CCNA2 orchestrates the PI3K/AKT signaling axis to propel prostate cancer metastasis

Introduction

Prostate cancer (PCa) is the most common malignancy in men worldwide, posing a significant public health burden with approximately 1.6 million new cases and 366,000 deaths annually (1). This staggering incidence reflects not only the aging global population but also increased screening and diagnostic capabilities, particularly in high-income countries.

PCa accounts for nearly one in four cancer diagnoses in men in the United States and Europe, and its prevalence is rising in developing nations as life expectancy increases. Its development involves a complex interplay of genetic, environmental, and lifestyle factors. Beyond inherited mutations such as BRCA2 or HOXB13, epigenetic modifications, dietary habits (e.g., high fat intake), and exposures like certain chemicals or chronic inflammation contribute to disease initiation and progression. For localized disease, treatments like radical prostatectomy and radiation therapy yield excellent outcomes, with over 90% of patients alive after 10 years (2). This high survival rate reflects decades of surgical refinement, advanced imaging for precise tumor targeting, and the indolent nature of many early-stage PCa, which often grow so slowly that patients may die of unrelated causes. Nevertheless, the decision between surgery and radiotherapy involves balancing efficacy with side effects such as urinary incontinence and erectile dysfunction, factors that heavily influence patient quality of life. In contrast, advanced PCa has a markedly worse prognosis (2,3).

Once the tumor metastasizes—most commonly to bone, lymph nodes, or liver—the 5-year survival rate drops dramatically, and curative interventions become unlikely. The biology of advanced disease involves enhanced angiogenesis, immune evasion, and resistance to conventional therapies, necessitating systemic approaches that often carry significant toxicity. The management of castration-resistant prostate cancer (CRPC) is particularly difficult, characterized by eventual metastasis or resistance to androgen deprivation therapy (ADT) (4,5), and limited benefit from chemotherapy like docetaxel (4). The potential emergence of aggressive neuroendocrine prostate cancer (NEPC) further complicates treatment, with palliative chemotherapy offering minimal survival benefit (6,7).

The molecule CCNA2 has emerged as a significant focus in oncology, acting as both a prognostic indicator and a potential therapeutic target. In non-small cell lung cancer (NSCLC), bioinformatics analyses identify CCNA2 as a hub gene for progression (8), and experimental studies show that targeting it can suppress malignancy (9). Similarly, in endometrial cancer, high CCNA2 expression correlates with higher risk (10). These findings underscore its role as a key regulatory molecule across various cancers.

Modern biomarker discovery has been revolutionized by machine learning and single-cell RNA sequencing (scRNA-seq) (11). These tools analyze complex datasets to uncover patterns for prognosis and treatment response prediction. scRNA-seq is especially valuable for elucidating tumor cellular heterogeneity. Integrating these approaches allows for more precise identification of genes associated with disease progression and therapy sensitivity in cancers such as PCa, thereby refining diagnostics and enabling tailored therapeutic strategies (12,13). We present this article in accordance with the TRIPOD and MDAR reporting checklists (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0511/rc).

Methods

Data sources

Three scRNA-seq samples of prostate adenocarcinoma (PRAD) were sourced from the publicly available dataset GSE168668 for primary analysis. Furthermore, bulk RNA-seq data along with corresponding clinical annotations from The Cancer Genome Atlas (TCGA)-PRAD cohort were incorporated into this study. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Single-cell RNA sequencing analysis

For the development and validation of the diagnostic model, multiple datasets, such as TCGA-PRAD, were employed. Initial data processing was performed using the Seurat package to create analysis objects and filter out low-quality cells. Standard quality control measures were applied, focusing on metrics including the percentage of mitochondrial genes, total cell counts, and gene detection rates. Specifically, cells with fewer than 200 detected genes and genes expressed in fewer than three cells were excluded.

Subsequently, data normalization was carried out by scaling the unique molecular identifier (UMI) counts per cell to a factor of 10,000, followed by a log-transformation. To reduce technical noise and highlight biological signals, the dataset was further scaled using the ScaleData function from Seurat (version 3.0.2). Dimensionality reduction was conducted via principal component analysis (PCA), informed by the top 20 most variably expressed genes. For downstream t-distributed stochastic neighbor embedding (t-SNE) and cell clustering, the first 14 principal components were retained. Finally, cellular subpopulations were identified by applying the FindClusters function with a resolution parameter of 0.5.

Functional enrichment studies

Data for this analysis were sourced from the TCGA database (https://portal.gdc.cancer.gov/), and downloaded in MINiML format. Messenger RNA differential expression was evaluated utilizing the Limma package (v3.40.2) within the R environment. To mitigate false discoveries, analyses were based on adjusted P values. Transcripts meeting the criteria of an adjusted P value <0.05 and an absolute log₂ fold change >1 were classified as differentially expressed.

To explore the biological roles of these targets, functional enrichment studies were performed. Gene Ontology (GO) annotation, encompassing the domains of molecular function (MF), biological process (BP), and cellular component (CC), provided a framework for functional characterization. Concurrently, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was employed to elucidate higher-order systemic functions and genomic information associated with the gene sets.

The potential oncogenic functions of the identified mRNAs were investigated by executing GO and KEGG enrichment analyses using the ClusterProfiler package in R. All statistical computations were carried out with R software (version 4.0.3), applying a significance threshold of P<0.05.

TCGA data processing

Prostate tumor RNA sequencing data, along with associated clinical annotations, were sourced from TCGA via its Genomic Data Commons portal (https://portal.gdc.cancer.gov). The STAR-counts data were converted to transcripts per million (TPM) format and subsequently normalized using a log₂(TPM+1) transformation. For the subsequent integrative analysis, only samples possessing both complete gene expression profiles and corresponding clinical records were retained.

Gene correlation analysis

Gene correlation analyses were conducted in R. For pairwise comparisons, the ggstatsplot package was employed to visualize associations. To assess relationships across multiple genes simultaneously, a correlation heatmap was generated using the pheatmap package. Spearman’s rank correlation method was applied to evaluate dependencies between quantitative variables that were not normally distributed. Results with a P value below 0.05 were considered statistically significant.

Clinical data analysis

The visualization was constructed as a Sankey diagram employing the ggalluvial package within the R environment. All statistical analyses were conducted using R software (version 4.0.3). A significance threshold of α=0.05 was applied, with P values falling below this level deemed statistically significant.

Survival analysis

Differences in survival outcomes between groups were evaluated within the Kaplan-Meier analysis framework using the log-rank test. For the Kaplan-Meier curves, both the P value and the hazard ratio (HR) with its 95% confidence interval (CI) were computed via the log-rank test and univariate Cox proportional hazards models. All computations were performed in R, version 4.0.3 (The R Foundation for Statistical Computing, 2020). Statistical significance was defined by a two-sided significance threshold of P<0.05.

Cells

PC3 (FH0195) and DU145 (ZQ0037) are both human-derived PCa cell lines, purchased from Fuheng Biotechnology Co., Ltd. (Shanghai, China) and Zhongqiao Xinzhou Biotechnology Co., Ltd. (Beijing, China), respectively. All basic experiments were repeated three times and were biological replicates.

Colony formation assay

Cells were plated in 6-well dishes and cultured for 14 days. Following this incubation, the resulting colonies were subjected to fixation with methanol and stained using a 0.1% solution of crystal violet. Colonies were then imaged and quantified with an Olympus microscope (Olympus IX73, Olympus Corporation, Tokyo, Japan).

Cell scratch assay

Cells were seeded into a 12-well plate at 1 mL per well to promote uniform distribution and incubated until reaching 80–90% monolayer density. Using a sterile 10 µL pipette tip, two or three parallel linear wounds were gently created in each well, ensuring consistent width. The wells were then rinsed gently three times with PBS to remove dislodged cellular debris. Migration was assessed at 12, 24, and 48 h. After each interval, cells were fixed with 4% paraformaldehyde, stained with crystal violet, and the wound areas were imaged under a microscope.

Transwell assay

Cell migration and invasion capacities were assessed in Transwell chambers, either coated with Matrigel (Corning, Inc., Corning, NY, USA) for invasion assays or left uncoated for migration studies. For the assay, transfected PCa cells (2×10⁴) were resuspended in 100 µL of serum-free medium (Gibco; Thermo Fisher Scientific, Waltham, MA, USA) and plated into the upper chamber. The lower compartment was filled with 500 µL of DMEM containing 10% fetal bovine serum (Shanghai ExCell Biology, Shanghai, China) as a chemoattractant. Chambers were incubated for 24 hours at 37 ℃ with 5% CO₂. Thereafter, cells on the lower membrane surface were fixed with 4% paraformaldehyde (Beyotime Institute of Biotechnology) for 10 minutes and stained with a 0.2–0.5% crystal violet solution (Sigma-Aldrich; Merck KGaA, Darmstadt, Germany) for an additional 10 minutes, both at room temperature. Stained cells were visualized and counted using a fluorescence imaging scanner (Pannoramic MIDI, 3DHISTECH, Budapest, Hungary). The migration assay protocol was identical, except that the Matrigel coating step was omitted.

Western blotting

During the Western blot experiment, protein sample preparation was performed first: the treated cell culture dish was placed on ice, the culture medium was discarded, and the cells were washed three times with ice-cold PBS, removing any residual liquid. Then, freshly prepared RIPA cell lysis buffer was added, and the lysate was collected using a cell scraper. The lysate was then heated in a boiling water bath for 10 minutes to denature the proteins, and the processed samples were stored at −20 ℃ for later use. Subsequently, an SDS-PAGE separating gel of appropriate concentration was prepared according to the molecular weight of the target protein. For protein electrophoresis, after determining the protein concentration of each sample, equal amounts were loaded. Electrophoresis was carried out at a constant voltage of 80 V. The run was stopped when the samples had moved from the stacking gel into the separating gel and migrated to the appropriate position. For the transfer step, the PVDF membrane was first activated with methanol and equilibrated in distilled water. After electrophoresis, the gel was cut to the appropriate size, and transfer was performed using a wet transfer system for 100 minutes.

Next, blocking and primary antibody incubation were carried out: the PVDF membrane after transfer was immersed in blocking buffer containing 5% non-fat milk and blocked at room temperature for 2 hours. After washing three times with 1× TBST, the membrane was incubated with the primary antibody diluted in antibody dilution buffer at 4 ℃ overnight. Finally, secondary antibody incubation and detection were performed: the membrane was taken out and allowed to warm to room temperature for 30 minutes. After washing three times with TBST, the corresponding secondary antibody was added and incubated for 2 hours at room temperature. After another wash, an appropriate amount of chemiluminescent substrate (ECL) was added, and the signal was detected using a chemiluminescence imaging system.

The antibodies for CCNA2 (Cat No. 66391-1-Ig) and GAPDH (Cat No. 10494-1-AP) were purchased from ProteinTech (Proteintech Group, Rosemont, IL, USA). The investigation subsequently assessed the protein expression levels of key components within this pathway. AKT (Cat No. R23412), p-AKT (Cat No. 341790), PI3K (Cat No. R22768), and p-PI3K (Cat No. 341468) were all purchased from Zenbio (Chengdu, China).

Statistical analysis

All data are expressed as the mean ± standard deviation. Statistical comparisons were performed with Student’s t-test using GraphPad Prism software (GraphPad Software, Boston, MA, USA; version 9.0). A threshold of P<0.05 was applied to determine statistical significance.

Results

Cellular population localization of functional genes

The initial phase of our investigation focused on identifying genes associated with metastasis in PCa. We analyzed three PRAD samples sourced from the GSE168668 dataset. Stringent thresholds were applied for cell selection: each cell was required to have between 2,000 and 6,000 detected RNA molecules, with mitochondrial RNA content not exceeding 15% (Figure 1A). Following this quality control, highly variable genes were identified using robust principal component analysis (RPCA). This feature set was then used to correct for batch effects across the samples (Figure 1B-1D). An analysis of variance (ANOVA) test subsequently revealed 2,000 genes exhibiting significant differential expression among the cell populations (2,000 highly variable genes in total, only visualizing the top 10, Figure 1E,1F).

Figure 1 Identification of metastasis-related genes. (A) Quality assessment of the scRNA-seq data across distinct cellular sub-populations. (B-D) The PCA visualizations after integration and batch-effect correction. (E,F) The identification of highly variable genes following count data normalization and batch removal. (G,H) Cellular stratification of the PRAD samples, achieved via t-SNE dimensional reduction. (I) The proportional cell distribution between prostate cancer groups characterized by the presence or absence of metastasis. (J) Corresponding spatial cell distribution map for these groups. (K) The associated functional enrichment analysis. DMSO, dimethyl sulfoxide; ECM, extracellular matrix; EMT, epithelial-mesenchymal transition; mt, mitochondrial; NKT, natural killer T; PC, principal component; PCA, principal component analysis; PRAD, prostate adenocarcinoma; RES, resveratrol; ROS, reactive oxygen species; scRNA-seq, single-cell RNA sequencing; t-SNE, t-distributed stochastic neighbor embedding; UMAP, uniform manifold approximation and projection.

Single-cell RNA sequencing analysis delineated the cellular composition of the three PRAD samples, identifying eight distinct clusters: epithelial cells, endothelial cells, B cells, NKT cells, dendritic cells, luminal epithelial cells, stem cells, and enteroendocrine cells (Figure 1G,1H). Figure 1I depicts the proportional composition of cells when comparing metastatic and non-metastatic PCa groups, while Figure 1J illustrates their respective spatial distributions. Expression profiles and functional enrichment of key genes across these groups are presented in Figure 1K.

Bioinformatic database analysis identified an association between CCNA2 expression and metastasis in PCa

To elucidate potential mechanisms of metastasis in PCa, transcriptomic data from metastatic cells (dataset GSE104935) were analyzed. This comparison revealed 5,631 genes with differential expression (fold change >2.00; P<0.05) linked to metastasis. Concurrently, prognostic data and gene expression profiles from the TCGA-PRAD cohort were examined, identifying 537 genes significantly associated with patient outcomes. By intersecting these genes with metastasis-related genes from the scRNA-seq analysis database, 25 candidate genes connected to metastasis and prognosis were selected (Figure 2A).

Figure 2 Identification of CCNA2 as a candidate gene in metastatic prostate cancer. (A) Venn diagram illustrating the overlap among metastasis-related genes, prognostic genes from the TCGA-PRAD cohort, and genes linked to metastasis from GEO datasets. (B) Feature importance ranking for the overlapping gene set, as determined by random forest analysis. (C,D) Results from Cox proportional hazards model performed on the candidate genes. (E-J) ROC curves assessing the predictive performance of the selected genes. (K) Prognostic ROC curves for CCNA2 at 1-, 3-, and 5-year intervals. (L) Time-dependent ROC analysis for CCNA2. AUC, area under the curve; CI, confidence interval; FPR, false positive rate; GEO, Gene Expression Omnibus; HR, hazard ratio; PRAD, prostate adenocarcinoma; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas; TPR, true positive rate.

A random forest analysis was subsequently employed to prioritize these candidates. Eight genes—KNL1, KIF2C, SAPCD2, BUB1, DEPDC1B, ASPM, CCNA2, and GTSE1—demonstrated the highest feature importance scores (MeanDecreaseGini > 0.095). Among these, only CCNA2 consistently showed robust predictive accuracy across multiple prognostic timelines, as evidenced by its area under the curve (AUC) for overall survival (OS) and disease-specific survival (DSS) at 1, 3, and 5 years. Based on this integrative bioinformatic screening, CCNA2 was selected as the foremost prognostic gene associated with metastasis (Figure 2B-2L).

The growth and metastatic potential of PCa cells are modulated by CCNA2 expression

The efficacy of different CCNA2-targeting siRNA sequences was evaluated by western blotting. This screen identified siCCNA2#1 and siCCNA2#2 as the most effective constructs for suppressing protein expression (Figure 3A-3D); these were therefore chosen for further functional investigation. In PC3 and DU145 PCa cell lines, silencing CCNA2 expression markedly reduced cellular proliferation (Figure 3E,3F). Additionally, wound-healing and Transwell chamber assays demonstrated that CCNA2 knockdown significantly attenuated the migratory and invasive capacities of both cell lines (Figure 3G-3P).

Figure 3 Silencing CCNA2 suppresses prostate cancer cell proliferation and motility. (A-D) Western blot analysis confirming the protein knockdown efficiency of CCNA2-targeting siRNAs. (E,F) Colony formation assay assessing the effect of CCNA2 knockdown on cellular proliferation (0.1% crystal violet). (G-J) Wound-healing (scratch) assay examining changes in cell migration following CCNA2 silencing. (K-P) Transwell migration and invasion assays analyzing the impact of CCNA2 knockdown on cell motility (0.2–0.5% crystal violet). *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. Ctrl, control; siRNA, small interfering RNA.

Immunofluorescence staining confirms high expression of CCNA2 in PCa tissues

PCa tissue microarrays (ZL-PRC1601) were purchased from Shanghai Zhuoli Biotech Company (Ethics Number: LLS M-15-01). In this study, immunofluorescence technology was used to detect the expression level and localization of CCNA2 (Cyclin A2) protein in PCa tissues and matched adjacent normal tissues, as shown in Figure 4. The results showed strong CCNA2-positive fluorescent signals in PCa tissues, which were mainly localized in the nucleus. In contrast, CCNA2 fluorescence signals were weak or almost undetectable in adjacent normal tissues. These findings indicate that CCNA2 is abnormally overexpressed in PCa and may be involved in tumor initiation and progression. This provides experimental evidence for CCNA2 as a potential diagnostic marker or therapeutic target for PCa.

Figure 4 Immunofluorescence of CCNA2 in prostate cancer tissues and adjacent tissues.

Association of CCNA2 with prognosis and survival in PCa patients

After confirming the function of CCNA2 through cellular experiments, we further conducted a detailed investigation into the impact of this gene on patient prognosis to enhance its clinical translational relevance. Figure 5A illustrates the association between CCNA2 expression levels and clinical outcomes within the TCGA cohort. The visualization comprises three panels: the top section ranks CCNA2 expression from low to high across samples, the middle section plots corresponding survival time and status, and the bottom panel provides a heatmap of expression values. Kaplan-Meier analysis, assessed via the log-rank test, demonstrated a significant survival difference between groups based on CCNA2 expression (Figure 5B; HR 2.05, P<0.001, 95% CI: 1.338–3.142). This identifies CCNA2 as a risk factor, where higher expression correlates with worse prognosis. Furthermore, the prognostic model exhibited strong predictive accuracy for 1-, 3-, and 5-year survival in PRAD, with AUC values of 0.715, 0.626, and 0.568, respectively, as shown in the receiver operating characteristic (ROC) curve (Figure 5C). Additionally, CCNA2 expression was found to be significantly elevated in both paired and unpaired tumor samples compared to normal tissues in the TCGA-PRAD dataset (Figure 5D,5E). Finally, a Sankey diagram (Figure 5F) delineates the relationship between high and low CCNA2 expression levels and clinical parameters, including patient status, T-stage, N-stage, and M-stage.

Figure 5 Development of a CCNA2-based prognostic model. (A) Comparison of CCNA2 expression levels between high- and low-CCNA2 groups. (B) Progression-free survival curves stratified by the expression of CCNA2. (C) A predictive model assessing 1-, 3-, and 5-year survival in PRAD patients. (D,E) CCNA2 expression levels compared between tumor and normal tissues in both matched pairs and unpaired samples from the TCGA-PRAD cohort. (F) Sankey diagram illustrating associations between CCNA2 expression and clinical parameters. ***, P<0.001. AUC, area under the curve; CI, confidence interval; exp, expression; HR, hazard ratio; pM, pathological metastasis; pN, pathological node; PRAD, prostate adenocarcinoma; pT, pathological tumor; TCGA, The Cancer Genome Atlas; TPM, transcripts per million.

Development of prognostic models

After elucidating the role of the single gene, we also constructed a prognostic model using multiple genes, further enhancing the clinical significance of this study. A prognostic model was developed by incorporating eight candidate genes: KNL1, KIF2C, SAPCD2, BUB1, DEPDC1B, ASPM, CCNA2, and GTSE1. The risk score for each patient was calculated using the following formula: Riskscore = (0.4425 × KNL1) + (1.1803 × KIF2C) + (1.1194 × SAPCD2) + (−2.1977 × BUB1) + (0.3256 × DEPDC1B) + (−2.7906 × ASPM) + (3.1411 × CCNA2) + (−0.4514 × GTSE1).

Based on their gene expression profiles, samples were stratified into high-risk and low-risk groups via least absolute shrinkage and selection operator (LASSO) regression. Individuals in the high-risk group demonstrated a significantly worse clinical outcome compared to those in the low-risk group (Figure 6A-6C). The model’s predictive accuracy for 1-, 3-, and 5-year survival in PRAD was evaluated using time-dependent ROC analysis, yielding AUC values of 1.000, 0.789, and 0.827, respectively (Figure 6D,6E). The nomogram integrating the prognostic genes for clinical outcome prediction is shown in Figure 6F.

Figure 6 Development of a prognostic signature based on candidate genes. (A,B) The LASSO coefficient path and cross-validation plots of prostate cancer. (C) A composite visualization shows: the rank-ordered risk scores (top), corresponding survival time and status for each sample (middle), and a heatmap displaying the multi-gene expression profile for the prognostic signature (bottom). (D) Kaplan-Meier survival analysis for the eight-gene panel (KNL1, KIF2C, SAPCD2, BUB1, DEPDC1B, ASPM, CCNA2, GTSE1). (E) Time-dependent ROC curves evaluating the model’s predictive accuracy for 1-, 3-, and 5-year overall survival in PRAD patients. (F) A nomogram integrating the prognostic genes for clinical outcome prediction. AUC, area under the curve; CI, confidence interval; HR, hazard ratio; LASSO, least absolute shrinkage and selection operator; PRAD, prostate adenocarcinoma; ROC, receiver operating characteristic.

Construction of diagnostic models

Given the possibility of multicollinearity among the candidate genes, we recognized that relying on a single modeling approach may be insufficient and could be affected by multicollinearity among predictor genes. To address this limitation and to more rigorously validate the robustness of our findings, we subsequently constructed over 100 alternative prognostic models using diverse machine learning and statistical algorithms. The diagnostic efficacy of KNL1, KIF2C, SAPCD2, BUB1, DEPDC1B, ASPM, CCNA2, and GTSE1 for PCa was evaluated using the expression data from the TCGA-PRAD dataset as the training set and the expression data from the GSE16560, GSE54460, and GSE70769 datasets as the validation set. The results indicated that several algorithm combinations demonstrated robust predictive performance for diagnosing PRAD patients within the training set. To confirm the efficacy of our diagnostic model, the expression patterns of these genes were subsequently validated in the GSE16560, GSE54460, and GSE70769 datasets. Within the verification set, a minority of algorithm combinations yielded suboptimal outcomes, whereas the majority produced superior predictions. Notably, the Stepglm[both]+GBM algorithm combination emerged as the most effective diagnostic model, distinguished by its highest mean AUC value (Figure 7A). For enhanced clarity, we also detailed the gene count encompassed by each algorithm combination (Figure 7B).

Figure 7 Development of diagnostic models incorporating integrated machine learning approaches. (A) The diagnostic performance of various algorithmic pairings in classifying PRAD. (B) The quantities of genetic features included in each distinct model configuration. V1, TCGA-PRAD; V2, GSE16560; V3, GSE54460; V4, GSE70769. AUC, area under the curve; Enet, elastic net; Freq, frequency; GBM, gradient boosting machine; glmBoost, boosted generalized linear model; LASSO, least absolute shrinkage and selection operator; LDA, linear discriminant analysis; plsRglm, partial least squares regression for generalized linear models; PRAD, prostate adenocarcinoma; RF, random forest; Stepglm, stepwise generalized linear model; SVM, support vector machine; XGBoost, extreme gradient boosting.

Evaluating the association between CCNA2 expression and compound sensitivity in metastatic PRAD

To explore the potential interaction between the key gene CCNA2 and therapeutic compounds for PRAD, molecular docking was performed as a computational screening approach. This analysis was conducted using the CB-Dock2 tool, a recognized platform for evaluating molecular interactions. Binding affinity was estimated by the Vina score, where values lower than −5.0 kcal/mol indicate favorable predicted binding, with increasing negativity correlating with greater affinity. The docking simulation evaluated the interaction of CCNA2 with six PRAD-related agents: flutamide, bicalutamide, abiraterone, enzalutamide, darolutamide, and rezvilutamide. The results predicted potential binding conformations between CCNA2 and each of the tested drugs (Figure 8A-8F). However, these findings are purely computational. Therefore, no conclusion can be drawn regarding enhanced pharmacological activity based on these in silico data alone, and the current observations should be considered hypothesis-generating, warranting further experimental validation to determine whether any physiologically meaningful interaction exists.

Figure 8 Molecular docking reveals strong interactions between CCNA2 and prostate cancer therapeutics. The binding affinity of CCNA2 was evaluated with six PRAD-targeted drugs. (A) Flutamide, (B) bicalutamide, (C) abiraterone, (D) enzalutamide, (E) darolutamide, and (F) rezvilutamide. CID, compound identifier; PRAD, prostate adenocarcinoma.

Functional enrichment profiling of CCNA2 in PCa

PCa patients from the TCGA-PRAD cohort were stratified according to their CCNA2 expression into high and low expression groups. This stratification was performed to investigate the biological function of CCNA2 in PCa (Figure 9A,9B). KEGG pathway analysis revealed that CCNA2 is significantly associated with several key processes, including adrenergic signaling in cardiomyocytes, the p53 signaling pathway, and apoptotic pathways across multiple species (Figure 9C,9D). Furthermore, GO enrichment results demonstrate that CCNA2 participates in essential cell division processes such as chromosome segregation, organelle fission, and mitotic nuclear division (Figure 9E,9F).

Figure 9 Functional profiling of CCNA2 in prostate cancer. (A) Differential expression between these subsets is visualized in a volcano plot and a heatmap. (B) The cohort was stratified into CCNA2 high- and low-expression groups. G1, high-expression group of CCNA2; G2, low-expression group of CCNA2. (C,D) Resulting differentially expressed genes were subjected to KEGG pathway enrichment analysis. (E,F) GO functional enrichment analysis. (G,H) Analyzing the correlation between CCNA2 and PI3K/AKT signaling pathway through GSEA. (I) Changes in pathway proteins after knocking down CCNA2. GO, Gene Ontology; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; NES, normalized enrichment score.

Following this, a gene set enrichment analysis (GSEA) was performed, revealing a close association between CCNA2 and the PI3K/AKT signaling pathway (Figure 9G,9H). Upon silencing CCNA2, a marked reduction in the expression of phosphorylated PI3K and phosphorylated AKT was observed; however, the total protein levels of PI3K and AKT remained relatively unchanged (Figure 9I). These findings indicate that CCNA2 may contribute to metastasis in PCa through modulation of the PI3K/AKT pathway.

Discussion

PCa represents the most frequently diagnosed male malignancy globally, with considerable geographical disparities observed in both its occurrence and fatality rates. Globally, this disease constituted roughly 1.6 million newly reported cases and an estimated 366,000 fatalities in 2024, positioning it as the second most prevalent cancer and the leading cause of cancer mortality in men, based on GLOBOCAN 2024 data (1). Developed nations report especially elevated incidence rates, a phenomenon largely attributed to the extensive adoption of prostate-specific antigen (PSA) screening, which has enhanced the capacity for early diagnosis (14,15).

The clinical presentation of PCa is often non-specific, which hinders timely detection and management. A significant number of patients presenting with lower urinary tract symptoms (LUTS) are in fact diagnosed with benign prostatic hyperplasia (BPH) rather than cancer. For instance, while LUTS are frequently linked to PCa, evidence shows this association is weak, as many symptomatic individuals do not have a malignancy (16). This common misperception can result in undue psychological distress and invasive clinical procedures (17). Importantly, research indicates that asymptomatic individuals with raised PSA levels face an increased cancer risk. One key study demonstrated that men with a PSA level of ≥3.0 ng/mL in the absence of voiding symptoms were found to have a substantially higher likelihood of harboring PCa (18). These observations underscore the critical role of routine screening for enabling earlier identification of this disease, even when typical symptoms are not present.

The advent of single-cell methodologies has profoundly advanced the comprehension of PRAD, offering unprecedented resolution in delineating its heterogeneous nature. By analyzing transcriptomic profiles at the cellular level, scRNA-seq reveals intricate dynamics of the tumor microenvironment (TME) and the cross-talk between diverse cellular components. These insights are crucial for discovering new biomarkers and informing treatment strategies. Research illustrates that scRNA-seq can clarify how intratumoral heterogeneity contributes to metastasis, enabling detailed studies of identification of distinct cell subsets linked to unfavorable prognosis (19,20).

The integration of multiple biological data types—genomics, transcriptomics, proteomics, and metabolomics—through multi-omics analysis enables a more holistic view of disease mechanisms, thereby supporting clinical decision-making. When augmented by machine learning (ML) and artificial intelligence (AI), this approach becomes a powerful strategy for uncovering latent patterns and associations in complex diseases such as cancer. Within oncology, the application of AI to multi-omics data is particularly significant. As Nicora et al. noted, ML is increasingly employed to delineate cancer subtypes and discover biomarkers from integrated datasets, a crucial step for precision medicine (21). This capability is further demonstrated by Wang et al., who show that graph convolutional networks can effectively combine multi-omics information for classifying patients and identifying biomarkers, highlighting the potential of such technologies to refine diagnostics and treatments (22). The capacity of AI to process vast datasets allows researchers to decipher complex interactions across different omics layers, which is fundamental for understanding cancer as a multi-factorial pathology (23).

Furthermore, integrated multi-omics analyses have proven effective in revealing novel therapeutic targets. For instance, in head and neck cancer, Costa et al. described how distinct co-expression networks derived from multi-omic data can pinpoint new targets for intervention, underscoring the necessity of combining various omics perspectives to gain deeper biological insights (24). This view is corroborated by Subramanian et al., who maintain that multi-omics integration is essential for progressing systemic biological understanding and developing targeted therapies (25).

The gene CCNA2 has garnered significant research attention for its potential as a prognostic indicator and its contribution to oncogenesis. In estrogen receptor-positive breast cancer, elevated CCNA2 expression has demonstrated notable predictive value for various clinical outcomes, including rates of distant metastasis-free, disease-free, recurrence-free survival, and OS. Furthermore, its expression is strongly linked to metastasis, indicating an influence on treatment efficacy (8,26). Within colorectal cancer (CRC), integrated bioinformatic analyses of multiple datasets have established CCNA2 as a promising novel biomarker, highlighting its role in controlling cellular proliferation and programmed cell death (27). Supporting this, research by Liu et al. into the molecular drivers of CRC advancement has detailed the participation of pivotal genes and pathways, including CCNA2, in disease pathogenesis (28). These collective findings underscore the gene’s critical function in CRC and its viability as a potential therapeutic target.

The broader implications of CCNA2 as an immunological biomarker have also been explored. Jiang et al. examined its function within the TME and its effect on treatment responses in various malignancies, proposing that CCNA2 can modulate immune cell infiltration and tumor-immune system crosstalk, thereby enhancing its prognostic and therapeutic relevance (29). In summary, these studies reveal the multifunctional nature of CCNA2 in oncology, encompassing prognostic value, cell cycle control, epithelial-mesenchymal transition (EMT), immune modulation, and metastatic mechanism. Its consistent identification across diverse cancers positions CCNA2 as a possible universal biomarker and target, warranting deeper investigation into its molecular actions and regulatory circuits during tumor progression.

This study presents a systematic multi-omics investigation into the role of CCNA2 in metastasis and tumor progression in PRAD.

First, starting from the key factor influencing patient prognosis—metastasis in PCa—we employed integrated bioinformatics approaches, including single-cell RNA sequencing, bulk transcriptomic data from TCGA and GEO, and multiple machine learning algorithms, to identify genes associated with metastasis and clinical outcomes.

Next, using single-cell sequencing, we localized the genes associated with metastasis in PCa at the cellular level, revealing that CCNA2 is enriched in specific cell populations (e.g., epithelial and stem-like cells), with notable compositional shifts between the metastatic and non-metastatic groups.

Subsequently, we validated the most significant gene through basic experiments: CCNA2 knockdown in PCa cell lines significantly suppressed proliferation, migration, and invasion, confirming its functional role.

To further enhance the clinical applicability of the experimental results, we constructed a multigene prognostic model incorporating CCNA2 together with seven other candidate genes, and developed a machine learning-based diagnostic model that demonstrated high predictive accuracy across multiple independent validation cohorts. Additionally, we performed molecular docking between the CCNA2 protein structure and endocrine therapy drugs for PCa, and conducted immunofluorescence analysis on pathological tissues and adjacent paracancerous tissues from PCa patients, further confirming the significance of CCNA2.

Finally, we performed mechanistic experiments to elucidate the functions of these genes, demonstrating that CCNA2 modulates the PI3K/AKT signaling pathway, thereby driving tumor progression and metastasis. CCNA2 promotes tumor progression through PI3K/AKT pathway activation via two interconnected mechanisms. First, the CDK2/cyclin A2 complex directly phosphorylates AKT at its carboxyl-terminal regulatory sites. Liu et al. demonstrated that AKT activity fluctuates across the cell cycle, mirroring cyclin A expression; mechanistically, phosphorylation of S477 and T479 at the AKT extreme carboxy terminus by CDK2/cyclin A promotes AKT activation by facilitating or functionally compensating for S473 phosphorylation (30). Second, CCNA2 participates in a broader regulatory network with PI3K/AKT signaling. A recent network pharmacology and experimental validation study by Zhang et al. identified CCNA2 as one of five core targets (alongside SRC, HRAS, CDK2, and AKT2) enriched in the PI3K-AKT signaling pathway; in vitro experiments confirmed that CCNA2 mRNA and protein levels were upregulated following pathway stimulation, and molecular docking demonstrated stable binding between CCNA2 and pathway-related compounds (31).

The present research provides an integrative evaluation of metastasis-related genes to clarify their influence on prognosis in PRAD patients. Among these genes, CCNA2 was characterized as the most pivotal prognostic marker. Prior multi-omics investigations and experimental data have established that CCNA2 promotes cell-cycle progression at the G1/S and G2/M checkpoints, thereby modifying the TME across multiple cancer types (32). Additional studies indicate that in pancreatic adenocarcinoma, CCNA2 modulates tumor progression via the p53 signaling cascade (33,34). However, the particular function of CCNA2 as a metastasis-related gene in PRAD had not been explored previously. Our investigation, employing PRAD tissue microarrays and in vitro cellular assays, clarifies the substantial contribution of CCNA2 in this malignancy.

Conclusions

This investigation integrates single-cell profiling with multiple machine learning approaches to reveal important connections between metastasis-related genes and clinical outcomes in PRAD. In particular, CCNA2 promotes tumor progression through the PI3K/AKT signaling axis. By proposing new biomarkers and possible targets for treatment, these findings contribute meaningful perspectives for improving early diagnosis and clinical management of PRAD patients.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD and MDAR reporting checklists. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0511/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0511/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0511/prf

Funding: This study was supported by Natural Science Foundation of Heilongjiang Province (Nos. LH2019H030 and PL2024H044).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0511/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. The 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. Crocetti E, Ravaioli A, Falcini F, et al. The burden of prostate cancer and metastatic prostate cancer in Italy in 2024. Eur J Cancer Prev 2025;34:387-91. [Crossref] [PubMed]
2. Lim J, Malek R, Jr S, et al. Prostate cancer in multi-ethnic Asian men: Real-world experience in the Malaysia Prostate Cancer (M-CaP) Study. Cancer Med 2021;10:8020-8. [Crossref] [PubMed]
3. Wu XC, Yu YZ, Zuo YZ, et al. Identification of UAP1L1 as a critical factor for prostate cancer and underlying molecular mechanism in tumorigenicity. J Transl Med 2022;20:91. [Crossref] [PubMed]
4. Lin H, Shen Z, Liu H, et al. Upregulation of GRIM-19 augments the sensitivity of prostate cancer cells to docetaxel by targeting Rad23b. Clin Exp Pharmacol Physiol 2020;47:76-84. [Crossref] [PubMed]
5. Chen X, Wei C, Zhao J, et al. Carnosic acid: an effective phenolic diterpenoid for prevention and management of cancers via targeting multiple signaling pathways. Pharmacol Res 2024;206:107288. [Crossref] [PubMed]
6. Wang Y, Wu N, Wang K, et al. Specific classification and new therapeutic targets for neuroendocrine prostate cancer: A patient-based, diagnostic study. Front Genet 2022;13:955133. [Crossref] [PubMed]
7. Zhang Y, Zheng D, Zhou T, et al. Androgen deprivation promotes neuroendocrine differentiation and angiogenesis through CREB-EZH2-TSP1 pathway in prostate cancers. Nat Commun 2018;9:4080. [Crossref] [PubMed]
8. Zhang L, Wang S, Wang L. Comprehensive analysis pinpoints CCNA2 as a prognostic and immunological biomarker in non-small cell lung cancer. BMC Pulm Med 2025;25:14. [Crossref] [PubMed]
9. Zhao X, Ha M, Zhou L, et al. Berberine diminishes the malignant progression of non-small cell lung cancer cells by targeting CDCA5 and CCNA2. J Nat Med 2025;79:530-42. [Crossref] [PubMed]
10. Gonzalez-Bosquet J, Shahi M, Yadav S, et al. ECPPF stratification identifies occult high-risk subgroups in stage I, grade 1 or 2, ≤50% invasive endometrial cancer: Candidates for adjuvant therapy. Gynecol Oncol 2025;196:113-20. [Crossref] [PubMed]
11. Wang Y, Wang J, Liu J, et al. Immune-related diagnostic markers for benign prostatic hyperplasia and their potential as drug targets. Front Immunol 2024;15:1516362. [Crossref] [PubMed]
12. Fang W, Liu X, Maiga M, et al. Digital PCR for Single-Cell Analysis. Biosensors (Basel) 2024;14:64. [Crossref] [PubMed]
13. Zhou X, Wu H, Wen H, et al. Advances in Single-Cell Printing. Micromachines (Basel) 2022;13:80. [Crossref] [PubMed]
14. Steele CB, Li J, Huang B, et al. Prostate cancer survival in the United States by race and stage (2001-2009): Findings from the CONCORD-2 study. Cancer 2017;123:5160-77. [Crossref] [PubMed]
15. Iyengar S, Hall IJ, Sabatino SA. Racial/Ethnic Disparities in Prostate Cancer Incidence, Distant Stage Diagnosis, and Mortality by U.S. Census Region and Age Group, 2012-2015. Cancer Epidemiol Biomarkers Prev 2020;29:1357-64.
16. Palmstedt E, Månsson M, Kollberg KS, et al. How a population-based cohort of men estimate lifetime risk of prostate cancer in a survey before entering a prostate cancer screening trial in Sweden? BMJ Open 2024;14:e083562. [Crossref] [PubMed]
17. Haq MS, Thomas PM, Almonte M, et al. Prostate Cancer Presenting With an Unusual Presentation of Rectal Pain and Bleeding. Cureus 2021;13:e18766. [Crossref] [PubMed]
18. Frånlund M, Carlsson S, Stranne J, et al. The absence of voiding symptoms in men with a prostate-specific antigen (PSA) concentration of ≥3.0 ng/mL is an independent risk factor for prostate cancer: results from the Gothenburg Randomized Screening Trial. BJU Int 2012;110:638-43. [Crossref] [PubMed]
19. Song H, Weinstein HNW, Allegakoen P, et al. Single-cell analysis of human primary prostate cancer reveals the heterogeneity of tumor-associated epithelial cell states. Nat Commun 2022;13:141. [Crossref] [PubMed]
20. Wang Z, Su Y, Zhao L, et al. Pan-Cancer Single-Cell Transcriptomic Analysis Reveals Divergent Expression of Embryonic Proangiogenesis Gene Modules in Tumorigenesis. Cancer Med 2024;13:e70373. [Crossref] [PubMed]
21. Nicora G, Vitali F, Dagliati A, et al. Integrated Multi-Omics Analyses in Oncology: A Review of Machine Learning Methods and Tools. Front Oncol 2020;10:1030. [Crossref] [PubMed]
22. Wang T, Shao W, Huang Z, et al. MOGONET integrates multi-omics data using graph convolutional networks allowing patient classification and biomarker identification. Nat Commun 2021;12:3445. [Crossref] [PubMed]
23. Arjmand B, Hamidpour SK, Tayanloo-Beik A, et al. Machine Learning: A New Prospect in Multi-Omics Data Analysis of Cancer. Front Genet 2022;13:824451. [Crossref] [PubMed]
24. Costa RL, Boroni M, Soares MA. Distinct co-expression networks using multi-omic data reveal novel interventional targets in HPV-positive and negative head-and-neck squamous cell cancer. Sci Rep 2018;8:15254. [Crossref] [PubMed]
25. Subramanian I, Verma S, Kumar S, et al. Multi-omics Data Integration, Interpretation, and Its Application. Bioinform Biol Insights 2020;14:1177932219899051. [Crossref] [PubMed]
26. Gong K, Zhou H, Liu H, et al. Identification and Integrate Analysis of Key Biomarkers for Diagnosis and Prognosis of Non-Small Cell Lung Cancer Based on Bioinformatics Analysis. Technol Cancer Res Treat 2021;20:15330338211060202. [Crossref] [PubMed]
27. Li J, Zhou L, Liu Y, et al. Comprehensive Analysis of Cyclin Family Gene Expression in Colon Cancer. Front Oncol 2021;11:674394. [Crossref] [PubMed]
28. Liu X, Liu X, Qiao T, et al. Identification of crucial genes and pathways associated with colorectal cancer by bioinformatics analysis. Oncol Lett 2020;19:1881-9. [Crossref] [PubMed]
29. Jiang A, Zhou Y, Gong W, et al. CCNA2 as an Immunological Biomarker Encompassing Tumor Microenvironment and Therapeutic Response in Multiple Cancer Types. Oxid Med Cell Longev 2022;2022:5910575. [Crossref] [PubMed]
30. Liu P, Begley M, Michowski W, et al. Cell-cycle-regulated activation of Akt kinase by phosphorylation at its carboxyl terminus. Nature 2014;508:541-5. [Crossref] [PubMed]
31. Zhang LC, Liu YN, La XQ, et al. The potential mechanism of Neu5Gc inducing colorectal cancer based on network pharmacology and experimental validation. Naunyn Schmiedebergs Arch Pharmacol 2023;396:705-18. [Crossref] [PubMed]
32. Rosendahl J, Zarna C, Håkansson J, et al. Gene-Expression Analysis of Human Fibroblasts Affected by 3D-Printed Carboxylated Nanocellulose Constructs. Bioengineering (Basel) 2023;10:121. [Crossref] [PubMed]
33. Jiang P, Zhang M, Gui L, et al. Expression patterns and prognostic values of the cyclin-dependent kinase 1 and cyclin A2 gene cluster in pancreatic adenocarcinoma. J Int Med Res 2020;48:300060520930113. [Crossref] [PubMed]
34. Zhang Y, Yang J, Wang X, et al. GNG7 and ADCY1 as diagnostic and prognostic biomarkers for pancreatic adenocarcinoma through bioinformatic-based analyses. Sci Rep 2021;11:20441. [Crossref] [PubMed]