An O-glycosylation-based prognostic signature for biochemical recurrence in prostate cancer identifies GALNTL6 as a potential promoter of malignant phenotypes

Introduction

Prostate cancer (PCa) places a substantial disease burden on men and remains an important cause of cancer-related death (1). Although localized prostate cancer generally has a relatively good prospect, a considerable number of cases will develop into metastatic disease and castration-resistant prostate cancer (CRPC), and this type of cancer is generally still incurable (2,3). The key challenge in clinical treatment is to predict and prevent biochemical recurrence (BCR). BCR usually occurs before the cancer spreads to other parts of the body, which means the treatment has failed (4,5). Existing prognostic methods are not sufficient for early identification of high-risk patients (6,7). It is now highly necessary to identify new molecular biomarkers to understand the fundamental mechanisms that drive disease development, so as to implement personalized treatment plans.

Protein glycosylation is one of the most common post-translational alterations, which is crucial for regulating communication between cells, cell adhesion, signal transduction, and immune responses (8-11). Protein glycosylation includes N-linked and O-linked forms. O-glycosylation begins with GALNT-mediated addition of N-acetyl galactosamine (GalNAc) to serine or threonine residues in the Golgi apparatus (12-14). This distinct subcellular localization contributes to the unique regulatory dynamics and functional diversity of O‑glycosylation in cellular processes.

To elaborate, O-glycosylation is increasingly regarded as a key factor in cancer development (15-17). Abnormal O-glycosylation is related to tumor invasion, metastasis, and resistance to various cancer treatments. Accumulating studies have highlighted aberrant O-glycosylation as a potential therapeutic target in gastrointestinal cancers because of its role in tumor progression (18-20). However, the contribution of O-glycosylation to PCa progression and BCR, as well as its clinical relevance, remains insufficiently defined.

In this study, we constructed an O‑glycosylation‑based prognostic model for PCa BCR that performed robustly across independent cohorts. GALNTL6, an understudied GALNT family member, was identified as the most critical predictor. Functional validation through gain-of-function experiments in PCa cells demonstrated that GALNTL6 overexpression promotes malignant phenotypes. Transcriptomic analysis following its overexpression revealed activation of the PI3K‑AKT pathway and upregulation of antioxidant response genes, suggesting its role in driving these processes. Thus, interference with the GALNTL6-mediated PI3K-AKT-antioxidant axis may provide a potential therapeutic avenue for reducing recurrence in advanced PCa. We present this article in accordance with the MDAR and TRIPOD reporting checklists (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0315/rc).

Methods

Data acquisition and processing

Transcriptomic and clinical data were collected from seven publicly available prostate cancer cohorts, including The Cancer Genome Atlas Prostate Adenocarcinoma (TCGA-PRAD), GSE107299, DKFZ, E-MTAB-6128, GSE21034, GSE54460, GSE70769, and GSE94767. These datasets were retrieved from TCGA, cBioPortal, Array Express, and the GEO database. For model development and external validation, gene expression and recurrence-related clinical data were obtained from these cohorts. Gene expression values were log2-transformed before downstream analyses. For construction of the O-glycosylation scores (OGs) model, processed expression values of O-glycosylation-related regulators in prostate cancer samples with available recurrence data were used. The OGs was not calculated from direct tumor-versus-normal expression differences, but from prognostic modeling within prostate cancer cohorts.

In the TCGA-PRAD cohort, clinicopathological and follow-up data were further collected. Staging was defined according to the 8th edition of the American Joint Committee on Cancer (AJCC) system. TCGA-PRAD was used as the training cohort, and the remaining independent cohorts were used for external validation. Thirty established O-glycosylation regulators were selected from an MSigDB gene set related to O-glycan biosynthesis for prognostic analysis. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Somatic mutation analysis

We characterized and visualized somatic mutation profiles of the 30 O-glycosylation regulators using the maftools package.

Signature construction and generation of OGs

Prognostic modeling was performed using ten candidate machine-learning approaches, including random survival forest, least absolute shrinkage and selection operator (LASSO), stepwise Cox regression, CoxBoost, plsRcox, SuperPC, elastic net, ridge regression, gradient boosting machine (GBM), and survival-support vector machine (SVM). A total of 117 candidate algorithms or algorithm combinations were evaluated in the TCGA cohort under leave-one-out cross-validation (LOOCV). Their performance was further tested in six independent cohorts, and Harrell’s concordance index, also known as the C-index, was used to select the optimal model. The C-index measures the discriminative ability of a survival prediction model by evaluating how well the model correctly ranks patients according to their recurrence risk. A C-index of 0.5 indicates predictive performance similar to random chance, whereas a C-index of 1.0 indicates perfect discrimination. The CoxBoost plus elastic net model with α=0.7 achieved the best overall performance and was therefore selected to generate the final O-glycosylation signature. CoxBoost was used for feature selection, and elastic net Cox regression was subsequently applied to determine the final model coefficients. The OGs value for each patient was calculated using the following formula: OGs = Σ βi × Expi, where βi represents the model coefficient of each selected O-glycosylation-related gene, and Expi represents the corresponding normalized gene expression value. Thus, the OGs score represented a weighted prognostic index rather than an unweighted average of GALNT family member expression. Each selected gene contributed to the final score according to its model coefficient, which reflected its prognostic weight in the final model. To classify patients into high- and low-OGs groups, the optimal cutoff value was determined separately in each cohort using the surv_cutpoint function in the R package survminer, with BCR-free survival as the outcome variable. Patients with OGs values greater than the cohort-specific optimal cutoff were assigned to the high-OGs group, whereas those with OGs values equal to or lower than the cutoff were assigned to the low-OGs group. Kaplan-Meier analysis was then performed to compare BCR-free survival between the two groups.

Evaluation of the O-glycosylation signature and nomogram construction

The prognostic performance of the O-glycosylation signature was evaluated using forest plots and time-dependent receiver operating characteristic (ROC) analysis. A nomogram integrating clinicopathological variables and the derived risk score was constructed to estimate BCR risk at 1, 3, and 5 years. Calibration and ROC analyses were used to assess model performance.

Drug sensitivity prediction based on the signature

The oncoPredict package was used to estimate drug half maximal inhibitory concentration (IC₅₀) values in the TCGA-PRAD cohort with Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Therapeutics Response Portal (CTRP) as reference datasets. Pearson correlation analysis was then used to assess the association between risk scores and predicted IC₅₀ values.

Cell Lines, cell culture, and GALNTL6 expression screening

Human prostate cancer cell lines DU145, PC3, 22Rv1, LNCaP, C4-2, and C4-2B, as well as the normal prostate epithelial cell line BPH-1 and the HEK293T cell line, were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). All cell lines were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin solution at 37 ℃ in a humidified incubator with 5% CO₂. Total RNA was collected from all cell lines under identical culture conditions, and GALNTL6 mRNA expression was assessed by quantitative real-time polymerase chain reaction (qRT-PCR).

GALNTL6 small interfering RNA (siRNA) design, screening, and transfection

Three siRNAs targeting human GALNTL6 (si-GALNTL6-1, si-GALNTL6-2, and si-GALNTL6-3), along with a non-targeting negative control siRNA (si-NC), were designed and synthesized by Beijing Tsingke Biotech Co., Ltd., Guangzhou Branch. The sense and antisense sequences of all siRNAs are provided in Table S1. Cells were seeded 1 day before transfection to achieve approximately 60–80% confluence at the time of transfection. siRNAs were transfected into DU145 and 22Rv1 cells using Lipofectamine RNAiMAX according to the manufacturer’s instructions, at a final concentration of 50 nM. The medium was replaced with fresh complete medium 24 h after transfection. Cells were collected 48 h after transfection for qRT-PCR, Western blotting, or subsequent functional assays. The knockdown efficiencies of the three GALNTL6-targeting siRNAs were initially evaluated by qRT-PCR and/or Western blotting. Subsequent functional experiments were performed using the siRNA with the highest knockdown efficiency, and consistent phenotypic effects were confirmed using at least two independent siRNAs. All experiments were independently repeated at least three times.

Western blot analysis

Antibodies against GALNTL6 (1:1,000; bs-16229R, BIOSS, Beijing, China) and β-actin (1:5,000; 20536-1-AP, Proteintech, Chicago, IL, USA) were used for Western blotting, with β-actin as the loading control. Signal intensity of each band was determined with ImageJ (National Institutes of Health, USA).

qRT-PCR

Total RNA was extracted from cultured cells using TRIzol reagent according to the manufacturer’s instructions. RNA concentration and purity were assessed using a NanoDrop spectrophotometer. One microgram of total RNA from each sample was reverse-transcribed into complementary DNA (cDNA) using HiScript III RT SuperMix for qPCR (+gDNA wiper) (Vazyme, Nanjing, China; Cat. No. R323-01).

qRT-PCR was performed using SYBR Green qPCR Mix on a real-time PCR instrument. Each 20-µL reaction contained 10 µL of 2× SYBR Green qPCR Mix, 0.4 µL each of forward and reverse primers (10 µM), 2 µL of cDNA template, and 7.2 µL of nuclease-free water. The amplification protocol consisted of an initial denaturation step at 95 ℃ for 30 s, followed by 40 cycles of 95 ℃ for 5 s and 60 ℃ for 30 s. Melt-curve analysis was performed after amplification to confirm product specificity.

ACTB (β-actin) was used as the internal reference gene. Each sample was analyzed in triplicate. Relative GALNTL6 mRNA expression was calculated using the 2^−ΔΔCt method, where ΔCt = Ct(target gene) − Ct(reference gene), and ΔΔCt = ΔCt(experimental group) − ΔCt(control group). Primer sequences are listed in Table S2. All qRT-PCR experiments were performed using at least three independent biological replicates.

Construction of GALNTL6-overexpressing lentiviral vectors, lentiviral packaging, and establishment of stable cell lines

The full-length human GALNTL6 coding sequence [RefSeq transcript NM_001034845; coding sequence (CDS) length, 1,806 bp] was cloned downstream of the cytomegalovirus (CMV) promoter in a lentiviral overexpression vector to generate the GALNTL6 overexpression vector. After sequence verification, the GALNTL6 CDS was cloned downstream of the CMV promoter in a pLVX-C-FLAG-PGK-Puro backbone to generate the GALNTL6 overexpression vector (GALNTL6-OE). The corresponding empty vector (Vector/NC) was used as a negative control. Details regarding the vector backbone, cloning sites, tag information, and sequence verification are provided in Table S3.

Lentiviruses were produced using HEK293T packaging cells. HEK293T cells were seeded in 10-cm dishes and transfected at approximately 70–80% confluence with the GALNTL6-OE vector or empty vector, together with the packaging plasmid psPAX2 and envelope plasmid pMD2.G, using Lipofectamine 3000 (Invitrogen). The mass ratio of transfer vector, psPAX2, and pMD2.G was 4:3:1, corresponding to 10 µg, 7.5 µg, and 2.5 µg of plasmid DNA, respectively, per 10-cm dish. The culture medium was replaced with fresh complete medium 6 h after transfection. Lentiviral supernatants were collected at 48 and 72 h after transfection and filtered through a 0.45-µm filter before use for target-cell transduction. DU145 and 22Rv1 cells were transduced with GALNTL6-OE or control lentiviruses at a multiplicity of infection (MOI) of 10 in the presence of 8 µg/mL polybrene. The medium was replaced with fresh complete medium 24 h after infection. At 48 h after infection, cells were subjected to selection with 2 µg/mL puromycin for 15 days. Surviving cells were collected after complete death of the uninfected control cells to establish stable GALNTL6-overexpressing cell lines. GALNTL6 overexpression was validated by qRT-PCR and Western blotting. All in vitro experiments were performed with at least three independent biological replicates.

Cell-based functional assays and statistical analysis

Cell proliferation assay [Cell Counting Kit-8 (CCK-8)]

At 24 h after transfection, DU145 and 22Rv1 cells were seeded in 96-well plates at a density of 2×10³ cells per well, with five technical replicates per group. At 0, 24, 48, 72, and 96 h after seeding, 10 µL of CCK-8 reagent (Cat. No. MA0218, Meilunbio, Dalian, China) was added to each well, followed by incubation at 37 ℃ for 2 h. Absorbance was measured at 450 nm using a microplate reader. Blank wells containing medium and CCK-8 reagent without cells were included, and blank-corrected optical density values were used to reflect relative cell proliferative capacity.

Colony formation assay

At 24 h after transfection, DU145 and 22Rv1 cells were seeded into 6-well plates at a density of 500 cells per well. Cells were cultured in complete medium for 10–14 days, with medium changed every 3 days. Once visible colonies had formed, cells were fixed with 4% paraformaldehyde for 15 min and stained with 0.1% crystal violet for 20 min. After washing and air-drying, whole-well images were obtained. A colony containing more than 50 cells was defined as one colony. Colony numbers were quantified using ImageJ software.

Transwell migration and invasion assays

Cell migration and invasion were assessed using 24-well Transwell chambers with 8-µm pore-size inserts (Corning). For migration assays, 5×10⁴ cells suspended in 200 µL serum-free RPMI-1640 medium were added to the upper chamber, while 600 µL complete medium containing 10% FBS was added to the lower chamber as a chemoattractant. For invasion assays, the upper membrane was precoated with 50 µL Matrigel (Cat. No. 354234, Corning Incorporated, Corning, NY, USA) diluted 1:8 in serum-free RPMI-1640 medium and incubated at 37 ℃ for 1 h to allow gel formation. Subsequently, 1×10⁵ cells suspended in 200 µL serum-free medium were added to the upper chamber, and 600 µL complete medium containing 10% FBS was added to the lower chamber. Both migration and invasion assays were incubated for 24 h at 37 ℃ in a humidified incubator containing 5% CO₂. After incubation, non-migrated or non-invaded cells on the upper surface of the membrane were gently removed using cotton swabs. Cells on the lower surface were fixed with 4% paraformaldehyde for 15 min and stained with 0.1% crystal violet for 20 min. Five randomly selected, non-overlapping fields were imaged for each insert under an inverted microscope at ×100 magnification. The numbers of migrated or invaded cells were counted using ImageJ software, and the average cell number across the five fields was used for statistical analysis. All images were acquired using identical microscope settings and exposure conditions.

Biological replicates and statistical analysis

All cell-based assays were performed in three independent biological replicates. Quantitative data are presented as mean ± standard deviation (SD). CCK-8 time-course data were analyzed using two-way analysis of variance (ANOVA) followed by Šídák’s multiple-comparisons test. Colony numbers and the numbers of migrated or invaded cells were compared between two groups using an unpaired two-tailed Student’s t-test. For comparisons among three or more groups, one-way ANOVA followed by Tukey’s multiple-comparisons test was used. For data that did not meet the assumptions of normality, the Mann-Whitney U test or Kruskal-Wallis test was applied, as appropriate. Statistical analyses were performed using GraphPad Prism (version 9.0). All tests were two-sided, and P<0.05 was considered statistically significant.

Statistical analysis

Statistical analyses and figure generation were performed in R (version 4.0.5). Pearson correlation was used for continuous variables. Group comparisons were conducted using the Chi-squared test, Student’s t-test, or Wilcoxon rank-sum test, as appropriate. Repeated-measures ANOVA was used for longitudinal data. Optimal cutoff values for OGs-based stratification were determined separately in each cohort using the surv_cutpoint function of the R package survminer, with BCR-free survival as the outcome variable. Survival analysis, Cox regression, and time-dependent ROC analysis were performed using the survival and timeROC packages. P<0.05 was considered statistically significant.

Results

Landscape of O-glycosylation regulators in PCa

In the TCGA-PRAD cohort, mutation analysis of 30 O-glycosylation regulators detected 23 alterations across 480 PCa samples, yielding an overall mutation frequency of 1.67% (Figure 1A). The genes GALNT12 and GALNT10 exhibited the highest mutation frequency, and missense and synonymous mutations were the most common types. CNV in the O-glycosylation regulators was observed at a high frequency (Figure 1B,1C), and the correlation between O-glycosylation regulators was illustrated (Figure 1D). These observations imply that abnormal O-glycosylation regulator patterns may participate in PCa progression and could be of potential therapeutic interest. Because these regulators showed heterogeneous genomic and transcriptional patterns, we did not use their raw expression differences directly as the OGs score. Instead, this heterogeneity motivated the subsequent machine-learning-based construction of a weighted prognostic signature. We then assessed the clinical significance of regulator expression and found that higher GALNTL6 expression was associated with BCR, increased prostate-specific antigen (PSA) levels, and more advanced disease stage, as illustrated by the annotated heatmap (Figure 1E).

Figure 1 Genomic and transcriptional features of O-glycosylation regulators in prostate cancer. (A) Among 480 TCGA-PRAD patients, 23 (4.79%) showed genetic alterations in 30 O-glycosylation regulators. Each column denotes a single sample. TMB is displayed in the upper bar plot, regulator-specific alteration frequencies are listed on the right, and mutation classes are summarized in the right-hand bar plot. (B) Genomic locations of CNV events involving O-glycosylation regulators across chromosomes. (C) CNV alteration frequencies of the O-glycosylation regulators. (D) Correlation matrix illustrating the relationships among O-glycosylation regulators. (E) Heatmap depicting the associations between regulator expression and clinicopathological variables, including BCR status, Gleason score, PSA level, pN stage, pT stage, and age. BCR, biochemical recurrence; CNV, copy number variation; N, lymph node; pN, pathological lymph node; PSA, prostate-specific antigen; pT, pathological tumor; TCGA-PRAD, The Cancer Genome Atlas Prostate Adenocarcinoma; T, tumor; TMB, tumor mutation burden.

Development of an O-glycosylation signature by integrative machine learning

Based on the selected regulators, we derived a consensus OGs using an integrative machine learning strategy. In the TCGA cohort, 117 candidate models were generated and then evaluated across six external validation cohorts using Harrell’s C-index for comparison (Figure 2A). In this analysis, the C-index was used to quantify how accurately each model ranked patients according to their BCR risk; therefore, a higher mean C-index indicated better and more generalizable prognostic discrimination across cohorts. In this analysis, the C-index was used to quantify how accurately each model ranked patients according to their BCR risk; therefore, a higher mean C-index indicated better and more generalizable prognostic discrimination across cohorts. In this analysis, the C-index was used to quantify how accurately each model ranked patients according to their BCR risk; therefore, a higher mean C-index indicated better and more generalizable prognostic discrimination across cohorts. In this analysis, the C-index was used to quantify how accurately each model ranked patients according to their BCR risk; therefore, a higher mean C-index indicated better and more generalizable prognostic discrimination across cohorts. In this analysis, the C-index was used to quantify how accurately each model ranked patients according to their BCR risk; therefore, a higher mean C-index indicated better and more generalizable prognostic discrimination across cohorts. Among all candidate models, the CoxBoost plus elastic net model with α=0.7 yielded the highest mean C-index (0.61) and was therefore selected as the optimal model. CoxBoost was first used for feature selection, and elastic net Cox regression with α=0.7 was subsequently applied under the LOOCV framework to construct the final prognostic model and determine the model coefficients (Figure 2B,2C). The final OGs value for each patient was calculated by multiplying the normalized expression level of each of the 20 selected genes by its corresponding model coefficient and then summing these products (Figure 2D). Patients in each cohort were stratified into high- and low-OGs groups according to the cohort-specific optimal cutoff derived from BCR-free survival analysis. In the TCGA cohort, higher OGs were associated with poorer BCR and more advanced pathological features, including pathological tumor (pT), pathological lymph node (pN), and pathological metastasis (pM) stage (Figure 2E). In the training cohort and all six validation cohorts, BCR-free survival was consistently worse among patients classified as high risk than among those classified as low risk (all P<0.01; Figure 2F).

Figure 2 Development and validation of the consensus O-glycosylation signature through integrative machine learning. (A) C-index values for different machine-learning algorithm combinations across the seven cohorts. The C-index measures the discriminative ability of each survival prediction model, with higher values indicating better performance in ranking patients according to BCR risk. (B) CoxBoost analysis results are presented. (C) Enet (α=0.7) regression was performed to obtain the optimal value using the LOOCV framework as illustrated. (D) Regression coefficients of the 20-gene O-glycosylation signature derived from the final model. OGs was calculated as the weighted sum of normalized expression values of the selected genes multiplied by their corresponding model coefficients. (E) Donut chart displaying the distribution of OGs, BCR status, and clinicopathological stages in the TCGA-PRAD cohort. (F) Kaplan-Meier survival curves comparing BCR-free survival between high- and low-OGs groups in the TCGA-PRAD training cohort and seven independent validation cohorts. Log rank test P values are shown. BCR, biochemical recurrence; cM, clinical metastasis; coef, coefficient; DKFZ, German Cancer Research Center; GBM, gradient boosting machine; LASSO, least absolute shrinkage and selection operator; LOOCV, leave-one-out cross-validation; M, metastasis; N, lymph node; OGs, O-glycosylation scores; pN, pathological lymph node; PSA, prostate-specific antigen; pT, pathological tumor; SVM, support vector machine; T, tumor; TCGA-PRAD, The Cancer Genome Atlas Prostate Adenocarcinoma.

Prognostic performance of the OGs

Across the DKFZ, E-MTAB-6128, GSE21034, and GSE54460 cohorts, univariate Cox regression indicated that higher OGs values were significantly associated with BCR (Figure 3A). The estimated hazard ratios were 1.579 (P=0.02), 2.120 (P=0.02), 1.482 (P=0.005), and 1.580 (P=0.03), respectively. Consistent with these findings, time-dependent receiver operating characteristic (ROC) analysis showed that the OGs retained predictive utility across multiple time points. In this study, time-dependent ROC analysis was used to evaluate the ability of the O-glycosylation signature to discriminate patients who developed BCR from those who remained recurrence-free at specific time points. The area under the ROC curve (AUC) was used as a quantitative measure of predictive discrimination, with higher AUC values indicating better predictive performance. The corresponding 1-, 3-, and 5-year AUCs were 0.727, 0.672, and 0.629 for DKFZ; 0.679, 0.692, and 0.693 for E-MTAB-6128; 0.697, 0.638, and 0.591 for GSE21034; and 0.571, 0.626, and 0.640 for GSE54460 (Figure 3B-3D).

Figure 3 Prognostic evaluation of the O-glycosylation signature. (A) Association of the OGs risk score with biochemical recurrence across seven independent cohorts shown by univariate Cox analysis. (B-D) Dynamic ROC curves presenting the accuracy of the O-glycosylation signature for estimating biochemical recurrence-free survival at 1, 3, and 5 years in different datasets. CI, confidence interval; cT, clinical tumor; DKFZ, German Cancer Research Center; GS, Gleason score; HR, hazard ratio; OGs, O-glycosylation scores; PSA, prostate-specific antigen; pT, pathological tumor; ROC, receiver operating characteristic.

Development of a prognostic nomogram for risk assessment

We developed a prognostic nomogram to predict 1-, 3-, and 5-year BCR in prostate cancer patients (Figure 4A). Decision-curve analysis suggested that the nomogram conferred greater clinical net benefit than conventional clinicopathological variables (Figure 4B). At all three time points, calibration curves showed that the estimated probabilities were well aligned with the actual outcomes, indicating good model reliability (Figure 4C-4E). AUC comparisons further demonstrated superior predictive performance of the nomogram over individual clinical factors. At 1, 3, and 5 years, the nomogram yielded AUC values of 0.747, 0.839, and 0.863, respectively (Figure 4F-4H).

Figure 4 Risk-estimation nomogram derived from the O-glycosylation signature. (A) Nomogram built from the OGs score together with clinicopathological factors. (B) Decision-curve profiles reflecting clinical utility of the model relative to single clinical variables. (C-E) Calibration profiles at 1, 3, and 5 years for BCR. (F-H) ROC-based comparisons of model discrimination against single clinical variables. BCR, biochemical recurrence; cM, clinical metastasis; GS, Gleason score; M, metastasis; N, lymph node; pN, pathological lymph node; PSA, prostate-specific antigen; pT, pathological tumor; ROC, receiver operating characteristic; T, tumor.

Drug sensitivity prediction based on the O-glycosylation signature

Compounds were ranked by Pearson correlation according to their relationships with the O-glycosylation signature, and the leading candidates from both databases are shown in Figure 5A,5B. These compounds represent candidate agents inferred from the CTRP and GDSC pharmacogenomic datasets rather than drugs experimentally tested in the present study. In CTRP, QS-11, a small-molecule modulator related to Wnt/β-catenin signaling, and myriocin, an inhibitor of serine palmitoyltransferase involved in sphingolipid biosynthesis, were associated with increased predicted IC₅₀ in the high-OGs group. By contrast, avrainvillamide, an indole alkaloid with reported antitumor activity; NSC95397, a CDC25 phosphatase inhibitor; manumycin A, a Ras/farnesyltransferase-related inhibitor; and tigecycline, a glycylcycline antibiotic reported to inhibit mitochondrial translation, were associated with increased predicted IC₅₀ in the low-OGs group (Figure 5A). In GDSC, SB505124_1194, a transforming growth factor-β receptor inhibitor, was associated with increased predicted IC₅₀ in the high-OGs group, whereas AZD7762_1022, a checkpoint kinase inhibitor, was associated with increased predicted IC₅₀ in the low-OGs group (Figure 5B).

Figure 5 Drug sensitivity related to the O-glycosylation signature. (A) CTRP-derived candidate compounds linked to the OGs score and their estimated IC₅₀ profiles across different OGs strata. (B) GDSC-derived candidate compounds linked to the OGs score and their estimated IC₅₀ profiles across different OGs strata. The identified compounds represent computationally predicted drug-sensitivity candidates, including signaling inhibitors, metabolic pathway modulators, and antibiotics with reported anticancer-related mechanisms. CTRP, Cancer Therapeutics Response Portal; GDSC, Genomics of Drug Sensitivity in Cancer; IC₅₀, half maximal inhibitory concentration; OGs, O-glycosylation scores.

GALNTL6 is a key O-glycosylation-related regulator in PCa

To determine which O-glycosylation-related regulator should be selected for further experimental validation, we prioritized candidate genes based on their contribution to the OGs model, prognostic relevance, tumor-versus-normal expression pattern, and association with clinicopathological features. GALNTL6 abundance was increased in prostate cancer tissues relative to normal prostate tissues (Figure 6A). This tumor-associated upregulation suggested that GALNTL6 may have biological relevance in PCa and supported its selection as a candidate regulator for further investigation. Cox regression analysis further indicated that GALNTL6 was a consistently unfavorable prognostic regulator for BCR among the O-glycosylation-related regulators analyzed in this study (Figure 6B,6C), supporting its potential importance in PCa recurrence. Patients were stratified according to the optimal cutoff, and Kaplan-Meier analysis showed that those with higher GALNTL6 expression had shorter BCR-free survival in TCGA-PRAD, GSE21034, and GSE70768 cohorts (Figure 6D-6F). In the TCGA-PRAD cohort, increased GALNTL6 expression was also associated with more aggressive clinicopathological features, including poorer BCR outcome and more advanced pT, pN, and pM stages (Figure 6G). Together, these findings indicated that GALNTL6 was not only incorporated into the OGs model but also showed consistent associations with tumor expression, BCR prognosis, and aggressive clinicopathological characteristics, thereby providing the rationale for selecting GALNTL6 for subsequent functional validation. To select appropriate cell models for functional experiments, we examined GALNTL6 expression in BPH-1 cells and several prostate cancer cell lines, including DU145, PC3, 22Rv1, LNCaP, C4-2, and C4-2B. Among them, DU145 and 22Rv1 cells exhibited relatively high expression levels and were therefore chosen for subsequent analyses (Figure 7A; Figure S1A,S1B). This cell-line screening further supported the feasibility of experimentally evaluating the biological role of GALNTL6 in PCa models.

Figure 6 Clinical relevance of GALNTL6 in PCa. (A) GALNTL6 mRNA expression in normal prostate tissues vs. PCa tumors. (B,C) Univariate and multivariate Cox regression analyses identifying GALNTL6 as an independent prognostic factor for BCR. Association between GALNTL6 expression and clinical parameters in TCGA-PRAD. (E-G) Kaplan-Meier curves showing BCR-free survival according to GALNTL6 expression in the TCGA-PRAD, GSE21034, and GSE70768 cohorts. *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. BCR, biochemical recurrence; CI, confidence interval; cM, clinical metastasis; HR, hazard ratio; M, metastasis; N, lymph node; ns, not significant; PCa, prostate cancer; pN, pathological lymph node; PSA, prostate-specific antigen; pT, pathological tumor; TCGA-PRAD, The Cancer Genome Atlas Prostate Adenocarcinoma.

Figure 7 Expression pattern of GALNTL6 and functional validation in PCa cell. (A) GALNTL6 mRNA expression was assessed by qRT-PCR in the indicated human prostate cancer cell lines and BPH-1. (B,C) Knockdown efficiency of three small interfering RNAs targeting GALNTL6 was validated by qRT-PCR and Western blotting in DU145 and 22Rv1 cells, with each GALNTL6 knockdown group compared with the corresponding si-NC group throughout the functional experiments. (D,E) Colony formation and CCK-8 assays. (F,G) Transwell invasion assays (crystal violet staining; magnification, ×100). (H,I) Wound-healing assays (magnification, ×100). Data are presented as mean ± standard deviation from three independent biological replicates. *, P<0.05; **, P<0.01; ***, P<0.001 versus si-NC. CCK-8, Cell Counting Kit-8; OD, optical density; PCa, prostate cancer; qRT-PCR, quantitative real-time polymerase chain reaction; si-NC, non-targeting negative control small interfering RNA.

DU145 and 22Rv1 cells with GALNTL6 knockdown were established for functional analysis and confirmed by qRT-PCR and Western blotting (Figure 7B,7C; Figure S1C,S1D). Knockdown of GALNTL6 significantly suppressed colony formation (Figure 7D) and cell proliferation (Figure 7E) in both cell lines (all P<0.01). In addition, reduced GALNTL6 expression markedly impaired invasive and migratory capacities, as demonstrated by Transwell invasion assays (Figure 7F,7G) and wound healing assays (Figure 7H,7I) (all P<0.05).

Overexpression of GALNTL6 enhances PCa functions

To functionally validate the oncogenic role of GALNTL6, we established stable OE-GALNTL6 DU145 and 22Rv1 cell lines. Western blot analysis confirmed the robust upregulation of GALNTL6 protein in both engineered cell lines compared to their respective control cells (Figure 8A). The OE-GALNTL6 cells formed a markedly greater number of colonies in clonogenic assays (Figure 8B) and exhibited enhanced cell viability in CCK-8 assays (Figure 8C). After GALNTL6 overexpression, Transwell assays showed increased migration and invasion of PCa cells (Figure 8D,8E). This pro-migratory effect was corroborated by wound healing experiments, which showed an accelerated gap closure rate in both OE-GALNTL6 cell lines (Figure 8F,8G). Collectively, these gain-of-function experiments establish that GALNTL6 acts as a functional promoter of prostate cancer cell aggressiveness in vitro.

Figure 8 Overexpression of GALNTL6 promotes malignant phenotypes in PCa cells. (A) Western blot analysis confirming GALNTL6 protein overexpression in stable DU145 and 22Rv1 cells. The three lanes for each cell line represent three independent biological replicates, and the corresponding densitometric quantification is shown in Figure S1E,S1F. (B,C) Colony formation and CCK-8 assays. (D,E) Transwell invasion assays (crystal violet staining; magnification, ×100). (F,G) Wound-healing assays (magnification, ×100). All experiments were performed using three independent biological replicates. Data are presented as mean ± standard deviation. **, P<0.01 versus EC. CCK-8, Cell Counting Kit-8; EC, empty vector control; OE, overexpression; PCa, prostate cancer.

Transcriptomic profiling reveals pathway alterations after GALNTL6 overexpression

To explore the molecular changes associated with GALNTL6 overexpression, RNA sequencing was performed in GALNTL6-overexpressing and control DU145 and 22Rv1 cells. Principal component analysis showed clear separation between the overexpression and control groups in both cell lines (Figure 9A). Differential expression analysis revealed broad transcriptional alterations after GALNTL6 overexpression in DU145 and 22Rv1 cells (Figure 9B,9C), with 368 overlapping differentially expressed genes between the two cell lines (Figure 9D). Gene Ontology enrichment analysis of these overlapping genes highlighted biological processes related to epithelial cell proliferation, epithelial cell migration, response to oxygen levels, cellular response to oxidative stress, and reactive oxygen species (ROS) metabolic process (Figure 9E). Kyoto Encyclopedia of Genes and Genomes pathway analysis further showed enrichment of the PI3K-AKT signaling pathway and prostate cancer-related pathways (Figure 9F).

Figure 9 Transcriptomic profiling reveals pathway alterations after GALNTL6 overexpression. (A) Principal component analysis of RNA sequencing data from GALNTL6-overexpressing and control cells. (B,C) Volcano plots of differentially expressed genes in DU145 and 22Rv1 cells. (D) Overlapping differentially expressed genes between the two cell lines. (E,F) Gene Ontology and Kyoto Encyclopedia of Genes and Genomes enrichment analyses of the overlapping genes. (G,H) Total AKT protein expression in GALNTL6-knockdown DU145 and 22Rv1 cells. (I,J) Phosphorylated AKT protein expression in GALNTL6-knockdown DU145 and 22Rv1 cells. β-actin was used as the loading control. Data are presented as mean ± standard deviation from three independent biological replicates. **, P<0.01; ***, P<0.001 versus si-NC. FDR, false discovery rate; NC, negative control; ns, not significant; PC, principal component; si-NC, non-targeting negative control small interfering RNA.

To further evaluate whether these transcriptomic findings were supported at the protein level, we performed Western blot analysis of total AKT and phosphorylated AKT in GALNTL6-silenced DU145 and 22Rv1 cells. GALNTL6 knockdown did not significantly change total AKT protein expression in either cell line (Figure 9G,9H). In contrast, phosphorylated AKT protein expression was significantly reduced after GALNTL6 knockdown (Figure 9I,9J). These findings provide protein-level evidence supporting an association between GALNTL6 and AKT pathway activity, consistent with the transcriptomic enrichment results. However, the direct mechanism by which GALNTL6 may regulate AKT phosphorylation remains to be determined.

Discussion

The malignant progression of PCa is closely associated with profound metabolic reprogramming (21-23), with aerobic glycolysis being a particularly critical hallmark change (24-27). Abnormal glucose metabolism in cancer cells serves a dual oncogenic role: it rapidly generates energy and, more importantly, serves as the source of nucleotide sugar donors for dynamic protein glycosylation (28-31). This establishes a critical metabolic-glycosylation nexus, where rewired glucose flux directly perturbs the cellular pools of these key substrates (32). Together, these findings suggest a close link between metabolic rewiring, altered glycosylation, and tumor progression.

Accumulating evidence underscores that specific O-glycosylation modifications play precise regulatory roles in the malignant progression of PCa (33). Furthermore, aberrant overexpression of specific GALNT family members is a potent driver of tumor progression across various cancer types (34-36). However, the prognostic relevance of O-glycosylation-related regulators in BCR and aggressive PCa remains insufficiently defined.

In this study, we integrated ten machine learning algorithms to develop a consensus O‑glycosylation‑based signature termed OGs. To validate its prognostic value, we systematically evaluated the OGs model across seven independent patient cohorts. Patients in the high-OGs group showed worse BCR-free survival, and Cox analyses supported OGs as a recurrence-associated risk score in most datasets. To enhance clinical utility, we incorporated OGs with conventional clinicopathological features to construct a nomogram, enabling individualized prognostic assessment. Drug sensitivity analysis further suggested potential therapeutic differences between OGs-defined subgroups, although these predictions require experimental validation.

Within the O-glycosylation signature, GALNTL6 was prioritized for functional validation because it was retained in the final model, showed elevated expression in PCa tissues, and was associated with poorer BCR-free survival and aggressive clinicopathological features. Recent evidence further supports the prognostic relevance of glycosyltransferase-related signatures in urological malignancies. Li et al. developed and validated an 11-gene glycosyltransferase-related prognostic model for bladder urothelial carcinoma using TCGA and GEO datasets, showing that glycosyltransferase-related gene signatures could stratify patient survival and were associated with tumor immune microenvironment features (37). Although their study focused on bladder urothelial carcinoma rather than prostate cancer, it provides relevant context for the broader prognostic value of dysregulated glycosylation-related gene networks in urological cancers. Together with our findings, these data suggest that glycosylation-related transcriptional programs may represent a clinically informative molecular layer across urological malignancies.

Functionally, GALNTL6 overexpression promoted proliferation, colony formation, migration, and invasion in DU145 and 22Rv1 cells, whereas GALNTL6 knockdown showed suppressive effects. Subsequent transcriptomic sequencing and functional enrichment analyses precisely directed our focus to the PI3K-AKT pathway. As a core regulatory hub governing cellular survival, proliferation, metabolism, and motility, the PI3K-AKT pathway’s abnormal activation is a hallmark of various solid tumors including PCa (38-41). Additional Western blot analysis showed that GALNTL6 modulation altered AKT phosphorylation, as indicated by changes in the phosphorylated AKT/total AKT ratio, supporting an association between GALNTL6 and AKT pathway activity. However, these data do not establish GALNTL6 as an upstream switch of the PI3K-AKT pathway or prove direct O-glycosylation of pathway components. Future glycoproteomic, co-immunoprecipitation, inhibition/rescue, and substrate-validation studies are required to test this mechanism.

In PCa, ROS can play both tumor-promoting and tumor-suppressive roles, depending on the biological context (42-46). At low levels, they can promote proliferation, but the chronically elevated ROS generated by hyperactive cancer metabolism also poses a serious threat to cell survival. To cope with this stress, cancer cells must activate adaptive pathways that restore redox balance (45,47,48). Our findings are consistent with recent research demonstrating that GALNT14 promotes lung adenocarcinoma progression by reducing intracellular ROS levels through its O-glycosylation activity (49). Meanwhile, PI3K/AKT signaling has been widely implicated in oxidative stress regulation and supports cell survival under conditions of metabolic stress (50-53). In this study, transcriptomic enrichment suggested a possible association between GALNTL6 overexpression and oxidative-stress-related transcriptional programs. However, we did not directly assess ROS levels, antioxidant enzyme activity, redox balance, or responses to oxidative-damage-based treatments. Therefore, whether GALNTL6 enhances antioxidant capacity or contributes to therapy resistance remains to be determined in future studies.

In summary, this study developed an O-glycosylation-based prognostic signature for BCR risk stratification in PCa and identified GALNTL6 as a candidate regulator associated with malignant phenotypes. The OGs-based nomogram showed favorable performance in the current retrospective datasets, although prospective validation is still required. This study also has several limitations, including its retrospective design, incomplete tissue-level protein validation, and the need for further mechanistic studies to define direct O-glycosylation substrates and oxidative-stress-related functional effects.

Conclusions

This study establishes an O-glycosylation-based signature for predicting BCR in PCa. We identified GALNTL6 as a candidate regulator within this signature, and functional assays indicated that GALNTL6 promotes malignant phenotypes in PCa cells. Transcriptomic enrichment and AKT phosphorylatio29 analysis suggest a possible association between GALNTL6 and AKT pathway activity, whereas its relationship with oxidative-stress-related programs requires further validation. These findings support the potential value of O-glycosylation-related molecular programs in PCa risk stratification and provide a basis for future mechanistic studies.

Acknowledgments

The authors would like to thank the contributors of the public databases used in this study for making their valuable datasets openly accessible.

Footnote

Reporting Checklist: The authors have completed the MDAR and TRIPOD reporting checklists. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0315/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0315/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0315/prf

Funding: This work was supported by the Science and Technology Development Fund (FDCT) of the Macau Special Administrative Region, China (grant Nos. 0116/2023/RIA2 and 006/2023/SKL, awarded to W.Z.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0315/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. All human sequencing data were obtained from publicly accessible, de-identified databases; therefore, informed consent was not required.

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. Siegel RL, Kratzer TB, Giaquinto AN, et al. Cancer statistics, 2025. CA Cancer J Clin 2025;75:10-45. [Crossref] [PubMed]
2. Sharifi N, Gulley JL, Dahut WL. Androgen deprivation therapy for prostate cancer. JAMA 2005;294:238-44. [Crossref] [PubMed]
3. Zeng C. Advances in cancer treatment: the role of new technologies and research. Cell Investigation 2025;1:100001.
4. Falagario UG, Abbadi A, Remmers S, et al. Biochemical Recurrence and Risk of Mortality Following Radiotherapy or Radical Prostatectomy. JAMA Netw Open 2023;6:e2332900. [Crossref] [PubMed]
5. Van den Broeck T, van den Bergh RCN, Briers E, et al. Biochemical Recurrence in Prostate Cancer: The European Association of Urology Prostate Cancer Guidelines Panel Recommendations. Eur Urol Focus 2020;6:231-4. [Crossref] [PubMed]
6. Trock BJ, Walsh PC. Impact of postoperative prostate-specific antigen disease recurrence and the use of salvage therapy on the risk of death. Cancer 2011;117:656-author reply 656. [Crossref] [PubMed]
7. Rodrigues DN, Butler LM, Estelles DL, et al. Molecular pathology and prostate cancer therapeutics: from biology to bedside. J Pathol 2014;232:178-84. [Crossref] [PubMed]
8. Deribe YL, Pawson T, Dikic I. Post-translational modifications in signal integration. Nat Struct Mol Biol 2010;17:666-72. [Crossref] [PubMed]
9. Beltrao P, Bork P, Krogan NJ, et al. Evolution and functional cross-talk of protein post-translational modifications. Mol Syst Biol 2013;9:714. [Crossref] [PubMed]
10. Keenan EK, Zachman DK, Hirschey MD. Discovering the landscape of protein modifications. Mol Cell 2021;81:1868-78. [Crossref] [PubMed]
11. Chen Q, Jiang K, Bronze MS, et al. Tumor microenvironment and macroenvironment: A new perspective on holistic oncology. Cancer Lett 2025;634:218076. [Crossref] [PubMed]
12. Bard F, Chia J. Cracking the Glycome Encoder: Signaling, Trafficking, and Glycosylation. Trends Cell Biol 2016;26:379-88. [Crossref] [PubMed]
13. Sanz-Martinez I, Pereira S, Merino P, et al. Molecular Recognition of GalNAc in Mucin-Type O-Glycosylation. Acc Chem Res 2023;56:548-60. [Crossref] [PubMed]
14. Magalhães A, Duarte HO, Reis CA. The role of O-glycosylation in human disease. Mol Aspects Med 2021;79:100964. [Crossref] [PubMed]
15. Gomes J, Mereiter S, Magalhães A, et al. Early GalNAc O-Glycosylation: Pushing the Tumor Boundaries. Cancer Cell 2017;32:544-5. [Crossref] [PubMed]
16. Wagner KW, Punnoose EA, Januario T, et al. Death-receptor O-glycosylation controls tumor-cell sensitivity to the proapoptotic ligand Apo2L/TRAIL. Nat Med 2007;13:1070-7. [Crossref] [PubMed]
17. Nguyen AT, Chia J, Ros M, et al. Organelle Specific O-Glycosylation Drives MMP14 Activation, Tumor Growth, and Metastasis. Cancer Cell 2017;32:639-653.e6. [Crossref] [PubMed]
18. Arai J, Hayakawa Y, Tateno H, et al. Impaired Glycosylation of Gastric Mucins Drives Gastric Tumorigenesis and Serves as a Novel Therapeutic Target. Gastroenterology 2024;167:505-521.e19. [Crossref] [PubMed]
19. Ferreira JA, Magalhães A, Gomes J, et al. Protein glycosylation in gastric and colorectal cancers: Toward cancer detection and targeted therapeutics. Cancer Lett 2017;387:32-45. [Crossref] [PubMed]
20. Zhou F, Ma J, Zhu Y, et al. The role and potential mechanism of O-Glycosylation in gastrointestinal tumors. Pharmacol Res 2022;184:106420. [Crossref] [PubMed]
21. Zadra G, Loda M. Metabolic Vulnerabilities of Prostate Cancer: Diagnostic and Therapeutic Opportunities. Cold Spring Harb Perspect Med 2018;8:a030569. [Crossref] [PubMed]
22. Cardoso HJ, Carvalho TMA, Fonseca LRS, et al. Revisiting prostate cancer metabolism: From metabolites to disease and therapy. Med Res Rev 2021;41:1499-538. [Crossref] [PubMed]
23. Zhou S, Liu Y, Zhang N, et al. Glycolytic enzyme PFKFB4 governs lipolysis by promoting de novo lipogenesis to drive the progression of hepatocellular carcinoma. Cancer Lett 2025;626:217774. [Crossref] [PubMed]
24. Fox JJ, Gavane SC, Blanc-Autran E, et al. Positron Emission Tomography/Computed Tomography-Based Assessments of Androgen Receptor Expression and Glycolytic Activity as a Prognostic Biomarker for Metastatic Castration-Resistant Prostate Cancer. JAMA Oncol 2018;4:217-24. [Crossref] [PubMed]
25. Yang X, Li C, Cheng Y, et al. YY1-glycolytic axis promotes high glucose-induced cancer stemness in endometrial cancer: A multi-omics guided therapeutic strategy. Cancer Lett 2025;632:217936. [Crossref] [PubMed]
26. Zhao Q, Chen WL, Melino G, et al. Targeting fructose metabolism for cancer therapy. Cancer Lett 2025;631:217914. [Crossref] [PubMed]
27. Wu C, Xu T, Zhang H, et al. Hypoxia and immunometabolism in the tumor microenvironment: insights into mechanisms and therapeutic potential. Cancer Lett 2025;631:217913. [Crossref] [PubMed]
28. Su Y, Luo Y, Zhang P, et al. Glucose-induced CRL4(COP1)-p53 axis amplifies glycometabolism to drive tumorigenesis. Mol Cell 2023;83:2316-2331.e7. [Crossref] [PubMed]
29. Ayodeji SA, Bao B, Teslow EA, et al. Hyperglycemia and O-GlcNAc transferase activity drive a cancer stem cell pathway in triple-negative breast cancer. Cancer Cell Int 2023;23:102. [Crossref] [PubMed]
30. Chen S, Huang Y, Zhang S, et al. Ribonucleotide reductases in cancer: mechanisms, clinical progress, and prospects. Cell Investigation 2025;1:100034.
31. Wang SC, Shen YC, Pan MR, et al. Deglycosylated PD-L1 is a biomarker for immune checkpoint blockade response: a real-world study in breast cancer patients in Taiwan. Cancer Lett 2025;631:217912. [Crossref] [PubMed]
32. Ji Y, Xu Z, Tang L, et al. O-GlcNAcylation of YBX1 drives a glycolysis-histone lactylation feedback loop in hepatocellular carcinoma. Cancer Lett 2025;631:217957. [Crossref] [PubMed]
33. Scott E, Hodgson K, Calle B, et al. Upregulation of GALNT7 in prostate cancer modifies O-glycosylation and promotes tumour growth. Oncogene 2023;42:926-37. [Crossref] [PubMed]
34. Song KH, Park MS, Nandu TS, et al. GALNT14 promotes lung-specific breast cancer metastasis by modulating self-renewal and interaction with the lung microenvironment. Nat Commun 2016;7:13796. [Crossref] [PubMed]
35. Hu Q, Tian T, Leng Y, et al. The O-glycosylating enzyme GALNT2 acts as an oncogenic driver in non-small cell lung cancer. Cell Mol Biol Lett 2022;27:71. [Crossref] [PubMed]
36. Park MS, Yang AY, Lee JE, et al. GALNT3 suppresses lung cancer by inhibiting myeloid-derived suppressor cell infiltration and angiogenesis in a TNFR and c-MET pathway-dependent manner. Cancer Lett 2021;521:294-307. [Crossref] [PubMed]
37. Li W, Zuo K, Zhao Q, et al. An 11-gene glycosyltransferases-related model for the prognosis of patients with bladder urothelial carcinoma: development and validation based on TCGA and GEO datasets. Transl Androl Urol 2024;13:2771-86. [Crossref] [PubMed]
38. Tang Y, Pan J, Huang S, et al. Downregulation of miR-133a-3p promotes prostate cancer bone metastasis via activating PI3K/AKT signaling. J Exp Clin Cancer Res 2018;37:160. [Crossref] [PubMed]
39. Wang Q, Wu S, Gu Y, et al. RASAL2 regulates the cell cycle and cyclin D1 expression through PI3K/AKT signalling in prostate tumorigenesis. Cell Death Discov 2022;8:275. [Crossref] [PubMed]
40. Chen C, Cai Q, He W, et al. AP4 modulated by the PI3K/AKT pathway promotes prostate cancer proliferation and metastasis of prostate cancer via upregulating L-plastin. Cell Death Dis 2017;8:e3060. [Crossref] [PubMed]
41. Meng F, Li H, Wang Y, et al. The mechanism of AKT activation in cancer. Cell Investigation 2025;1:100046.
42. Cheung EC, Vousden KH. The role of ROS in tumour development and progression. Nat Rev Cancer 2022;22:280-97. [Crossref] [PubMed]
43. Kuo CL, Ponneri Babuharisankar A, Lin YC, et al. Mitochondrial oxidative stress in the tumor microenvironment and cancer immunoescape: foe or friend? J Biomed Sci 2022;29:74. [Crossref] [PubMed]
44. Glorieux C, Liu S, Trachootham D, et al. Targeting ROS in cancer: rationale and strategies. Nat Rev Drug Discov 2024;23:583-606. [Crossref] [PubMed]
45. Liang X, Weng J, You Z, et al. Oxidative stress in cancer: from tumor and microenvironment remodeling to therapeutic frontiers. Mol Cancer 2025;24:219. [Crossref] [PubMed]
46. Prasad S, Gupta SC, Tyagi AK. Reactive oxygen species (ROS) and cancer: Role of antioxidative nutraceuticals. Cancer Lett 2017;387:95-105. [Crossref] [PubMed]
47. Tasdogan A, Ubellacker JM, Morrison SJ. Redox Regulation in Cancer Cells during Metastasis. Cancer Discov 2021;11:2682-92. [Crossref] [PubMed]
48. Di Stefano V, Cencioni C, Zaccagnini G, et al. p66ShcA modulates oxidative stress and survival of endothelial progenitor cells in response to high glucose. Cardiovasc Res 2009;82:421-9. [Crossref] [PubMed]
49. Tang B, Wang K, Ren Q, et al. GALNT14-mediated O-glycosylation drives lung adenocarcinoma progression by reducing endogenous reactive oxygen species generation. Cell Signal 2024;124:111477. [Crossref] [PubMed]
50. Wang J, Zhang J, Yu ZL, et al. The roles of dietary polyphenols at crosstalk between type 2 diabetes and Alzheimer's disease in ameliorating oxidative stress and mitochondrial dysfunction via PI3K/Akt signaling pathways. Ageing Res Rev 2024;99:102416. [Crossref] [PubMed]
51. Fu H, Sen L, Zhang F, et al. Mesenchymal stem cells-derived extracellular vesicles protect against oxidative stress-induced xenogeneic biological root injury via adaptive regulation of the PI3K/Akt/NRF2 pathway. J Nanobiotechnology 2023;21:466. [Crossref] [PubMed]
52. Xu G, Zhang J, Shao C, et al. Puerarin alleviates cerebral ischemia/reperfusion (CIR)-induced neurocyte oxidative stress and apoptosis via DNA demethylation-mediated PI3K/Akt activation. Phytomedicine 2025;145:157094. [Crossref] [PubMed]
53. Liu Y, Gong S, Li K, et al. Coptisine protects against hyperuricemic nephropathy through alleviating inflammation, oxidative stress and mitochondrial apoptosis via PI3K/Akt signaling pathway. Biomed Pharmacother 2022;156:113941. [Crossref] [PubMed]