Integrative analysis identifies a glycosylation-related lncRNA signature associated with prognosis in kidney renal clear cell carcinoma

Introduction

Kidney renal clear cell carcinoma (KIRC), an adenocarcinoma originating from renal tubular epithelium, accounts for 70–80% of renal cell carcinomas (1). The global annual incidence rate of KIRC continues to rise, with high mortality rate in the urinary system tumors (2). Some early-stage patients can survive for a long time after surgery, but about 20–30% of patients will experience recurrence and metastasis. Late-staged patients are insensitive to treatments like radiotherapy and chemotherapy (3). Despite being a highly lethal tumor, the prognosis evaluation of KIRC has long relied on traditional staging systems, including tumor-node-metastasis (TNM) staging and Fuhrman grading which cannot fully reflect the biological characteristics of tumors (such as molecular heterogeneity) or predict treatment response (4). Therefore, it is essential to seek other evaluation methods.

Biomarkers, measurable indicators of a biological state or condition, can be molecules, genes, gene products, cells, or physiological characteristics that provide information about health, disease, or responses to treatment (5). With their ability to different-stage diagnosis and prognosis, scientists and clinical doctors try to establish disease-related molecular models for diagnosis, prognosis, and treatment sensitivity assessment of KIRC. These biomarkers mainly consist of the molecules or genes that are related to metabolism of nicotinamide (6), migrasome (7) and topoisomerase II alpha (8). However, current diagnostic and prognostic signature for KIRC remains limited, and thus potential prognostic prediction models for KIRC are urgently needed to be developed.

Glycosylation is a critical post-translational modification, including sialylation, fucosylation, and N- and O-linked glycan (9). Aberrant glycosylation is a hallmark of cancer, affecting cell signaling, adhesion, immune evasion, and metastasis. Increased sialylation is linked to tumor metastasis, as it leads to a decrease in intercellular adhesion and an increase in tumor cell mobility (10). Siglec-Sialoglycan facilitates immune evasion by suppressing the activity of natural killer (NK) cells (11). Upregulated fucosylation boosts tumor cell proliferation via epidermal growth factor receptor (EGFR) signaling (12). N-linked glycan enhances the angiogenesis, and tumor cell transformation, survival and metastasis (13). Also, O-glycan truncation is strongly associated with tumor growth, migration and poor prognosis (14). In short, aberrant glycosylation can provide a favourable environment for tumor growth, immune escape and the transmission of tumor proliferation signals (15). In KIRC, the conserved oligomeric Golgi (COG) controls protein glycosylation, and all members of the COG complex subunit are strongly associated with the prognosis of KIRC patients (16).

Long non-coding RNAs (lncRNAs), a molecule with more than 200 nucleotides, possesses similar structure with messenger RNA (17). Although they cannot be translated into protein, lncRNAs are involved in cell growth, differentiation and others (18). Recent studies also have demonstrated that lncRNAs are critical for tumorigenesis, progression and prognosis. For example, lncRNA-ENTPD3-AS1 suppresses the growth of renal cell carcinoma by targeting miR-155/HIF-1 signaling (19). LncRNA-MALAT1 inhibits the expression of miR-194-5p to facilitate the development of KIRC (20). In addition, lncRNAs can act on the glycosylation process to influence the development of cancers. lncRNA-SNHG7 supports the growth and metastasis of colorectal cancer by regulating the expression of acetylgalactose-aminotransferase GALNT7 (21). LncRNA-ST3Gal6-AS1 inhibits colorectal cancer cell proliferation, metastasis, and promoted cell apoptosis though activating ST3Gal6-mediateα-2, 3 sialylation (22). Particularly, glycosylation-related lncRNAs (GRLncs) play an important role in tumor prognosis. The prediction model based on glycosyltransferase-related lncRNAs (GT-lncRNAs) could independently predict the prognosis of the patients with endometrial cancer (23). However, the effect of GRLncs on the prognosis of KIRC has not been elucidated.

In this study, we developed a prognostic model for KIRC based on GRLncs by univariate Cox regression analysis, least absolute shrinkage and selection operator (LASSO) regression and multivariate Cox regression. The prediction performance and robustness of the model were evaluated using Kaplan-Meier (KM) survival analysis, time-dependent receiver operating characteristic (ROC) analysis, Cox regression analyses, principal component analysis (PCA), nomogram construction, and internal validation based on randomly divided training and testing cohorts. To further support the biological relevance of the signature, we performed functional validation of three constituent lncRNAs (DLGAP1-AS2, EPB41L4A-DT, and AC084876.1) in KIRC cell lines and validated their expression patterns in clinical KIRC tissues by Real-time quantitative polymerase chain reaction (RT-qPCR). Collectively, this study provides a novel glycosylation-related prognostic framework that may complement existing risk stratification strategies and facilitate future mechanistic studies and clinical translation in KIRC. We present this article in accordance with the TRIPOD and MDAR reporting checklists (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0289/rc).

Methods

Data source

RNA-seq data [normalized by fragments per kilobase of transcript per million mapped reads (FPKM)] and clinical data of KIRC patients (including survival time, survival status, age, gender, tumor grade, tumor stage and TNM stage) were both downloaded from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/). The RNA-seq dataset comprised 611 samples, including 539 tumor tissues and 72 adjacent non-tumor tissues. Among 539 tumor samples, 537 had available clinical information. After matching transcriptomic and clinical datasets, 530 patients had both expression and clinical data available. Subsequently, 23 patients with overall survival (OS) <30 days were excluded, resulting in a final cohort of 507 patients. The patient selection workflow is illustrated in Figure 1. No imputation was performed for missing clinical variables; therefore, only patients with complete data were retained (complete-case analysis). Baseline characteristics including age, sex, tumor grade, stage, and TNM classification were compared between included and excluded patients using appropriate statistical tests (Table S1), which showed no significant differences between the two groups (all P>0.05). Glycosylation-related genes (GRGs) were downloaded from GeneCards database (https://www.Genecards.org/). Consistent with previous studies (24), genes with a GeneCards relevance score >7 were defined as GRGs and included in subsequent analyses, yielding a total of 136 GRGs.

Figure 1 The workflow of this study. Experimental validation was performed for three of the five identified GRLncs (DLGAP1-AS2, EPB41L4A-DT, AC084876.1) due to technical constraints; AC093278.2 and AC005261.3 were not tested functionally. CCK8, Cell Counting Kit 8; Cor, correlation coefficient; DEGs, differentially expressed genes; EDU, 5-ethynyl-2'-deoxyuridine; FDR, false discovery rate; GO, Gene Ontology; GRLncs, glycosylation-related lncRNAs; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; KIRC, kidney renal clear cell carcinoma; KM, Kaplan-Meier; PCA, principal component analysis; ROC, receiver operating characteristic; RT-qPCR, real-time quantitative polymerase chain reaction; TCGA, The Cancer Genome Atlas.

Identification of differentially expressed GRGs and functional enrichment analysis

Differential expression analysis between KIRC tumor tissues and adjacent normal tissues was performed using the “limma” R package. Glycosylation-related differentially expressed genes (DEGs) were screened out with a false discovery rate (FDR) <0.05 and |log₂ fold change (FC) >1| as criteria (25). P values were adjusted for multiple testing using the Benjamini-Hochberg (BH) method to control the FDR. To explore the biological functions of the differentially expressed GRGs, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis were conducted using the “ClusterProfiler” package. Terms with adjusted P values <0.05 were considered significantly enriched.

Construction of the prognostic model based on GRLncs

Pearson correlation analysis was performed using the “limma” R package to determine the correlation between GRGs and GRLncs. LncRNAs with an absolute correlation coefficient (|R|) >0.3 and P<0.001 with at least one GRG were considered GRLncs (26). A total of 968 GRLncs were identified. Next, the extracted clinical information from 507 KIRC patients and the 968 GRLncs were integrated and analyzed by univariate Cox regression analysis to obtain the GRLncs associated with KIRC prognosis based on the criterion of P<0.001. Finally, LASSO regression (maxit =1000, nfolds =10, standardize = TRUE, lambda =1) and multivariate Cox regression (stepwise regression) were further used to screen the GRLncs for constructing a prognostic model for KIRC.

The validation of predictive performance of the prognostic model

The validation of predictive performance of the prognostic model was performed by both performance assessment methods (such as KM analysis, ROC, Nomogram) and internal validation.

During the performance assessment process, the risk score, univariate and multivariate independent prognostic analysis and PCA were used for assessing the predictive performance of the prognostic model. Here the risk score of each KIRC patient was calculated by the following formula:

Riskscore=0.76898×Exp(DLGAP1-AS2)−0.5709×Exp(EPB41LA4-DT)−0.4844×Exp(AC093278.2)+0.42592×Exp(AC084876.1)+0.32022×Exp(AC005261.3)[1]

where Exp(lncRNA) denotes the normalized expression level (FPKM) of the corresponding lncRNA.

At first, the above 507 KIRC patients that were used for screening prognostic GRLncs were divided into high-risk group and low-risk group based on median risk score. Then KM survival analysis was used to show the OS of these two groups. The accuracy of the prognostic model was evaluated by ROC curves, which were built using “survivalROC” R package. The univariate and multivariate Cox regression analysis was used for assessing which factors were associated with prognosis. PCA was performed with “scatterplot3d” R package to obtain the expression patterns of optimal GRLncs in the low-risk group and high-risk group. By combining risk score with clinicopathological variables, such as age, gender, stage and M-stage, the nomograms (bootstrapping with 1,000 resampling as an internal validation) were constructed to predict 1-, 3- and 5-year survival rate of KIRC patients. Additionally, the calibration curves tested consistency between the actual OS rates and the predicted survival rates at 1, 3 and 5 years.

Internal validation was applied to further examine the predictive ability of the above model. Although the clinical data used in internal validation was the same as the above performance assessment methods, but the 507 KIRC patients were randomly divided into the train set (n=255) and the test set (n=252) by stratified random sampling and multiple iterations (1,000 times). Stratified random sampling was performed using the “create Data Assignment” function in the caret package. 1000 times of iterations could reduce the risk of overfitting. The clinical characteristics of the 507 KIRC patients were shown in Table S2. For these two sets, we conducted KM survival analysis and ROC curves separately to prove the prediction robustness and reliability of the prognostic model.

Gene set enrichment analysis (GSEA) and tumor microenvironment (TME) analysis

The GSEA was performed to investigate the enrichment of glycosylation-related pathways in KIRC tumor tissues and the enriched pathways in two risk groups. At first, the genes that were not expressed in more than half of the KIRC samples were excluded from the expression matrix (27). The filtered expression data underwent log2 (FPKM + 1) transformation. Then GSEA was carried out with GSEA 4.1.0 (http://www.broad.mit.edu/gsea/) based on the filtered expression matrix of genes and the corresponding phenotypes. The expression matrices were formatted as .gct files, and the corresponding phenotypics were formatted as .cls files. GSEA was applied to the C2 canonical pathway gene sets (MSigDB v7.4). Statistical significance of enrichment scores was computed using 1000 phenotype permutations. |Normalized enrichment score (NES)| >1, P<0.05 and FDR <0.25 were considered as the thresholds for statistical significance.

The ESTIMATE algorithm was applied to evaluate TME characteristics. The infiltration abundance of 17 immune cell types was quantified via ssGSEA (GSVA package), and the expression levels of 35 immune checkpoint genes were compared between the high- and low-risk groups. Based on tumor immune dysfunction and exclusion (TIDE) related gene sets [cytotoxic T lymphocyte (CTL), dysfunction, exclusion, and myeloid-derived suppressor cell (MDSC) signatures], the mean expression of each signature was calculated, and the composite TIDE score was defined as dysfunction + exclusion − CTL, with patients stratified into immunotherapy responder and non-responder groups by the median score. Drug sensitivity was predicted using ridge regression (glmnet package, alpha =0) with 10-fold cross-validation based on the GDSC database, and the estimated IC50 values were compared between the two groups. P<0.05 was considered statistically significant.

Single cell RNA-seq analysis

The single cell RNA-seq dataset of KIRC (accession No. GSE171306) was from Tumor Immune Single Cell Hub 2 (TISCH2) database (http://tisch.comp-genomics.org/). For this dataset, LogNormalize normalization was performed for data scaling. Graph-based Louvain clustering was employed using 20 significant principal components (PCs) for graph construction, with the clustering resolution parameter set at 0.5. Based on major-lineage level annotations and malignant cell identity, all cells were classified into three types: malignant cells, immune cells, and stromal cells. Due to the gene visualization functionality provided by TISCH2, users are allowed to search for relevant genes in the selected datasets. The violin plot was displayed to show the expression distribution of the GRLncs from the prognostic model in malignant cells, stromal cells and immune cells.

Additionally, we investigated the GRGs in malignant cells, stromal cells and immune cells of KIRC. At first, we downloaded the cluster-averaged expression matrix of KIRC from TISCH2 database, in which the “Celltype_malignancy” expression matrix consisted of the gene expression of malignant cells, stromal cells and immune cells. The expressed genes in malignant cells, stromal cells and immune cells were then separately intersected with the 136 GRGs mentioned above to obtain the GRGs in these three cell types. Then the intersection of the GRGs in these three cell types and the 18 glycosylation-related DEGs from the whole tissues (detected by bulk RNA-seq) was extracted using a Venn diagram (https://www.bioinformatics.com.cn/).

Clinical samples and RT-qPCR validation

To further validate the expression patterns of prognostic GRLncs in clinical specimens, 10 paired KIRC tumor tissues and adjacent normal tissues were obtained from the Biobank of West China Hospital, Sichuan University. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Biomedical Ethics Committee of West China Hospital, Sichuan University (Approval No. 2025-986). Written informed consent was obtained from all participants.

Total RNA was extracted from tissue samples using TRIzol reagent (Invitrogen, USA) according to the manufacturer’s instructions. RNA was reverse-transcribed into cDNA using a reverse transcription kit, and RT-qPCR was performed using SYBR Green Master Mix on a QuantStudio Real-Time PCR System (Applied Biosystems, USA). U6 was used as the internal control. Relative expression levels were calculated using the 2^−ΔΔCt method. Primer sequences are listed in Table S3.

In vitro experimental validation of the prognostic model

Cell culture

The human KIRC cell lines Caki-1 and 786-O were provided by Servicebio, China. Caki-1 cells were cultured with Dulbecco’s Modified Eagle Medium (DMEM) (Sigma-Aldrich, USA) containing 10% fetal bovine serum (FBS; Hyclone, USA) and 1% penicillin-streptomycin (Gibco, USA). 786-O cells were kept in Roswell Park Memorial Institute (RPMI)-1640 medium (Hyclone, USA) containing 10% FBS and 1% penicillin-streptomycin. Both cell lines were kept at 37 ℃ in a 5% CO₂ incubator. The medium was replaced once every 3 days.

Antisense oligonucleotides (ASO) transfection

According to the instructions (28), ASO transfection was performed. In brief, when the cell density reached 30–50%, 50nM DLGAP1-AS2, EPB41L4A-DT, AC084876.1 and non-targeting ASO [negative control (NC)] (synthesized by RiboBio, China) were transfected into cells with Lipo2000 (Invitrogen, USA). After 48 hours (h) of transfection, the cells were collected for subsequent experiments. The corresponding nucleotide sequences of these ASOs are shown in Table S4.

Total RNA extraction and RT-qPCR

The total RNA was extracted via Trizol (Life technologies, USA) method and then reversely transcribed into cDNA with RevertAid First Strand cDNA Synthesis Kit (ThermoFisher, USA). The obtained cDNA was used as a template for qPCR. The reaction system of qPCR was as follows: (2×) mix (10 µL), (10 µM) forward primer (1 µL), (10 µM) reverse primer (1 µL), and cDNA (50 ng), and DEPC water (to 20 µL). The reaction procedures were as follows: 95 ℃ for 30 seconds, 95 ℃ for 5 seconds, 60 ℃ for 30 seconds, and 40 cycles. Using U6 as an internal reference (29), the relative expression of lncRNAs was quantified through 2^−ΔΔCT method. All the primers used in this study were designed by IDT PrimerQuest online tool (https://sg.idtdna.com/pages/tools/primerquest), and their sequences are shown in Table S3.

5-ethynyl-2′-deoxyuridine (EdU) incorporation assay

The Caki-1 cells in the logarithmic phase were inoculated in 96-well plates (Corning, USA) and placed in a cell incubator (ThermoFisher, USA) overnight. Then the adherent cells were subjected to ASO transfection. After transfection for 48 h, the cells were harvested for EdU incorporation assay according to the instructions (RiboBio, China). The EdU positive cells were defined as proliferative cells. Three replicate wells were set up for each group, and the independent experiments repeated twice.

Cell Counting Kit 8 (CCK8)

Caki-1 and 786-O cells were seeded into 96-well plates (4×10³ cells per well). After ASO intervention for 48 h, 10 µL CCK-8 reagent (Dojindo, Japan) was added into each well and incubated for 4 h at 37 ℃. Then the optical density (OD) value at the wavelength of 450 nm was detected by microplate reader (ThermoFisher, USA). The cell viability was calculated by the following formula: cell viability = (OD value in experimental group − OD value in blank group)/(OD value in control group − OD value in blank group) ×100%. Three replicate wells were set up for each group, and the independent experiments repeated twice.

Clone formation assay

The Caki-1 and 786-O cells in the logarithmic phase were inoculated in 6-well plates (1×10³ cells per well) and placed in an incubator overnight. Then ASO was transfected into the adherent cells. After 10 days of culture, the cells were harvested and fixed with 4% paraformaldehyde (Sigma-Aldrich, USA) for 30 min. Then the cells were stained with 0.1% crystal violet (Beyotime, China) for 15 min, and washed with PBS. After natural drying, the cells were photographed and the formed clones were observed. Three replicate wells were set up for each group, and the independent experiments repeated twice.

Transwell assay

After ASO transfection for 48h, Caki-1 and 786-O cells were collected and suspended with serum-free medium. Then the cells in serum-free medium were added to the upper chambers with matrigel (2×10⁵ cells per well), and 500 µL medium containing 10% FBS was added to the lower chambers. The transwell chambers (Merck, Germany) were placed into 24-well plates and incubated in an incubator. After 48 h, the untransferred cells in upper chambers were removed, and the transferred cells were fixed with 4% paraformaldehyde for 30 min and stained with 0.1% crystal violet for 15 min. Then the transferred cells were photographed under an optical microscope (SDPTOP, China). Three replicate wells were set up for each group, and the independent experiments repeated twice.

Statistical analysis

Bioinformatics analysis was performed by R software (version R 4.1.3). Normal and non-normal distributed continuous variables were analyzed by unpaired Student’s t-test and Wilcoxon rank-sum test, respectively. Chi-squared test was used to analyze categorical data. The FDR was corrected by BH method. Univariate Cox regression analysis was used to analyze the relationship between GRLncs and OS, and LASSO regression and multivariate Cox analysis were used to screen GRLncs for constructing a prognostic model. KM survival analysis and log-rank test were used to analyze the OS of patients in the high-risk and low-risk groups.

For experimental validation, the statistical analysis was performed by IBM SPSS Statistics V19 package (Armonk, NY, USA). Student t-test was performed to analyze the data from two completely randomized design groups. P value of <0.05 was considered to be statistically significant.

Results

Glycosylation is a crucial metabolic pathway that promotes the progression of KIRC

As mentioned above, we first obtained 136 GRGs with the correlation score greater than 7 from GeneCards database (Table S5). Among these GRGs, 18 glycosylation-related DEGs were differentially expressed in KIRC (FDR <0.05 and |log₂ FC| >1), in which 11 DEGs were down-regulated and 7 DEGs were up-regulated (Figure 2A). KEGG pathway analysis showed that these glycosylation-related DEGs were mainly enriched in mucin type O-glycan biosynthesis and N-glycan biosynthesis. GO analysis demonstrated that these DEGs were located in apical plasma membrane and apical part of the cell, and possessed the molecular functions consisting of UDP-glycosyltransferase activity, acetyl glucosamine transferase activity, etc. More importantly, GO analysis also explained that these DEGs were mainly enriched in glycosylation biological processes, including protein glycosylation and macromolecular glycosylation (Figure 2B-2D). Furthermore, the enrichment of glycosylation-related pathways in KIRC was confirmed by GSEA analysis (Figure 2E,2F). These glycosylation-related DEGs enriched in glycosaminoglycan biosynthesis chondroitin and sulfate proteoglycan biosynthetic process indicated that glycosylation-related biological processes and pathways are associated with KIRC. As reported previously, aberrant glycosylation facilitates tumor growth, immune escape and malignant signal transduction (30,31). In fact, glycosylation-related DEGs were indeed expressed in malignant cells, immune cells and stromal cells from KIRC tumor tissues (Figure 2G), further proving their importance in the progression of KIRC. Therefore, these findings support the potential involvement of glycosylation-related pathways in KIRC biology and provide a rationale for investigating their functional roles in future studies.

Figure 2 GO and KEGG analysis of the 18 differentially expressed GRGs in KIRC. (A) Volcano map showed the 136 GRGs in KIRC. Red dots represented the 7 significantly up-regulated DEGs and blue dots represented the 11 significantly down-regulated DEGs in KIRC. (B) The enriched BP, CC, MF, KEGG pathways of the 18 differentially expressed GRGs. (C) Circle graph showed the KEGG pathways and GO terms that the 18 differentially expressed GRGs were enriched in. (D) Bubble map showed the enriched KEGG pathways and GO terms of the 18 differentially expressed GRGs. The larger the bubble, the more genes are enriched. The larger the log10 (P. adjust), the higher the significance. Z-Score (a method for data standardization) indicates activation status, negative values represent inhibition, and positive values represent activation. (E) GSEA (based on KEGG analysis) showed the enrichment of glycosylation-related pathways in KIRC. (F) GSEA (based on GO analysis) showed the enrichment of glycosylation-related pathways in KIRC. (G) Venn diagram showed the intersection of the GRGs in malignant cells, immune cells, stromal cells (from single cell RNA-seq data) and the 18 differentially expressed GRGs from the whole tissues (detected by bulk RNA-seq). BP, biological process; CC, cellular component; DEGs, differentially expressed genes; FC, fold change; GO, Gene Ontology; GRGs, glycosylation-related genes; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes; KIRC, kidney renal clear cell carcinoma; MF, molecular function; NES, normalized enrichment score.

The GRLncs are successfully used for constructing the prognostic model of KIRC

A total of 968 GRLncs were identified by their correlation with the 18 glycosylation-related DEGs through Pearson correlation analysis, based on |Cor| >0.3 and P<0.001. Subsequent univariate Cox regression analysis showed that 207 GRLncs were associated with the prognosis of KIRC (P<0.001) (Table S6). Then, LASSO regression and multivariate Cox regression analyses further analyzed these 207 GRLncs. Additional single-cell RNA-seq data were employed to screen out 5 critical GRLncs (DLGAP1-AS2, EPB41L4A-DT, AC005261.3) expressed in malignant cells, immune cells and stromal cells of KIRC tumor tissues (Figure 3A-3D). Hazard ratios (HRs) were calculated which revealed AC093278.2 and EPB41L4A-DT as low-risk protective lncRNAs (HR <1), whereas the DLGAP1-AS2, AC084876.1 and AC005261.3 were high-risk lncRNAs for KIRC prognosis (HR >1) (Table S7). They had co-expression relationships with 4 glycosylation-related mRNAs, including advanced glycation end product receptor (AGER), dystroglycan 1 (DAG1), core 2 β-1, 6-N-acetylglucosaminyltransferase (GCNT3) and protein O-linked mannose β-1,4-N-acetylglucosaminyltransferase 2 (POMGNT2) (Figure 3E). These co-expression relationships provided the basis for defining the identified lncRNAs as GRLncs. Furthermore, the network suggested potential biological associations between the identified GRLncs and GRGs in KIRC. However, whether these lncRNAs directly regulate glycosylation-related pathways requires further experimental investigation.

Figure 3 The five GRLncs for building the prognostic model and their expression in KIRC tissues. (A) The expression of the five GRLncs in KIRC and normal tissues from TCGA database (blue: low expression; red: high expression). (B-D) Violin diagrams showed the expression of AC005261.3, DLGAP1-AS2 and EPB41L4A-DT in malignant cells, immune cells and stromal cells. (E) Sankey map showed the co-expression relationship between the five GRLncs and glycosylation-related mRNAs (AGER, DAG1, GCNT3 and POMGNT2), and the effect of five GRLncs on KIRC prognosis (yellow: protect; pink: risk). AGER, advanced glycation end product receptor; DAG1, dystroglycan 1; GCNT3, core 2 β-1, 6-N-acetylglucosaminyltransferase; N, normal; GRLncs, glycosylation-related lncRNAs; POMGNT2, protein O-linked mannose β-1; T, tumor.

The prognostic model shows promising predictive performance for KIRC patient prognosis

To evaluate the predictive performance of the prognostic model, we first calculated the risk score of 507 KIRC patients from TCGA database. Based on median risk score, we divided these patients into high-risk group and low-risk group (Figure 4A). The KM survival analysis demonstrated that the patients in the low-risk group had significantly longer OS than those in the high-risk group (Figure 4B). Moreover, survival status scatter plots showed that the mortality rate of KIRC patients rose up with the increase of risk score (Figure 4C), suggesting that higher risk scores were associated with poorer OS.

Figure 4 The correlation between the predictive signature and the prognosis of KIRC patients. (A) Based on the median risk score, the KIRC patients were divided into high-risk group and low-risk group. (B) KM survival analysis showed the OS of high-risk and low-risk KIRC patients. (C) The survival status of KIRC patients with different risk score. Blue indicates the alive patients, and pink indicates the dead patients. (D-E) GSEA showed the enriched pathways in the low-risk patients. GSEA, gene set enrichment analysis; KIRC, kidney renal clear cell carcinoma; KM, Kaplan-Meier; OS, overall survival.

Due to different prognosis of patients in the high- and low-risk groups, we conducted GSEA to study the possible differences between these two groups. GSEA analysis based on GO and KEGG analysis showed that adherens junction, tight junction, regulation of actin cytoskeleton, ubiquitin mediated proteolysis, transforming growth factor-beta (TGF-β) signaling pathway, cadherin binding, beta-catenin binding, negative regulation of vascular permeability and adherens junction organization were significantly enriched in the low-risk group (Figure 4D,4E, Table S8). However, homologous recombination, linoleic acid, base excision, base excision repair, P53 signaling pathway, primary immunodeficiency, synaptonemal complex organization, U1 snRNP, homologous chromosome segregation and granulocyte macrophage colony stimulating factor production were enriched in the high-risk group, but all the enrichments were not significant (FDR >0.25; Table S8). Evidence has shown that the adhesion-related signaling, TGF-β signaling, regulation of actin cytoskeleton and negative regulation of vascular permeability enhance intercellular adhesion, reduce vascular permeability, inhibit metastasis and maintain TME homeostasis in tumors (32). Therefore, these differences in biological processes may underlie the distinct prognostic characteristics of the two risk groups. However, these findings are based on pathway enrichment analyses and should be regarded as hypothesis-generating rather than evidence of direct molecular mechanisms.

The prediction accuracy of the prognostic model was validated by ROC curve. The results revealed that the AUC for evaluating the prediction accuracy of 1-, 3- and 5-year survival rate were 0.727, 0.74 and 0.765, respectively (Figure 5A), indicating that the five-GRLnc signature showed favorable prognostic performance in the TCGA cohort. As shown in Figure 5B, the AUC for risk score, age, gender, grade, stage, T-stage and M-stage were 0.758, 0.596, 0.482, 0.658, 0.712, 0.676 and 0.625, respectively, suggesting that the GRLnc-based risk score demonstrated potentially superior prognostic performance compared with conventional clinicopathological variables in this cohort. In addition, although univariate Cox regression analysis showed that the age, grade, stage, T-stage, N-stage, M-stage and risk score were significantly associated with OS of KIRC patients (Figure 5C), multivariate Cox regression analysis confirmed that the age and risk score were independent predictors of OS of KIRC patients (Figure 5D).

Figure 5 The predictive performance and reliability of the prognostic model based on the 5 GRLncs. (A) ROC curves showed the predictive accuracy of the prognostic model in assessing the 1-, 3-, and 5-year survival. (B) ROC curves demonstrated the predictive accuracy of risk score, age, gender, grade, stage, T-stage and M-stage in KIRC prognosis, in which the predictive accuracy of risk score was better than the others. (C,D) The univariate Cox regression analysis (C) and multivariate Cox regression analysis (D) showed the correlation of risk score, age, gender, grade, stage, T-stage, M-stage and N-stage with the OS of KIRC. (E) The relationship between the expression of the 5 prognostic GRLncs and clinicopathological variables in KIRC patients. (F-I) PCA showed that the high-risk group and low-risk group could be distinguished based on the 5 GRLncs in the model, but not all genes, glycosylation-related gene set, or GRLncs. AUC, area under the curve; GRLncs, glycosylation-related lncRNAs; KIRC, kidney renal clear cell carcinoma; M, metastasis; N, node; PCA, principal component analysis; ROC, receiver operating characteristic; T, tumor.

To further verify the predictive ability of the above prognostic model, we investigated the expression of these GRLncs and its relationship with the risk score. The results showed that the expression of high-risk factors—AC005261.3, DLGAP1-AS2 and AC084876.1 was higher in the high-risk group than that in the low-risk group, while the expression of protective factors—EPB41L4A-DT and AC093278.2 was lower in the high-risk group (Figure 5E). PCA based on the whole transcriptome, GRGs, or GRLncs alone showed considerable overlap between the two risk groups (Figure 5F-5H). In contrast, PCA based on the five-GRLnc signature clearly separated the high- and low-risk patients (Figure 5I), indicating that the selected signature provided better discrimination of patient risk status.

In addition, we assessed the 1-, 3- and 5-year OS of KIRC patients by constructing nomogram (Figure S1A). The calibration curves showed that the actual OS was highly consistent with the predicted OS (Figure S1B-S1D), suggesting that the nomogram could be used to predict 1-, 3-, and 5-year survival of clinical KIRC patients. To further assess the reliability of the model in predicting the prognosis of KIRC, we constructed the survival curves of KIRC patients based on age (≥65 and <65 years), gender, stage, T-stage and M-stage. Regardless of age, gender, stage, T and M stages, the KIRC patients with low-risk score had a higher OS than those with high-risk score (Figure S2A-S2J). These findings indicate that the prognostic performance of the GRLnc signature was maintained across multiple clinicopathological subgroups, supporting its potential prognostic value independent of conventional clinical variables.

In summary, the five-GRLnc signature showed promising prognostic performance for KIRC in TCGA-based analyses and showed potential value for risk stratification.

Differences in TME, immunotherapy response, and drug sensitivity between high- and low-risk groups

To further investigate the biological and clinical implications of the GRLnc-based risk signature, we comprehensively compared the TME, immunotherapy-related features, and predicted drug sensitivity between the high- and low-risk groups. We first applied the ESTIMATE algorithm to evaluate stromal and immune components within the TME. The results revealed that the high-risk group had a significantly higher Stromal Score and a significantly lower Immune Score compared with the low-risk group (Figure 6A), suggesting distinct microenvironmental characteristics and a potentially more stromal-enriched tumor milieu in high-risk patients. Further quantification of 17 immune cell types via ssGSEA demonstrated that the low-risk group was characterized by higher infiltration of Mast cells, Th17 cells, Dendritic cells and M2 Macrophages, whereas the high-risk group exhibited elevated infiltration of monocytes, Th2 cells, NK cells, Th1 cells, Tfh cells, neutrophils, plasma cells, exhausted T cells, and Treg CD8+ T cells (Figure 6B). These findings suggest substantial differences in immune landscape between the two risk groups and indicate a more complex immune regulatory environment in high-risk tumors. Analysis of immune checkpoint gene expression revealed markedly different expression profiles between the two risk groups. Multiple checkpoint molecules, including CD70, TNFSF9, KLRC1, CD80, ICOS, TIGIT, CD27, PDCD1, TNFRSF18, LAG3, and CTLA4 were significantly upregulated in the high-risk group, whereas NT5E, KIR3DL1, IDO1, KIR2DL3, ENTPD1, VSIR, TNFRSF4, PDCD1LG2, KIR2DL1, CD274, HAVCR2, ICOSLG, CD40, and CD276 were more highly expressed in the low-risk group (Figure 6C). Additionally, TIDE-related feature scores were calculated to evaluate the potential association between the GRLnc signature and immunotherapy response (Figure S2K). Although no statistically significant difference was observed in the overall TIDE score between the two groups (P>0.05), the marked differences in immune cell infiltration and immune checkpoint expression indicate that the GRLnc signature captures biologically relevant immunological characteristics of KIRC. Further validation in independent immunotherapy-treated cohorts will be required to determine its predictive value for immune checkpoint blockade response. Finally, drug sensitivity based on the GDSC database revealed that the low-risk group was more sensitive to BI-2536, axitinib, and tozasertib indicated by lower IC50 values, whereas the high-risk group displayed greater sensitivity to erlotinib, gefitinib, afatinib, vinblastine and sorafenib (Figure 6D). These findings suggest that patients classified into different GRLnc risk groups may exhibit distinct therapeutic vulnerabilities, indicating that the proposed signature could potentially assist in treatment stratification and personalized therapeutic decision-making.

Figure 6 Comprehensive landscape of microenvironment, immunity, and drug response in high- and low-risk groups. (A) Violin diagram showed the stromal score, immune score and estimate score in high-risk group and low-risk group. (B) ssGSEA algorithm was used to calculate the infiltration levels of 17 immune cells in high- and low-risk groups. (C) Comparison of expression levels of immune-related genes. (D) Box plots of IC50 sensitivity to targeted drugs. Blue represents the low-risk group and red represents the high-risk group. *, P<0.05; **, P<0.01; ***, P<0.001. CD8_T_cells, CD8-positive T cells; M1_macrophages, M1-type macrophages; M2_macrophages, M2-type macrophages; NK_cells, natural killer cells; ssGSEA, single sample gene set enrichment analysis; Tfh, follicular helper T cells; Th1, T helper 1 cells; Th17, T helper 17 cells; Th2, T helper 2 cells; Treg, regulatory T cells.

The internal validation supports the robustness of the GRLnc prognostic signatures

To further examine the prediction the stability and prognostic performance of the the GRLnc signature, we randomly divided 507 KIRC patients from TCGA database into the testing set (n=252) and the training set (n=255). The training set was used for model construction, whereas the testing set was used for internal validation of the model. KM survival analysis demonstrated patients in the low-risk group had significantly longer OS than those in the high-risk group in both the training and testing cohorts (Figure 7A,7B). Similarly, survival status scatter plots exhibited that the proportion of deaths increased with increasing risk scores in both cohorts (Figure 7C,7D). In addition, the ROC curves demonstrated that the AUCs of the 1-, 3- and 5-year survival rate were 0.692, 0.733 and 0.727 in the testing set, respectively, and 0.775, 0.757 and 0.804 in the training set, respectively (Figure 7E,7F). All the above results suggested that GRLnc signature showed consistent prognostic performance across different random partitions of the TCGA cohort.

Figure 7 The internal validation of the prognostic model. (A) KM survival curves of the high-risk group and low-risk group in the testing set. (B) KM survival curves of the high-risk group and low-risk group in the training set. (C) Scatter plot showed the survival status of high-risk group and low-risk group in the testing set. (D) Scatter plot showed the survival status of the high-risk group and low-risk group in the training set. (E) ROC curves showed the AUC of the 1-, 3-, and 5-year survival of the high-risk group and low-risk group in the testing set. (F) ROC curves showed the AUC of 1-, 3- and 5-year survival of the high-risk group and low-risk group in the training set. AUC, area under the curve; KM, Kaplan-Meier; ROC, receiver operating characteristic .

Clinical validation and in vitro functional characterization of three prognostic GRLncs

To further validate the clinical relevance of the identified GRLncs, we first examined the expression levels of DLGAP1-AS2, AC084876.1, and EPB41L4A-DT in 10 paired KIRC tumor tissues and adjacent normal tissues by RT-qPCR (table available at https://cdn.amegroups.cn/static/public/tau-2026-0289-1.xlsx). Consistent with the prognostic model, DLGAP1-AS2 and AC084876.1 tended to be upregulated in tumor tissues compared with adjacent normal tissues, although the differences did not reach statistical significance (Figure S3). In contrast, EPB41L4A-DT expression was significantly decreased in KIRC tissues (Figure S3), supporting its potential role as a protective factor in KIRC progression.

Subsequently, we investigated the biological functions of these GRLncs in KIRC cells. ASO was applied to knockdown their expression in Caki-1 and 786-O cell lines (Figure 8A-8C, Figure S4A-S4C, table available at https://cdn.amegroups.cn/static/public/tau-2026-0289-1.xlsx). Due to the inability to synthesize AC005261.3-ASO and AC093278.2-ASO, only the role of DLGAP1-AS2, EPB41L4A-DT and AC084876.1 in Caki-1 and 786-O cells was investigated in this study. The proliferative ability of Caki-1 and 786-O cells after DLGAP1-AS2, EPB41L4A-DT and AC084876.1 knockdown was detected by CCK8, EdU incorporation assay and clone formation assay. In CCK8 assay, we found that the cell viability of Caki-1 and 786-O cells was enhanced in the EPB41L4A-DT knockdown group compared to the control group, while the cell viability was reduced in the DLGAP1-AS2 and AC084876.1 knockdown groups (Figure 8D-8F, Figure S4D-S4F). Moreover, the results from EdU incorporation assay and clone formation assay were similar to CCK8 assay (Figure 8G,8H; Figure S4G). Moreover, we also investigated the role of these lncRNAs in the migration and invasion of Caki-1 and 786-O cells. As shown in Figure 8I, Figure S4H, and table available at https://cdn.amegroups.cn/static/public/tau-2026-0289-1.xlsx. DLGAP1-AS2 and AC084876.1 knockdown suppressed the migration of Caki-1 and 786-O cells, while EPB41L4A-DT knockdown held the opposite effects. As mentioned above, DLGAP1-AS2 and AC084876.1 were identified as high-risk factors, while EPB41L4A-DT was a protective factor for good prognosis of KIRC patients. Notably, the observed biological phenotypes were consistent with the risk directions predicted by the prognostic model. Specifically, knockdown of the high-risk lncRNAs DLGAP1-AS2 and AC084876.1 suppressed KIRC cell proliferation and migration, whereas silencing of the protective lncRNA EPB41L4A-DT promoted these malignant phenotypes. These findings provide biological support for the prognostic relevance of the identified GRLncs. Taken together, the clinical expression analysis and in vitro functional experiments support the potential involvement of DLGAP1-AS2, AC084876.1, and EPB41L4A-DT in KIRC progression and further strengthen the biological plausibility of the GRLnc-based prognostic signature.

Figure 8 The effect of the GRLncs in the prognostic model on Caki-1 cells verified by in vitro experiments. (A-C) The expression of AC084876.1, DLGAP1-AS2 and EPB41L4A-DT in Caki-1 cells after ASO transfection for 48 h (n=6 biological replicates). (D-F) The cell viability of Caki-1 cells after transfection with AC084876.1-ASO, DLGAP1-AS2-ASO and EPB41L4A-DT-ASO for 48 h (n=3). (G,H) The proliferation ability of Caki-1 cells after transfection with AC084876.1-ASO, DLGAP1-AS2-ASO and EPB41L4A-DT-ASO for 48 h. In EdU incorporation assay (G), cells (1×10³ per well) were seeded into 96-well plates (n=3). In colony formation assay (H), cells (1×10³ per well) were seeded into 6-well plates and cultured for 10 days (n=3). (I) The migration ability of Caki-1 cells after transfection with AC084876.1-ASO, DLGAP1-AS2-ASO and EPB41L4A-DT-ASO for 48 h. Cells (2×10⁵ per well) were resuspended with serum-free medium and seeded in the upper chambers. The medium containing 10% serum were added into the lower chambers. After 48 hours, the migrated cells were stained with crystal violet. The pictures of the migrated cells were captured (n=10 fields of view from three biological replicates). *, P<0.05; **, P<0.01; ***, P<0.001. ASO, antisense oligonucleotide; EDU, 5-ethynyl-2'-deoxyuridine; GRLncs, glycosylation-related lncRNAs; NC, negative control.

Discussion

In this study, we developed a prognostic signature based on 5 GRLncs (DLGAP1-AS2, EPB41L4A-DT, AC093278.2, AC084876.1 and AC005261.3) for patients with KIRC. This signature effectively stratified patients into high- and low-risk groups with significantly different OS outcomes and demonstrated favorable prognostic performance compared with conventional clinicopathological variables. Multiple complementary analyses, including time-dependent ROC curves, Cox regression, principal component analysis, nomogram construction, and internal validation using independent training and testing cohorts, consistently supported the robustness and prognostic value of the model. Furthermore, immune landscape and drug sensitivity analyses suggested that the GRLnc signature was associated with distinct TME characteristics and potential therapeutic vulnerabilities, indicating its potential value for individualized risk assessment and treatment stratification. Importantly, clinical RT-qPCR validation together with in vitro functional experiments provided biological evidence supporting the roles of three signature lncRNAs (DLGAP1-AS2, EPB41L4A-DT, and AC084876.1), with their expression patterns and functional phenotypes being consistent with the predicted risk directions of the prognostic model. Collectively, these findings support the biological relevance of the proposed GRLnc signature and highlight its potential as a promising prognostic framework for KIRC.

Aberrant glycosylation is increasingly recognized as a hallmark of cancer and contributes to multiple aspects of tumor progression, including cell proliferation, invasion, metastasis, immune escape, and therapeutic resistance. Previous studies have demonstrated that several glycosylation-related molecules are dysregulated in KIRC. For instance, fucosyltransferase 3 is a predictor of OS and relapse-free survival (33), while core 1 β1, 3-galactosyltransferase (C1GALT1) promote tumor progression through regulation of O-glycan structure on mucin (MUC) 1 (34). In addition, UDP-glycosyltransferase (UGT) catalyzes the covalent addition of sugars from nucleotide UDP-glycan donors to functional groups of endo-exogenous substances, and activates endogenous oncogenic signaling molecules via glucuronidation (35). Consistent with these findings, our transcriptomic analyses identified 18 differentially expressed GRGs in KIRC that were primarily enriched in glycosylation-associated biological processes and pathways, including mucin type O-glycan biosynthesis, N-glycan biosynthesis, protein glycosylation, macromolecular glycosylation, UDP-glycosyltransferase activity and acetylglucosamine transferase activity. These observations further support the close association between glycosylation-related biological processes and KIRC progression. Emerging evidence has indicated that different glycosylation modifications contribute to tumor progression through distinct mechanisms. O-glycan truncation promotes CD44v6-mediated activation of the receptor tyrosine kinase RON, enhancing invasiveness and metastasis (36). Irregular changes in N-glycosylation in the Golgi induce epithelial-mesenchymal transition (EMT) and cancer metastasis (37). EGFR sialylation leads to sustained activation of the phosphoinositide 3-kinase (PI3K)-AKT pathway, thereby promoting proliferation and anti-apoptosis (38). Sialyl Lewis antigens interact with E-selectin to promote distant metastasis (39). Fucosylation enhances the interaction of PD-1 and PD-L1 to facilitate tumor immune escape (40). In agreement with these observations, our GSEA and TME analyses demonstrated that the high-risk group exhibited enrichment of tumor-promoting biological processes together with distinct immune landscape characteristics, whereas the low-risk group was associated with pathways involved in tissue homeostasis. Furthermore, differential expression of immune checkpoint molecules and predicted drug sensitivities between the two risk groups suggests that glycosylation-related lncRNAs may reflect not only prognostic differences but also distinct immune states and therapeutic vulnerabilities. Although these findings do not establish direct mechanistic links, they support the hypothesis that glycosylation-related biological processes may participate in regulating tumor progression and the immune microenvironment in KIRC.

Based on the GRG set, we identified 968 GRLncs, from which five prognostic GRLncs were selected to construct the prognostic signature. Increasing evidence indicates that GRLncs participate in multiple aspects of tumor biology. In breast cancer, GRLnc mIR4435-2HG promotes proliferation, migration, and invasion by activating the Wnt/β-catenin axis, and mIR4435-2HG has also been implicated in progression of prostate, lung, and stomach malignancies (41). Likewise, depletion of GRLnc TGFB2-AS1 inhibits proliferation, migration and invasion of hepatocellular cancer cells (42). In addition, GRLncs also hold an effect on the prognosis of breast cancer and glioma (43,44). Collectively, these findings support the biological significance of GRLncs across diverse malignancies and provide a rationale for investigating their roles in KIRC. To note, GRGs were defined using a GeneCards relevance score cutoff (>7), consistent with previous studies (24). As different cutoffs may influence the composition of the GRG set and subsequent model construction, future studies should evaluate the robustness of the identified signature across different gene-selection criteria.

Numerous molecular prognostic signatures have been developed for human malignancies based on protein-coding genes, lncRNAs, miRNAs, and other biological processes (45-47). These studies demonstrate that transcriptome-derived signatures can effectively stratify patients according to clinical outcomes and facilitate individualized risk assessment. In KIRC, prognostic models based on migrasome-associated lncRNAs (7), mitophagy-related genes (48), and necroptosis-associated genes (49) have also been reported. However, despite increasing evidence supporting the critical role of aberrant glycosylation in KIRC progression, prognostic models specifically based on glycosylation-related lncRNAs remain largely unexplored.

In this study, we developed a prognostic model comprising 5 GRLncs (DLGAP1-AS2, EPB41L4A-DT, AC093278.2, AC084876.1 and AC005261.3) for KIRC. In comparison with conventional clinical variables (e.g., age, gender, TNM stage, and tumor grade), this model exhibited favorable prognosis performance. The time-dependent ROC analysis yielded AUC values ranging from 0.727 to 0.765, exceeding the AUCs for age (0.596), gender (0.482), TNM stage (0.712) and tumor grade (0.658). These findings suggest that the proposed GRLnc signature may serve as a promising prognostic tool that provides information complementary to traditional clinical indicators. In the prognostic model, DLGAP1-AS2, AC084876.1 and AC005261.3 were identified as high-risk factors, while AC093278.2 and EPB41L4A-DT were protective factors for good prognosis of KIRC patients. Also, these results were confirmed by performance assessment methods and internal validation. Similar to our results, previous studies have implicated these lncRNAs in tumor progression. DLGAP1-AS2 is increased in multiple malignancies and promotes tumor growth through activation of oncogenic pathways such as Wnt/β-catenin signaling (50) and TGF-β/Smad signaling pathways (51), thereby facilitating proliferation, EMT, and migration (51,52). AC084876.1 and AC005261.3 are associated with the poor prognosis of renal cell carcinoma (53) and colon cancer (54), whereas AC093278.2 and EPB41L4A-DT have already been reported to exert protective effects in KIRC (55,56). On this basis, these identified GRLncs may not only serve as prognostic biomarkers, but may also represent potential therapeutic targets in KIRC.

Notably, three signature lncRNAs were further validated in clinical specimens and functional experiments. RT-qPCR analysis of 10 paired KIRC tissues revealed that the protective lncRNA EPB41L4A-DT was significantly downregulated in tumor tissues, whereas the high-risk lncRNAs DLGAP1-AS2 and AC084876.1 exhibited a trend toward increased expression in tumors. Furthermore, knockdown experiments demonstrated that silencing DLGAP1-AS2 and AC084876.1 suppressed KIRC cell proliferation and migration, whereas inhibition of EPB41L4A-DT promoted these malignant phenotypes. Importantly, these biological effects were fully consistent with the risk directions predicted by the prognostic model, thereby providing experimental support for the biological relevance and prognostic significance of these signature lncRNAs.

Our findings further revealed co-expression relationships between the five GRLncs and four glycosylation-related mRNAs (AGER, DAG1, GCNT3 and POMGNT2). Previous studies have demonstrated that these genes are involved in tumor development and progression. AGER has been implicated in driving cancer growth, aggression, and metastasis through the fueling of chronic inflammation in the TME (57). Abnormal expression of DAG1 plays a role in both the process of tumor progression and the maintenance of the malignant phenotype (58). In contrast, GCNT3, co-expressed with the protective factor AC093278.2 in the prognostic model, exhibits a tumor growth-suppressive effect (59). POMGNT2, involved in the glycosylation of α-Dystroglycan, also acts as a tumor suppressor (60). Given these co-expression patterns, we hypothesize that the identified GRLncs may participate in KIRC progression through biological programs associated with glycosylation. Consistent with this possibility, GSEA and TME analysis suggest that immunodeficiency and activation of tumor-promoting signals were associated with poor prognosis of high-risk KIRC patients, while the maintenance of TME homeostasis might contribute to the better prognosis outcome of low-risk KIRC patients. Therefore, we hypothesize that the co-expressed GRLncs and mRNAs could modulate immunity, tumor-promoting signals, and TME homeostasis in KIRC. Nevertheless, the GRLnc-glycosylation network identified in this study was derived from co-expression analyses and therefore reflects potential biological associations rather than direct regulatory interactions. Consequently, the proposed links between GRLncs, glycosylation, immune regulation, and tumor progression should be regarded as hypothesis-generating. Further mechanistic studies, including validation of glycosylation-related targets and signaling pathways following lncRNA perturbation, will be required to establish causal relationships. Taken together, our findings highlight the potential importance of glycosylation-related lncRNAs in KIRC and provide a biologically supported prognostic framework as well as a foundation for future mechanistic investigations.

Several molecular signatures have been proposed for KIRC risk stratification. Among the most widely recognized, ClearCode34 is composed of 34 protein-coding genes, and classifies tumors into ccA and ccB molecular subtypes, demonstrating improved prognostic accuracy over conventional clinicopathological models, such as the University of California Los Angeles Integrated Staging System and the Mayo Clinic Stage, Size, Grade, and Necrosis score (61). In another study, Mehra et al. developed a 15-gene (15G) signature derived from a discovery cohort of 91 patients and externally validated in independent datasets, remaining significantly associated with disease-free survival and disease-specific survival after adjustment for clinicopathological variables and the cell cycle progression score (62). Other transcriptome-based approaches have also been proposed. Wang et al. constructed a 16-RNA prognostic score model comprising 3 lncRNAs, 6 miRNAs, and 7 mRNAs, which achieved favorable predictive performance in both training and validation cohorts (63). More recently, Xu et al. developed an intratumor heterogeneity (ITH)-based prognostic model (ITHscore) consisting of three genes (UBE2C, MOCOS, and MELTF), which was derived from DEPTH-defined ITH features and validated in an independent external cohort. The ITHscore demonstrated excellent predictive performance, with a 5-year AUC of 0.957 in the training cohort and 0.820 in the validation cohort, and was associated with immunosuppressive TMEs and signatures of immune checkpoint blockade resistance (64).

Compared with these previously reported models, our GRLnc signature possesses several distinctive features. First, unlike protein-coding gene-based classifiers (e.g., ClearCode34, the 15-gene signature, and the ITHscore) or multi-omic signatures, our model focuses specifically on GRLncs, thereby capturing an emerging biological process that has been increasingly implicated in tumor progression, immune regulation, and therapeutic response. While the recently proposed ITHscore reflects tumor aggressiveness through intratumor heterogeneity and immune evasion characteristics, our glycosylation-centered design provides complementary mechanistic information that may not be captured by existing molecular classification systems. To our knowledge, this is the first prognostic model specifically based on GRLncs in KIRC. Second, the model consists of only five lncRNAs, which may facilitate clinical implementation through cost-effective and technically accessible assays. Third, beyond computational model construction, we experimentally validated the biological functions of three signature lncRNAs, demonstrating their involvement in KIRC cell proliferation and migration and thereby strengthening the biological plausibility of the risk score. With respect to predictive performance, the time-dependent AUC values of our model (0.727–0.765) are comparable to those reported for several recently published transcriptomic signatures (Wang et al.’s validation cohort 0.714–0.778). Although the ITHscore achieved higher predictive accuracy, direct comparisons should be interpreted cautiously because differences in patient populations, study design, modeling strategies, clinical endpoints, and validation frameworks may substantially influence performance estimates. Importantly, whereas ClearCode34 and several other models have undergone external validation, our current study is limited to internal validation due to the lack of publicly available cohorts containing complete expression profiles of all five GRLncs. Therefore, future multicenter studies and prospective validation efforts will be necessary to further establish the robustness and clinical utility of this glycosylation-related prognostic framework.

Recently, Costa Filho et al. proposed a framework for assessing the translational readiness of genomic risk stratification models, emphasizing external validation, reproducibility, biological plausibility, comparison with existing models, clinical utility, and prospective evaluation (65). Viewed within this framework, our GRLnc signature demonstrates several strengths. The model showed stable prognostic performance in both training and testing cohorts and remained an independent predictor of OS after adjustment for conventional clinicopathological variables. Moreover, we compared our signature with several representative molecular prognostic models for KIRC and found that it provides a distinct glycosylation-centered biological perspective. The biological relevance of the model was further supported by enrichment analyses and functional validation experiments, in which three signature GRLncs consistently influenced KIRC cell proliferation and migration in accordance with their predicted risk effects. In addition, the integration of the risk score into a nomogram suggests potential clinical applicability. However, several criteria required for clinical translation have not yet been fully satisfied. In particular, independent external validation and cross-platform reproducibility could not be assessed because no publicly available cohort contains complete expression data for all five GRLncs. Furthermore, the model has not undergone prospective evaluation to determine whether its use can improve clinical decision-making or patient outcomes. Therefore, while our findings support the GRLnc signature as a biologically meaningful and promising prognostic framework for KIRC, further multicenter and prospective studies will be required before clinical implementation can be considered.

Several limitations of this study should be acknowledged. First, this retrospective study was based exclusively on TCGA data, and although internal validation demonstrated stable prognostic performance, independent external validation was not feasible because no public cohort currently contains complete expression profiles of all five GRLncs. Therefore, the generalizability of the model requires further confirmation. Second, while functional experiments supported the roles of three signature GRLncs in KIRC progression, the biological functions of AC093278.2 and AC005261.3 remain to be investigated. Future studies employing xenograft or genetically engineered animal models will be necessary to further confirm their functional significance. Third, our findings linking GRLncs to glycosylation are currently supported by co-expression and pathway analyses, whereas direct regulatory mechanisms require further experimental validation. Finally, the selection of the GeneCards score cutoff (>7) was based on literature precedent, but sensitivity analyses using different cutoffs were not performed. Future studies should address these limitations through multicenter external validation, cross-platform reproducibility assessment, prospective clinical evaluation, comprehensive functional characterization of all model lncRNAs, and mechanistic investigations of glycosylation regulation.

Conclusions

In conclusion, we developed a prognostic model based on five GRLncs (DLGAP1-AS2, EPB41L4A-DT, AC093278.2, AC084876.1, AC005261.3) that demonstrated promising predictive performance in the TCGA dataset. The identified GRLncs may provide insights into the biological processes associated with KIRC progression and warrant further mechanistic and clinical investigation.

Acknowledgments

We would like to thank Servicebio for the cell lines donation and the Biobank of West China Hospital, Sichuan University, for providing KIRC tissues.

Footnote

Reporting Checklist: The authors have completed the TRIPOD and MDAR reporting checklists. Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0289/rc

Data Sharing Statement: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0289/dss

Peer Review File: Available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0289/prf

Funding: This work was supported by Natural Science Foundation of Sichuan Province of China (No. 2023NSFSC1567) and National Natural Science Foundation of China (No. W2433199).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tau.amegroups.com/article/view/10.21037/tau-2026-0289/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Biomedical Ethics Committee of West China Hospital, Sichuan University (approval No. 2025-986). Written informed consent was obtained from all participants.

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. Jiang H, Chen H, Wan P, et al. Decreased expression of HADH is related to poor prognosis and immune infiltration in kidney renal clear cell carcinoma. Genomics 2021;113:3556-64. [Crossref] [PubMed]
2. Ljungberg B, Albiges L, Abu-Ghanem Y, et al. European Association of Urology Guidelines on Renal Cell Carcinoma: The 2022 Update. Eur Urol 2022;82:399-410. [Crossref] [PubMed]
3. Hsieh JJ, Purdue MP, Signoretti S, et al. Renal cell carcinoma. Nat Rev Dis Primers 2017;3:17009. [Crossref] [PubMed]
4. Kirkali Z, Algaba F, Scarpelli M, et al. What does the urologist expect from the pathologist (and what can the pathologists give) in reporting on adult kidney tumour specimens? Eur Urol 2007;51:1194-201. [Crossref] [PubMed]
5. Bravo À, Cases M, Queralt-Rosinach N, et al. A knowledge-driven approach to extract disease-related biomarkers from the literature. Biomed Res Int 2014;2014:253128. [Crossref] [PubMed]
6. Dong P, Zhao L, Shi P, et al. Nicotinamide metabolism-related signature and lncRNA regulatory network in kidney renal clear cell carcinoma. PeerJ 2026;14:e21300. [Crossref] [PubMed]
7. Zhao L, Geng F, Ji X, et al. Construction of a risk and prognostic model for migrasome-associated lncRNAs in renal cell carcinoma. Sci Rep 2025;15:26760. [Crossref] [PubMed]
8. Shi X, Song B, Wang B, et al. Development and validation of a TOP2A based index clinical model for kidney renal clear cell carcinoma. Transl Oncol 2026;64:102625. [Crossref] [PubMed]
9. Christiansen MN, Chik J, Lee L, et al. Cell surface protein glycosylation in cancer. Proteomics 2014;14:525-46. [Crossref] [PubMed]
10. Hang Q, Li W, Guo J, et al. Inhibitory effects of β-galactoside α2,6-sialyltransferase 1 on the Hippo pathway in breast cancer cells. J Biol Chem 2025;301:110266. [Crossref] [PubMed]
11. Leviatan Ben-Arye S, Schmidt EN, Jame-Chenarboo Z, et al. An atlas of human siglecs integrates expression, affinity, and cis/trans sialoglycan recognition profiles. Nat Commun 2026;17:5755. [Crossref] [PubMed]
12. Li N, Hu B, Liu C, et al. Targeting FUT1-mediated EGFR fucosylation and COX-2 reverses gemcitabine resistance by activating immunogenic cell death in hepatocellular carcinoma. Bioorg Chem 2025;166:109111. [Crossref] [PubMed]
13. Huang Y, Song J, Chen J, et al. The significance of protein N-glycosylation in the pathogenesis of lung cancer and its clinical implications. Cancer Lett 2025;631:217849. [Crossref] [PubMed]
14. Iyer SC, Srinivasan DK, Parameswaran R. GalNAc-Transferases in Cancer. Biomedicines 2025;14:5. [Crossref] [PubMed]
15. Qusairy Z, Rada M. Glycosylation in cancer: mechanisms, diagnostic markers, and therapeutic applications. Mol Cell Biochem 2025;480:4941-59. [Crossref] [PubMed]
16. Zhang Y, Lai H, Tang B. Abnormal Expression and Prognosis Value of COG Complex Members in Kidney Renal Clear Cell Carcinoma (KIRC). Dis Markers 2021;2021:4570235. [Crossref] [PubMed]
17. Hu X, Sood AK, Dang CV, et al. The role of long noncoding RNAs in cancer: the dark matter matters. Curr Opin Genet Dev 2018;48:8-15. [Crossref] [PubMed]
18. Wei JW, Huang K, Yang C, et al. Non-coding RNAs as regulators in epigenetics Oncol Rep 2017;37:3-9. (Review). [Crossref] [PubMed]
19. Wang J, Zou Y, Du B, et al. SNP-mediated lncRNA-ENTPD3-AS1 upregulation suppresses renal cell carcinoma via miR-155/HIF-1α signaling. Cell Death Dis 2021;12:672. [Crossref] [PubMed]
20. Ye Y, Zhang F, Chen Q, et al. LncRNA MALAT1 modified progression of clear cell kidney carcinoma (KIRC) by regulation of miR-194-5p/ACVR2B signaling. Mol Carcinog 2019;58:279-92. [Crossref] [PubMed]
21. Li Y, Zeng C, Hu J, et al. Long non-coding RNA-SNHG7 acts as a target of miR-34a to increase GALNT7 level and regulate PI3K/Akt/mTOR pathway in colorectal cancer progression. J Hematol Oncol 2018;11:89. [Crossref] [PubMed]
22. Hu J, Shan Y, Ma J, et al. LncRNA ST3Gal6-AS1/ST3Gal6 axis mediates colorectal cancer progression by regulating α-2,3 sialylation via PI3K/Akt signaling. Int J Cancer 2019;145:450-60. [Crossref] [PubMed]
23. Shen X, Liu Y, Zhang J, et al. Predictive Significance of Glycosyltransferase-Related lncRNAs in Endometrial Cancer: A Comprehensive Analysis and Experimental Validation. ACS Omega 2025;10:8023-41. [Crossref] [PubMed]
24. Bai HY, Li TT, Sun LN, et al. Development of a Novel Prognostic Model for Lung Adenocarcinoma Utilizing Pyroptosis-Associated LncRNAs. Anal Cell Pathol (Amst) 2025;2025:4488139. [Crossref] [PubMed]
25. Gu J, He Z, Huang Y, et al. Clinicopathological and Prognostic Value of Necroptosis-Associated lncRNA Model in Patients with Kidney Renal Clear Cell Carcinoma. Dis Markers 2022;2022:5204831. [Crossref] [PubMed]
26. Sun YF, Chen L, Xia QJ, et al. Identification of necroptosis-related long non-coding RNAs prognostic signature and the crucial lncRNA in bladder cancer. J Cancer Res Clin Oncol 2023;149:10217-34. [Crossref] [PubMed]
27. Chen X, Zhang Y, Wang F, et al. A Novel Assessment Model Based on Molecular Subtypes of Hypoxia-Related LncRNAs for Prognosis of Bladder Cancer. Front Cell Dev Biol 2021;9:718991. [Crossref] [PubMed]
28. Xiu B, Chi Y, Liu L, et al. LINC02273 drives breast cancer metastasis by epigenetically increasing AGR2 transcription. Mol Cancer 2019;18:187. [Crossref] [PubMed]
29. Liu Y, Shi M, He X, et al. LncRNA-PACERR induces pro-tumour macrophages via interacting with miR-671-3p and m6A-reader IGF2BP2 in pancreatic ductal adenocarcinoma. J Hematol Oncol 2022;15:52. [Crossref] [PubMed]
30. Croci DO, Cerliani JP, Pinto NA, et al. Regulatory role of glycans in the control of hypoxia-driven angiogenesis and sensitivity to anti-angiogenic treatment. Glycobiology 2014;24:1283-90. [Crossref] [PubMed]
31. Mustafa M, Ahmad R, Tantry IQ, et al. Apoptosis: A Comprehensive Overview of Signaling Pathways, Morphological Changes, and Physiological Significance and Therapeutic Implications. Cells 2024;13:1838. [Crossref] [PubMed]
32. Zhou W, Fong MY, Min Y, et al. Cancer-secreted miR-105 destroys vascular endothelial barriers to promote metastasis. Cancer Cell 2014;25:501-15. [Crossref] [PubMed]
33. Borzym-Kluczyk M, Radziejewska I, Darewicz B. Glycosylation of proteins in healthy and pathological human renal tissues. Folia Histochem Cytobiol 2012;50:599-604. [Crossref] [PubMed]
34. Chou CH, Huang MJ, Chen CH, et al. Up-regulation of C1GALT1 promotes breast cancer cell growth through MUC1-C signaling pathway. Oncotarget 2015;6:6123-35. [Crossref] [PubMed]
35. Allain EP, Rouleau M, Lévesque E, et al. Emerging roles for UDP-glucuronosyltransferases in drug resistance and cancer progression. Br J Cancer 2020;122:1277-87. [Crossref] [PubMed]
36. Mereiter S, Martins ÁM, Gomes C, et al. O-glycan truncation enhances cancer-related functions of CD44 in gastric cancer. FEBS Lett 2019;593:1675-89. [Crossref] [PubMed]
37. Xu X, Peng Q, Jiang X, et al. Altered glycosylation in cancer: molecular functions and therapeutic potential. Cancer Commun (Lond) 2024;44:1316-36. [Crossref] [PubMed]
38. Gong A, Zhao X, Pan Y, et al. The lncRNA MEG3 mediates renal cell cancer progression by regulating ST3Gal1 transcription and EGFR sialylation. J Cell Sci 2020;133:jcs244020. [Crossref] [PubMed]
39. Tozawa K, Okamoto T, Kawai N, et al. Positive correlation between sialyl Lewis X expression and pathologic findings in renal cell carcinoma. Kidney Int 2005;67:1391-6. [Crossref] [PubMed]
40. Bastian K, Scott E, Elliott DJ, et al. FUT8 Alpha-(1,6)-Fucosyltransferase in Cancer. Int J Mol Sci 2021;22:455. [Crossref] [PubMed]
41. Chen D, Tang P, Wang Y, et al. Downregulation of long non-coding RNA MR4435-2HG suppresses breast cancer progression via the Wnt/β-catenin signaling pathway. Oncol Lett 2021;21:373. [Crossref] [PubMed]
42. Liu W, Huai R, Zhang Y, et al. Down-regulation expression of TGFB2-AS1 inhibits the proliferation, migration, invasion and induces apoptosis in HepG2 cells. Genes Genomics 2019;41:951-9. [Crossref] [PubMed]
43. Lv W, Tan Y, Zhou X, et al. Landscape of prognosis and immunotherapy responsiveness under tumor glycosylation-related lncRNA patterns in breast cancer. Front Immunol 2022;13:989928. [Crossref] [PubMed]
44. Wu X, Wang H, Li S, et al. Mining glycosylation-related prognostic lncRNAs and constructing a prognostic model for overall survival prediction in glioma: A study based on bioinformatics analysis. Medicine (Baltimore) 2023;102:e33569. [Crossref] [PubMed]
45. Dai S, Huang Y, Liu T, et al. Development and validation of RNA binding protein-applied prediction model for gastric cancer. Aging (Albany NY) 2021;13:5539-52. [Crossref] [PubMed]
46. Tang R, Wu Z, Rong Z, et al. Ferroptosis-related lncRNA pairs to predict the clinical outcome and molecular characteristics of pancreatic ductal adenocarcinoma. Brief Bioinform 2022;23:bbab388. [Crossref] [PubMed]
47. Zhao Y, Xu L, Wang X, et al. A novel prognostic mRNA/miRNA signature for esophageal cancer and its immune landscape in cancer progression. Mol Oncol 2021;15:1088-109. [Crossref] [PubMed]
48. Duan D, Guo Y, Li J, et al. Mitophagy related gene signature for prognosis and therapeutic evaluation in KIRC. Sci Rep 2025;15:27273. [Crossref] [PubMed]
49. Zhang X, Jin Q, Cheng G, et al. Identification of necroptosis-associated mRNA biomarkers in kidney clear cell carcinoma. Front Immunol 2025;16:1545486. [Crossref] [PubMed]
50. Lu J, Xu Y, Xie W, et al. Long noncoding RNA DLGAP1-AS2 facilitates Wnt1 transcription through physically interacting with Six3 and drives the malignancy of gastric cancer. Cell Death Discov 2021;7:255. [Crossref] [PubMed]
51. Cao H, Zhu J, Zhang L, et al. lncRNA DLGAP1-AS2 Modulates the TGF-β/Smad Signaling Pathway to Regulate Epithelial-Mesenchymal Transition in Renal Cell Carcinoma. Biochem Genet 2026; Epub ahead of print. [Crossref] [PubMed]
52. Zhao Y, Wu X, Zheng Y, et al. Prognostic value of LncRNA DLGAP1-AS2 in bladder cancer and its regulatory effect on tumor progression. Discov Oncol 2025;16:1578. [Crossref] [PubMed]
53. Cao H, Tong H, Zhu J, et al. A Glycolysis-Based Long Non-coding RNA Signature Accurately Predicts Prognosis in Renal Carcinoma Patients. Front Genet 2021;12:638980. [Crossref] [PubMed]
54. Chen W, Chen Y, Liu L, et al. Comprehensive Analysis of Immune Infiltrates of Ferroptosis-Related Long Noncoding RNA and Prediction of Colon Cancer Patient Prognoses. J Immunol Res 2022;2022:9480628. [Crossref] [PubMed]
55. Zhou Z, Yang Z, Cui Y, et al. Identification and Validation of a Ferroptosis-Related Long Non-Coding RNA (FRlncRNA) Signature to Predict Survival Outcomes and the Immune Microenvironment in Patients With Clear Cell Renal Cell Carcinoma. Front Genet 2022;13:787884. [Crossref] [PubMed]
56. Sun Z, Jing C, Xiao C, et al. Long Non-Coding RNA Profile Study Identifies an Immune-Related lncRNA Prognostic Signature for Kidney Renal Clear Cell Carcinoma. Front Oncol 2020;10:1430. [Crossref] [PubMed]
57. Mondal S, Chakraborty P, Ghatak S, et al. Receptor for Advanced Glycation End Products (RAGE) Ligand Axis as a Mediator of Inflammation and Oxidative Stress in Cancer: Implications for Cancer Progression and Therapeutic Targeting. Curr Pharm Des 2026; Epub ahead of print. [Crossref] [PubMed]
58. Aggelis V, Craven RA, Peng J, et al. VHL-dependent regulation of a β-dystroglycan glycoform and glycogene expression in renal cancer. Int J Oncol 2013;43:1368-76. [Crossref] [PubMed]
59. Fernández LP, Sánchez-Martínez R, Vargas T, et al. The role of glycosyltransferase enzyme GCNT3 in colon and ovarian cancer prognosis and chemoresistance. Sci Rep 2018;8:8485. [Crossref] [PubMed]
60. Quereda C, Pastor À, Martín-Nieto J. Involvement of abnormal dystroglycan expression and matriglycan levels in cancer pathogenesis. Cancer Cell Int 2022;22:395. [Crossref] [PubMed]
61. Brooks SA, Brannon AR, Parker JS, et al. ClearCode34: A prognostic risk predictor for localized clear cell renal cell carcinoma. Eur Urol 2014;66:77-84. [Crossref] [PubMed]
62. Mehra R, Nallandhighal S, Cotta B, et al. Discovery and Validation of a 15-Gene Prognostic Signature for Clear Cell Renal Cell Carcinoma. JCO Precis Oncol 2024;8:e2300565. [Crossref] [PubMed]
63. Wang PY, Yang S, Bao YJ. An Integrative Analysis Framework for Identifying the Prognostic Markers from Multidimensional RNA Data of Clear Cell Renal Cell Carcinoma. Am J Pathol 2022;192:671-86. [Crossref] [PubMed]
64. Filho VOC, Passos PRC, Noronha MM, et al. Development and Validation of an Intratumor Heterogeneity-Based Prognostic Model for Clear Cell Renal Cell Carcinoma. JCO Precis Oncol 2025;9:e2500709. [Crossref] [PubMed]
65. Costa Filho VO, Passos PRC. Precision over quantity: identifying trustworthy genomic risk stratification models. Einstein (Sao Paulo) 2025;23:eCE1726. [Crossref] [PubMed]