesis). Previous studies have reported that KCNN3 is a KCa2.x family subtype primarily expressed in endothelial cells and plays a crucial regulatory role in vasodilation (38). Based on these findings, we hypothesize that KCNN3 may influence disease progression in non-cirrhotic portal hypertension and HCC by regulating the functional state of the vascular system. This remains unconfirmed and requires further experimental validation and elucidation. Another key gene, calcium/calmodulin-dependent protein kinase IV (CAMK4), functions as a multifaceted serine/threonine protein kinase that plays a crucial role in regulating diverse cellular processes including T cell differentiation, podocyte function, and tumor cell proliferation and apoptosis (39,40). Its activation mechanism is closely linked to intracellular calcium ion levels: Elevated calcium concentrations promote the binding of calmodulin to CAMK4, thereby removing its pseudo-substrate from the catalytic domain. This allows CAMK4 to access its actual substrate and ATP, completing its activation (41). Previous studies have confirmed that CAMK4 expression is significantly downregulated in HCC, and this low expression correlates with poor patient prognosis and shorter overall survival. Functional studies indicate that upregulating CAMK4 expression effectively induces apoptosis in HCC cells and inhibits their proliferative capacity (42). Findings from our bioinformatics analysis further clarify the tumor-suppressive role of CAMK4. The results implicate its involvement in HCC cell cycle biology (e.g., mitotic and cell cycle processes) as a likely mechanism through which it constrains tumor growth.

This study employed the median cutoff value of the prognostic gene risk score to classify HCC patients into risk groups with distinct immune profiles. Previous reports indicate that substantial infiltration of NK cells and CD8+ T cells predicts favorable prognosis in early-stage HCC patients and correlates positively with the number of apoptotic tumor cells in human HCC (43). This aligns with our findings that patients in the low-risk group, who exhibit better survival rates, demonstrate significantly higher infiltration levels of both cell types compared to the high-risk group. CRGs play a crucial role in regulating the tumor immune microenvironment. For example, in head and neck cancers, a decrease in the local levels of membrane-associated calmodulin impairs the immune surveillance capacity of CD8+ T cells (44). In breast cancer, CaMKK2 expressed by tumor-associated macrophages promotes disease progression by inhibiting T-cell antitumor activity, whereas the application of CaMKK2 inhibitors blocks tumor growth in a CD8+ T-cell-dependent manner and induces a favorable remodeling of the immune microenvironment (45).

As a first-line treatment for HCC, sorafenib’s mechanism of action is closely linked to the regulation of NK cell function. Previous studies have confirmed that sorafenib effectively induces NK cell activation in vitro and in tumor-bearing mouse models through IL-12, IL-18, and IL-1β secreted by macrophages, suggesting that enhancing NK cell function may be one of the key mechanisms underlying sorafenib’s antitumor effects (46,47). This also provides a reasonable explanation for the observed increased sensitivity to sorafenib in patients classified as low-risk in this study’s risk stratification. Additional studies indicate that sorafenib can target the RAF-MEK-ERK cascade and angiogenesis via VEGFR2, yet its clinical efficacy remains limited—sorafenib provides only a 2.8-month survival benefit over placebo in HCC patients (48). NK cells play a pivotal role in HCC immune surveillance, yet NK cell-centric therapeutic strategies alone demonstrate suboptimal overall efficacy (49). Consequently, exploring NK cell-associated combination therapies holds significant clinical importance. Subsequent studies further revealed that sorafenib enhances NK cell cytotoxicity in a time-dependent and dose-dependent manner via the RAS/RAF/ERK signaling pathway (50), and that expanded NK cells significantly amplify sorafenib’s antitumor effects (49). Based on the above research evidence, whether the combination of NK cell-based immunotherapy and sorafenib can further improve treatment outcomes and patient prognosis in low-risk HCC patients remains to be confirmed through more in-depth clinical studies and experimental validation.

Furthermore, this study found through enrichment analysis that the high-risk group was significantly enriched in cell cycle-related pathways (such as DNA replication and mitotic cell cycle transition), whereas the low-risk group was enriched in metabolism-related pathways such as fatty acid metabolism and bile acid metabolism. Consistent with previous literature, abnormal calcium signaling is closely associated with metabolic reprogramming and cell cycle dysregulation through various mechanisms. First, regarding metabolic regulation, CaMKK2, as a key effector of calcium signaling, activates AMPK—a central regulator of cellular energy metabolism—thereby coordinating metabolic homeostasis (15,51,52). When intracellular calcium signaling is disrupted, the CaMKK2-AMPK signaling axis becomes imbalanced, granting tumor cells a survival advantage under metabolic stress and promoting metabolic reprogramming to support rapid proliferation (53-55). In this study, the low-risk group exhibited active metabolic characteristics, which may be related to calcium signaling maintaining metabolic homeostasis through the AMPK pathway. Furthermore, regarding cell cycle regulation, CRGs are involved in the regulation of multiple cell cycle checkpoints (56). For example, CaMK regulates the transition from the G1 phase to the S phase of the cell cycle; its inhibition leads to G1/S phase arrest or G2/M phase transition defects (57,58). In hepatocytes, low expression of CAMK4 is closely associated with cell cycle arrest and impaired apoptosis (42). In this study, CAMK4 was significantly downregulated in tumor tissues, while cell cycle pathways were hyperactivated in the high-risk group, suggesting that calcium signaling dysregulation may be involved in the release of cell cycle checkpoints; however, the specific mechanism requires further validation.

This study established the crucial role of CRGs in HCC diagnosis, prognosis prediction, and immune regulation through systematic bioinformatics analysis. The constructed diagnostic and prognostic models demonstrated high accuracy and robustness, closely correlating with tumor immune characteristics and treatment response. These genes may serve as novel biomarkers to assist in early screening, risk stratification, and treatment strategy formulation for HCC. However, this study has certain limitations. First, the analysis primarily relied on retrospective data from public databases and requires validation using prospective clinical samples. Second, the functional roles of these genes and their specific mechanisms in HCC remain to be experimentally elucidated. This particularly applies to CAMK4 and KCNN3, which exhibit potential dual roles in influencing cell cycle/angiogenesis through calcium signaling pathways. Additionally, it remains unclear whether combining NK cell immunotherapy with sorafenib could further improve treatment outcomes and prognosis in low-risk HCC patients. Furthermore, the generalizability of the current model across different populations and HCC subtypes caused by various etiologies requires further validation; in particular, it is worth noting that conclusions regarding diagnostic biomarkers and prognostic models must be validated in independent clinical cohorts.

Conclusions

This study provides preliminary evidence that CRGs have diagnostic and prognostic value in HCC. Five core diagnostic genes were identified, and a nine-gene prognostic risk model was constructed. However, the efficacy of these findings requires further validation through quantitative comparisons with existing clinical indicators in independent prospective cohorts before their potential as molecular targets for precision diagnosis and treatment of HCC can be confirmed.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0204/rc

Peer Review File: Available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0204/prf

Funding: This work was supported by the Medical School Joint Innovation and Open Fund of Lishui University (grant No. X111002, awarded to C.Z.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jgo.amegroups.com/article/view/10.21037/jgo-2026-0204/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Zou R, Hao Y, Wang Y, et al. A multicenter retrospective analysis: Factors influencing hepatic adverse events induced by immunotherapy in advanced liver cancer. Cancer Rep (Hoboken) 2024;7:e1918. [Crossref] [PubMed]
2. Llovet JM, Kelley RK, Villanueva A, et al. Hepatocellular carcinoma. Nat Rev Dis Primers 2021;7:6. [Crossref] [PubMed]
3. Rumgay H, Arnold M, Ferlay J, et al. Global burden of primary liver cancer in 2020 and predictions to 2040. J Hepatol 2022;77:1598-606. [Crossref] [PubMed]
4. Choi S, Kim BK, Yon DK, et al. Global burden of primary liver cancer and its association with underlying aetiologies, sociodemographic status, and sex differences from 1990-2019: A DALY-based analysis of the Global Burden of Disease 2019 study. Clin Mol Hepatol 2023;29:433-52. [Crossref] [PubMed]
5. Muratori L, Lohse AW, Lenzi M. Diagnosis and management of autoimmune hepatitis. BMJ 2023;380:e070201. [Crossref] [PubMed]
6. Zhang X, Liang X, Wen Y, et al. RAC1 inhibition ameliorates IBSP-induced bone metastasis in lung adenocarcinoma. Cell Rep 2024;43:114528. [Crossref] [PubMed]
7. Song J, Xiao T, Li M, et al. Tumor-associated macrophages: Potential therapeutic targets and diagnostic markers in cancer. Pathol Res Pract 2023;249:154739. [Crossref] [PubMed]
8. Hanif H, Ali MJ, Susheela AT, et al. Update on the applications and limitations of alpha-fetoprotein for hepatocellular carcinoma. World J Gastroenterol 2022;28:216-29. [Crossref] [PubMed]
9. Tayob N, Kanwal F, Alsarraj A, et al. The Performance of AFP, AFP-3, DCP as Biomarkers for Detection of Hepatocellular Carcinoma (HCC): A Phase 3 Biomarker Study in the United States. Clin Gastroenterol Hepatol 2023;21:415-423.e4. [Crossref] [PubMed]
10. Trevisani F, D'Intino PE, Morselli-Labate AM, et al. Serum alpha-fetoprotein for diagnosis of hepatocellular carcinoma in patients with chronic liver disease: influence of HBsAg and anti-HCV status. J Hepatol 2001;34:570-5. [Crossref] [PubMed]
11. Wang W, Wei C. Advances in the early diagnosis of hepatocellular carcinoma. Genes Dis 2020;7:308-19. [Crossref] [PubMed]
12. Zheng S, Wang X, Zhao D, et al. Calcium homeostasis and cancer: insights from endoplasmic reticulum-centered organelle communications. Trends Cell Biol 2023;33:312-23. [Crossref] [PubMed]
13. Janke EK, Chalmers SB, Roberts-Thomson SJ, et al. Intersection between calcium signalling and epithelial-mesenchymal plasticity in the context of cancer. Cell Calcium 2023;112:102741. [Crossref] [PubMed]
14. Patergnani S, Danese A, Bouhamida E, et al. Various Aspects of Calcium Signaling in the Regulation of Apoptosis, Autophagy, Cell Proliferation, and Cancer. Int J Mol Sci 2020;21:8323. [Crossref] [PubMed]
15. Tokumitsu H, Sakagami H. Molecular Mechanisms Underlying Ca(2+)/Calmodulin-Dependent Protein Kinase Kinase Signal Transduction. Int J Mol Sci 2022;23:11025. [Crossref] [PubMed]
16. Villalobo A. Regulation of ErbB Receptors by the Ca(2+) Sensor Protein Calmodulin in Cancer. Biomedicines 2023;11:661. [Crossref] [PubMed]
17. Liu T, Han X, Zheng S, et al. CALM1 promotes progression and dampens chemosensitivity to EGFR inhibitor in esophageal squamous cell carcinoma. Cancer Cell Int 2021;21:121. [Crossref] [PubMed]
18. Tomaszewski WH, Waibl-Polania J, Chakraborty M, et al. Neuronal CaMKK2 promotes immunosuppression and checkpoint blockade resistance in glioblastoma. Nat Commun 2022;13:6483. [Crossref] [PubMed]
19. Zhao L, Lv X, Chen W, et al. Athero-oncology perspective: identifying hub genes for atherosclerosis diagnosis using machine learning. Front Immunol 2025;16:1616096. [Crossref] [PubMed]
20. Sun Z, Du T, Yang G, et al. Identification of exosome-related genes in NSCLC via integrated bioinformatics and machine learning analysis. Sci Rep 2025;15:22962. [Crossref] [PubMed]
21. Lv J, Yu CY, Xiong YZ, et al. Immunological biomarkers and gene signatures predictive of radiotherapy resistance in non-small cell lung cancer. Front Immunol 2025;16:1574113. [Crossref] [PubMed]
22. He W, Fan S, Luo Y, et al. Investigating the prognostic role of citrullination-related genes in breast cancer by combining transcriptomics, single-cell analysis and verification. Breast 2025;84:104588. [Crossref] [PubMed]
23. Zhou Y, Ma W, Hu H, et al. Angiogenesis related gene signatures predict prognosis and guide therapeutic strategies in renal clear cell carcinoma. Sci Rep 2025;15:17030. [Crossref] [PubMed]
24. He Y, Liu F, Li Q, et al. Identification of cuproptosis and ferroptosis-related subtypes and development of a prognostic signature in colon cancer. PLoS One 2025;20:e0307013. [Crossref] [PubMed]
25. Foglia B, Turato C, Cannito S. Hepatocellular Carcinoma: Latest Research in Pathogenesis, Detection and Treatment. Int J Mol Sci 2023;24:12224. [Crossref] [PubMed]
26. Chan YT, Zhang C, Wu J, et al. Biomarkers for diagnosis and therapeutic options in hepatocellular carcinoma. Mol Cancer 2024;23:189. [Crossref] [PubMed]
27. Ren Z, Ma X, Duan Z, et al. Diagnosis, Therapy, and Prognosis for Hepatocellular Carcinoma. Anal Cell Pathol (Amst) 2020;2020:8157406. [Crossref] [PubMed]
28. Fowler KJ, Chernyak V, Ronot M, et al. Hepatocellular Carcinoma: It Is Time to Focus on Prognosis. Radiology 2023;307:e220884. [Crossref] [PubMed]
29. Rahman MA, Orfali R, Dave N, et al. K(Ca) 2.2 (KCNN2): A physiologically and therapeutically important potassium channel. J Neurosci Res 2023;101:1699-710. [Crossref] [PubMed]
30. Orfali R, Albanyan N. Ca(2+)-Sensitive Potassium Channels. Molecules 2023;28:885. [Crossref] [PubMed]
31. Nam YW, Downey M, Rahman MA, et al. Channelopathy of small- and intermediate-conductance Ca(2+)-activated K(+) channels. Acta Pharmacol Sin 2023;44:259-67. [Crossref] [PubMed]
32. Ma B, Wu D, Cao E, et al. Structural mechanisms for inhibition and activation of human small-conductance Ca(2+)-activated potassium channel SK2. Nat Commun 2026;17:1770. [Crossref] [PubMed]
33. Hill MC, Simonson B, Roselli C, et al. Large-scale single-nuclei profiling identifies role for ATRNL1 in atrial fibrillation. Nat Commun 2024;15:10002. [Crossref] [PubMed]
34. Mu F, Liu C, Huo H, et al. The relationship between Sjögren’s syndrome and recurrent pregnancy loss: a bioinformatics analysis. Reprod Biomed Online 2024;49:104363. [Crossref] [PubMed]
35. Datta D, Yang S, Joyce MKP, et al. Key Roles of CACNA1C/Cav1.2 and CALB1/Calbindin in Prefrontal Neurons Altered in Cognitive Disorders. JAMA Psychiatry 2024;81:870-81. [Crossref] [PubMed]
36. Koot BG, Alders M, Verheij J, et al. A de novo mutation in KCNN3 associated with autosomal dominant idiopathic non-cirrhotic portal hypertension. J Hepatol 2016;64:974-7. [Crossref] [PubMed]
37. Kmeid M, Liu X, Ballentine S, et al. Idiopathic Non-Cirrhotic Portal Hypertension and Porto-Sinusoidal Vascular Disease: Review of Current Data. Gastroenterology Res 2021;14:49-65. [Crossref] [PubMed]
38. Han MZ, Wang Y, Sun KX, et al. Upregulating vascular endothelial K(Ca)2.3 channels alleviates pulmonary hypertension in mice. Mol Pharmacol 2025;107:100048. [Crossref] [PubMed]
39. Najar MA, Rex DAB, Modi PK, et al. A complete map of the Calcium/calmodulin-dependent protein kinase kinase 2 (CAMKK2) signaling pathway. J Cell Commun Signal 2021;15:283-90. [Crossref] [PubMed]
40. Xu H, Yong L, Gao X, et al. CaMK4: Structure, physiological functions, and therapeutic potential. Biochem Pharmacol 2024;224:116204. [Crossref] [PubMed]
41. Chow FA, Anderson KA, Noeldner PK, et al. The autonomous activity of calcium/calmodulin-dependent protein kinase IV is required for its role in transcription. J Biol Chem 2005;280:20530-8. [Crossref] [PubMed]
42. Li Z, Lu J, Zeng G, et al. MiR-129-5p inhibits liver cancer growth by targeting calcium calmodulin-dependent protein kinase IV (CAMK4). Cell Death Dis 2019;10:789. [Crossref] [PubMed]
43. Chew V, Tow C, Teo M, et al. Inflammatory tumour microenvironment is associated with superior survival in hepatocellular carcinoma patients. J Hepatol 2010;52:370-9. [Crossref] [PubMed]
44. Chimote AA, Gawali VS, Newton HS, et al. A Compartmentalized Reduction in Membrane-Proximal Calmodulin Reduces the Immune Surveillance Capabilities of CD8(+) T Cells in Head and Neck Cancer. Front Pharmacol 2020;11:143. [Crossref] [PubMed]
45. Racioppi L, Nelson ER, Huang W, et al. CaMKK2 in myeloid cells is a key regulator of the immune-suppressive microenvironment in breast cancer. Nat Commun 2019;10:2450. [Crossref] [PubMed]
46. Hage C, Hoves S, Strauss L, et al. Sorafenib Induces Pyroptosis in Macrophages and Triggers Natural Killer Cell-Mediated Cytotoxicity Against Hepatocellular Carcinoma. Hepatology 2019;70:1280-97. [Crossref] [PubMed]
47. Sprinzl MF, Reisinger F, Puschnik A, et al. Sorafenib perpetuates cellular anticancer effector functions by modulating the crosstalk between macrophages and natural killer cells. Hepatology 2013;57:2358-68. [Crossref] [PubMed]
48. Llovet JM, Ricci S, Mazzaferro V, et al. Sorafenib in advanced hepatocellular carcinoma. N Engl J Med 2008;359:378-90. [Crossref] [PubMed]
49. Yang J, Eresen A, Scotti A, et al. Combination of NK-based immunotherapy and sorafenib against hepatocellular carcinoma. Am J Cancer Res 2021;11:337-49.
50. Lohmeyer J, Nerreter T, Dotterweich J, et al. Sorafenib paradoxically activates the RAS/RAF/ERK pathway in polyclonal human NK cells during expansion and thereby enhances effector functions in a dose- and time-dependent manner. Clin Exp Immunol 2018;193:64-72. [Crossref] [PubMed]
51. Hsu CC, Peng D, Cai Z, et al. AMPK signaling and its targeting in cancer progression and treatment. Semin Cancer Biol 2022;85:52-68. [Crossref] [PubMed]
52. Steinberg GR, Hardie DG. New insights into activation and function of the AMPK. Nat Rev Mol Cell Biol 2023;24:255-72. [Crossref] [PubMed]
53. Lin C, Blessing AM, Pulliam TL, et al. Inhibition of CAMKK2 impairs autophagy and castration-resistant prostate cancer via suppression of AMPK-ULK1 signaling. Oncogene 2021;40:1690-705. [Crossref] [PubMed]
54. Penfold L, Woods A, Pollard AE, et al. AMPK activation protects against prostate cancer by inducing a catabolic cellular state. Cell Rep 2023;42:112396. [Crossref] [PubMed]
55. McAloon LM, Muller AG, Nay K, et al. CaMKK2: bridging the gap between Ca2+ signaling and energy-sensing. Essays Biochem 2024;68:309-20. [Crossref] [PubMed]
56. Goshima T, Habara M, Maeda K, et al. Calcineurin regulates cyclin D1 stability through dephosphorylation at T286. Sci Rep 2019;9:12779. [Crossref] [PubMed]
57. Skelding KA, Rostas JA, Verrills NM. Controlling the cell cycle: the role of calcium/calmodulin-stimulated protein kinases I and II. Cell Cycle 2011;10:631-9. [Crossref] [PubMed]
58. Masaki T, Shimada M. Decoding the Phosphatase Code: Regulation of Cell Proliferation by Calcineurin. Int J Mol Sci 2022;23:1122. [Crossref] [PubMed]