O-GlcNAcylation-related genes mediate tumor microenvironment characteristics and prediction of immunotherapy response in gastric cancer

Abstract

We aim to identify molecular clusters related to O-GlcNAcylation and establish a novel scoring system for predicting prognosis and immunotherapy efficacy in patients with gastric cancer (GC). The transcriptomic and clinical data are obtained from XENA-UCSC and GEO databases. The O-GlcNAcylation-related genes are obtained from the GSEA database. Consensus clustering analysis is employed to identify O-GlcNAcylation-related molecular clusters, and principal component analysis (PCA) is utilized to develop a novel prognostic scoring system for predicting GC outcomes and immunotherapy efficacy. The prognostic accuracy of the scoring system is assessed across five real-world cohorts. The biological function of actin alpha 2, smooth muscle (ACTA2) in GC is determined through experimental verification. Using 34 O-GlcNAcylation-related genes associated with prognosis in GC patients, these individuals are divided into two distinct subgroups characterized by different outcomes, tumor microenvironment profiles, and clinical case characteristics. The DEGs between the two subgroups are subsequently used to further divide the GC patients into two subgroups by consensus cluster analysis. PCA is used to construct a prognostic scoring system, which reveal that patients in the low-score subgroup have a better prognosis and greater benefit from immunotherapy. The accuracy of the scoring system is confirmed through validation in a cohort of patients receiving immunotherapy in the real world. ACTA2 promotes proliferation and inhibits apoptosis in GC cells. These findings suggest that we successfully establish molecular clusters associated with O-GlcNAcylation and develop a scoring system that demonstrates strong performance in predicting the prognosis of patients with GC and the effect of immunotherapy interventions.

Introduction

Gastric cancer (GC) is the fifth most common cause of cancer death worldwide [1]. Because GC is often diagnosed at an advanced stage, it has a high mortality rate and is the third leading cause of cancer-related death worldwide, with 784,000 deaths in 2018 [2]. Despite advances in checkpoint inhibitor-based cancer immunotherapy [3], the prognosis for patients with GC remains unfavorable, which poses challenges for clinicians. Therefore, it is necessary to further study the mechanisms of different responses to immunotherapy and to design prognostic and efficacy prediction tools.

O-linked N-acetylglucosaminylation is a form of glycosylation that occurs when monosaccharide O-GlcNAc is added to serine or threonine residues of nuclear or cytoplasmic proteins by O-GlcNAc transferase (OGT) and can be reversibly removed by O-GlcNAcase (OGA) [4]. O-GlcNAcylation is the subject of extensive research in the field of oncology. It is significantly upregulated in most cancer types and plays a key role in integrating nutrient fluxes with metabolic pathways essential for tumor cell proliferation and growth. Moreover, O-GlcNAcylation has a regulatory function on many proteins involved in the initiation and development of cancer [ 5, 6]. For example, O-GlcNAcylation plays a significant role in enhancing the stability of the Reticulon 2 protein, thereby contributing to GC progression [7]. Wu L et al. [8] focused on the role of O-GlcNAcylation in tumor immune evasion. Specifically, O-GlcNAcylation plays a catalytic role in this process by inhibiting the lysosomal degradation of programmed death-ligand 1 (PD-L1). This mechanism suggests that O-GlcNAcylation is a potential intervention target for enhancing the tumor immune response. OGT is also involved in transcriptional regulation by regulating transcription factors [ 9– 11]. Studies have shown that O-GlcNAcylation of CD36 enhances cellular fatty acid uptake activity, which is critical for GC invasion [12]. Although many studies have investigated the relationship between O-GlcNAcylation and GC, relatively few studies have investigated O-GlcNAcylation as a predictive biomarker. Thus, novel predictive biomarkers are needed to improve the accuracy of immune checkpoint inhibitor (ICI) therapy in GC patients.

In this study, 675 GC samples were obtained from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. We proposed and produced a feasible 2-subtype classification system based on O-GlcNAcylation-related genes and characterized molecular features at the transcriptomic level. Two gene subgroups were subsequently identified on the basis of the differentially expressed genes (DEGs) between the two distinct O-GlcNAcylation subgroups. Our analysis comprehensively covered alterations in transcription, somatic mutations and copy number variations. A reliable prognostic scoring system was constructed to predict the prognosis of GC patients. This system may have potential therapeutic and prognostic implications for GC management, especially immunotherapy.

Materials and Methods

Data source and processing

Gene expression data, somatic mutation data, and corresponding clinicopathological information of GC patients were downloaded from the UCSC-XENA database (TCGA-STAD, https://gdc.xenahubs.net). The GSE62254 dataset was downloaded from the GEO database ( https://www.ncbi.nlm.nih.gov/geo/). The RNA-seq data from the UCSC-XENA database were obtained as FPKM values and then converted to TPM values. Batch correction was performed via the R packages limma and sva (version 4.2.2). A total of 675 samples were included in this study: 375 from the TCGA-STAD dataset and 300 from the GSE62254 dataset. A total of 151 genes related to O-GlcNAcylation were retrieved from the GSEA database (GOBP_PROTEIN_O_LINKED_GLYCOSYLATION, http://www.gseamsigdb.org/gsea/msigdb/human/geneset/GOBP_PROTEIN_O_LINKED_GLYCOSYLATION; REACTOME_O_LINKED_GLYCOSYLATION http://www.gseamsigdb.org/gsea/msigdb/human/geneset/REACTOME_O_LINKED_GLYCOSYLATION ).

Unsupervised clustering of the O-GlcNAcylation modification pattern

To identify O-GlcNAcylation-related genes (OGRGs) associated with survival, a univariate regression analysis was performed with P < 0.01. On the basis of survival-related OGRGs, the R package ConsensusClusterPlus was used for unsupervised clustering to classify the samples. The optimal classification resulted in two clusters. Principal component analysis (PCA) was employed to determine whether each isoform is relatively independent of the other isoforms.

Construction of gene clusters

The R package limma (version 4.2.2) was used to identify the DEGs among 2 clusters with P < 0.05 and a |LogFC| > 1. Survival-related DEGs were identified via univariate Cox regression analysis with P < 0.00000001, and GC patients were classified into two distinct gene clusters on the basis of selected DEGs via the R package “ConsensuClusterPlus”.

Establishment of an O-GlcNAcylation-related prognostic scoring system

We adopted the PCA algorithm to create a scoring system based on survival-related DEGs among the clusters in STAD named the OGScore according to the formula OGScore = ∑ (PC1 + PC2), where PC1 represents the largest proportion of the variance in the initial expression lineage to be decomposed, followed by PC2. All patients were classified into low- and high-OGScore groups at the optimal cutoff point, which was calculated via the survminerR package (version 0.4.9).

Enrichment analysis and functional annotation

The Biocarta pathways, hallmark pathways, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, reactome pathways, and Wikipathways were downloaded from the MSigDB database. Gene Set Variation Analysis (GSVA) enrichment was performed to explore the heterogeneity of various biological processes via the “GSVA” package. For the purpose of gene set enrichment analysis, we utilized the “clusterProfiler” package (version 4.0) to perform functional enrichment analysis on Gene Ontology (GO) and KEGG.

Evaluating the TME and immune cell infiltration

To explore the differences in the tumor microenvironment (TME) between the two subgroups, we employed the ESTIMATE algorithm to analyze the stromal score, immune score, and ESTIMATE score between the two clusters. The infiltration level of 23 immune cells in each sample was calculated via the ssGSEA function from the R package GSVA, and the Wilcoxon test was performed to analyze immune cell abundance across different clusters. Spearman correlations between the OGScore and immune cells were also analyzed.

Genomic alteration analysis

Analysis of copy number variations was performed via the Gene Set Cancer Analysis (GSCA) database ( http://bioinfo.life.hust.edu.cn/GSCA/#/). To calculate the gene mutation status between the high- and low-OGScore groups, we utilized the R package “maftools”. Afterwards, a waterfall plot was generated to visualize the mutations. We also analyzed the differences in mutations between the high- and low-OGScore groups.

Calculation of the stemness index

We applied the methodology proposed by Malta et al. [13] to compute the stemness index for each sample obtained from a GC patient. This approach uses a one-class logistic regression machine-learning algorithm (OCLR) and was built on pluripotent stem cell samples, which have been shown to be strongly correlated with stem cell characteristics and capable of predicting cancer stemness. We subsequently conducted a Spearman correlation analysis to examine the relationships between the O-GlcNAcylation score and the mRNA/mDNA indices.

Immunotherapy response and antitumor drug sensitivity

Five datasets were utilized to assess the predictive efficacy of OGScores for immune therapy response: late stage non-squamous non-small cell lung cancer treated with bevacizumab erlotinib (GSE61676), melanoma treated with pembrolizumab, an anti-PD-1 antibody (GSE78220), advanced non-small cell lung carcinoma treated with anti-PD-1/PD-L1 (GSE135222), advanced urothelial cancer treated with atezolizumab, an anti-PD-L1 antibody (IMvigor210 cohort, http://research-pub.gene.com/IMvigor210CoreBiologies), and kidney renal clear cell carcinoma treated with nivolumab (anti-PD-1) and the mTOR inhibitor everolimus. The R package pRRophetic was used to predict the IC ₅₀ values of multiple anticancer drugs for each sample, while chemotherapeutic medications were sourced from the Drug Sensitivity in Cancer (GDSC) genome database ( https://www.cancerrxgene.org/), and the differences in the scores of high- and low-ranking arrays were compared.

Tissue specimens

GC patients with tumor node metastasis (TNM) stages I–II who underwent curative gastrectomy at the Department of Gastrointestinal Surgery of the First Affiliated Hospital of Nanjing Medical University provided tumor and adjacent non-cancerous gastric mucosa samples. Prior to surgery, these patients had not received radiotherapy or chemotherapy. The samples were collected in accordance with the Health Insurance Portability and Accountability Act (HIPAA) guidelines and were approved by the Ethics and Human Subject Committee of The First Affiliated Hospital of Nanjing Medical University (number: 2023-SR-987).

Cell culture and transfection

We used the human GC cell lines MKN-45, NCI-N87, HGC-27, and AGS, as well as GES-1 cells, a normal gastric mucosal epithelial cell line acquired from the Shanghai Institutes for Biological Sciences Cell Center (Shanghai, China). The AGS cell line was cultured in F12K medium (Invitrogen, Carlsbad, USA), and the MKN-45, NCI-N87, HGC-27 and GES-1 cell lines were cultivated in RPMI 1640 medium (Invitrogen). Both media were supplemented with 10% fetal bovine serum (Invitrogen) and 1% penicillin/streptomycin (Gibco, Carlsbad, USA) and maintained at 37°C in a 5% CO ₂ humidified atmosphere.

For cell transfection, GC cells were seeded into 6-well plates and grown overnight. The next day, when the cell plating density reached 20%-30%, GC cells were transfected with ACTA2-OE, si-ACTA2-1, si-ACTA2-2 (GenePharma, Shanghai, China), or negative control (Vector or si-NC) by Lipofectamine 2000 (Invitrogen, Carlsbad, USA) according to the manufacture’s instructions. The sequence of ACTA2-OE was 5′-CGAATGTGGGAATTACTTGAAGG-3′; the sequence of si-ACTA2-1 was 5′-GGAAGCGGAUCAUGGUGAUTT-3′; the sequence of si-ACTA2-2 was 5′-GGACGUGUUCGACAUGGAATT-3′; the sequence of Vector was 5′-UCACAGUGAACCGGUCUCUUU-3′; and the sequence of si-NC was 5′-GTTCTCCGAACGTGTCACGT-3′. At 48 h post-transfection, cells were harvested for RT-qPCR and western blot analysis,

RNA preparation and quantitative reverse transcriptase polymerase chain reaction (qRT-PCR)

RNA was extracted from the samples via TRIzol reagent (Invitrogen) according to the manufacturer’s instructions. The extracted RNAs were converted into cDNA via the PrimeScript RT Reagent (RR036A; TaKaRa, Kyoto, Japan). Subsequent qRT-PCR assays were performed using a 7500 Real-time PCR System (Applied Biosystems, Carlsbad, USA) and Universal SYBR Green Master Mix (Roche, Shanghai, China). The sequences of primers (GenePharma, Shanghai, China) used for amplification were as follows: actin alpha 2, smooth muscle ( ACTA2) forward 5′-TCTTCGCAATGTTTGACCAGT-3′, reverse 5′-GTTGAAAGCCTCCTTAAACTCCT-3′; and glyceraldehyde-3-phosphatedehydrogenase ( GAPDH) forward 5′-GTCAAGGCTGAGAACGGGAA-3′, reverse 5′-AAATGAGCCCCAGCCTTCTC-3′.

Western blot analysis

A KeyGEN (Nanjing, China) protein extraction kit was used to extract proteins from stably transfected cells, in strict accordance with the protocol provided. Protein concentrations were quantified with a Pierce bicinchoninic acid (BCA) kit (Pierce, Rockford, USA). The samples were separated by sodium dodecyl sulfate (SDS)-polyacrylamide gel electrophoresis and then transferred to a polyvinylidene fluoride (PVDF) membrane (Millipore, Billerica, USA). The membrane was subsequently blocked for 2 h using a 5% BSA solution in Tris-buffered saline with Tween (TBST) buffer and incubated with the primary antibody (1:1000; 23081-1-AP; Proteintech, Wuhan, China) at 4°C overnight. After primary antibody incubation, the membrane was washed three times with TBST buffer for 10 min each. The membrane was incubated with a horseradish peroxidase (HRP)-conjugated secondary antibody at room temperature for 2 h and then washed three times with TBST. The protein bands were then detected via an enhanced chemiluminescence HRP substrate (WBKL0100; Millipore) and a chemiluminescence imaging system.

Clonogenic assay

For the clonogenic assay, GC cells transfected with LV-miR-149-5p, LV-miR-149-5p inhibitor or negative control lentivirus were plated into 6-well plates at a density of 500 cells per well and cultured for two weeks. The cells in each well were subsequently fixed with 2 mL of methanol for 30 min and stained with crystal violet for 20 min to facilitate colony visualization and counting.

Immunohistochemical (IHC) analysis

GC tissues were preserved in 10% formalin and subsequently embedded in paraffin. The sections prepared from these blocks were processed with targeted primary antibodies. Following an overnight incubation at 4°C, the sections were gently washed twice and then exposed to an HRP-polymer-conjugated secondary antibody (Abcam, Cambridge, UK) at room temperature. After secondary antibody treatment, the samples were treated with a 3,3′-diaminobenzidine solution and counterstained with hematoxylin to highlight the cellular structures. Finally, the prepared slides were examined under a light microscope for histological analysis.

5-Ethynyl-2’-deoxyuridine (EdU) assay

To evaluate DNA synthesis and cell proliferation, we used an EdU analysis kit (RiboBio, Guangzhou, China). An initial population of 100,000 treated GC cells was cultured overnight in a 24-well plate. Subsequently, 10 μM EdU solution was added to each well, and the cells were incubated for another 24 h. The cells were fixed with 4% formalin at room temperature for 2 h and then permeabilized with 0.5% Triton X-100 for 10 min. Next, a 200 μL Apollo reaction cocktail was added to label the incorporated EdU for 30 min, while Hoechst 33342 (200 μL) was applied to stain the cell nuclei. Finally, a Nikon microscope (Nikon, Tokyo, Japan) was used to observe the red fluorescent EdU signals and blue fluorescent Hoechst signals to evaluate DNA replication and overall cell proliferation, respectively.

Flow cytometric analysis of cell apoptosis

Cell apoptosis was assessed via an Annexin V‐Alexa Fluor 647 Apoptosis Detection Kit according to the protocol provided by the US Everbright (Suzhou, China). An initial seeding of 6 × 10 ⁵ cells was distributed into 6-well plates and cultured for 24 h. Post-culture, the cells were harvested, resuspended in an appropriate buffer, and stained with 50 μg/mL Annexin V‐Alexa Fluor 647 and 10 μg/mL propidium iodide in the dark for 15 min. The stained cells were subsequently analyzed via a BD LSR flow cytometer (BD Biosciences, San Diego, USA) to quantify apoptotic populations.

Haematoxylin and eosin staining of tissue

To prepare the tissue samples which were initially fixed in alcohol for microscopic examination, a systematic rehydration process was employed. First, the slides containing the tissue samples were immersed in deionized water and gently shaken for 30 s to rehydrate the tissues. Next, the slides were placed into a container with hematoxylin and stirred for another 30 s, followed by another 30 s of rinsing in deionized water to remove excess stains. Following hematoxylin staining, the slides were immersed in 1% eosin Y solution. The samples were sequentially soaked in 95% alcohol and then in 100% alcohol. Then, xylene was applied to remove the alcohol. Finally, the slides were mounted with coverslips and histologically examined under a microscope.

Statistical analysis

The statistical analysis in this study was conducted via R software (version 4.2.2). Correlation coefficients were calculated via both Pearson and Spearman correlation analyses. The data of two independent groups were compared via either Student’s t test or the Mann-Whitney U test, following established statistical conventions as appropriate. We analyzed differential mutations and copy number loss and gain of genes via the chi-square test and Fisher’s exact test, plotted overall survival (OS) via Kaplan-Meier methods, and evaluated significant differences via log-rank tests. Unless otherwise indicated, statistical significance was defined as P < 0.05.

Results

Network analysis of O-GlcNAcylation-associated genes and construction of O-GlcNAcylation-associated clusters

The expression data of 133 of the 151 O-GlcNAcylation-associated genes initially downloaded were present in the merged dataset. Using these genes, a network was constructed to investigate their interactions and assess their prognostic impacts on patients with GC harboring these genes ( Figure 1A). The results of the univariate regression analysis are provided in Supplementary Table S1. This analysis identified 48 genes with significant prognostic implications ( Supplementary Table S2), many of which were strongly associated with each other. We discovered that 17 of these genes had a positive effect on GC prognosis, whereas the remaining 31 genes were recognized as risk factors. Furthermore, to comprehensively evaluate their effectiveness in prognosis prediction, Kaplan-Meier survival analysis was performed on the basis of the grouping of O-GlcNAcylation-related gene expression levels. The results showed that the expression of 41 O-GlcNAcylation-related genes was correlated with the OS of GC patients, with a significant level of P = 0.001 or P < 0.001. Among them, low expression level of ADAMTS1, ADAMTS3, ADAMTS6, ADAMTS8, ADAMTSL1, ADAMTSL3, B3GTN9, GALNT11, GALNT10, GALNT15, GALNT16, GXYLT2, MUC15, SBSPON, PLOD2, SPON1, ST3GAL4, ST3GAL1, ST6GALNAC3, TET1, THBS2, THBS1, THSD7B, VEGFB, THSD7A and TMTC1 was correlated with high OS of GC patients, while high expression level of ADAMTSL5, B3GNT3, C1GALT1, DPM2, FKTN, FUT3, FUT6, GALNT3, GALNT8, MUC13, POMGNT1, SLC35C1, POMK, ST6GALNAC4 and TET2, was correlated with high OS of GC patients ( Supplementary Figure S1 ).

Figure 1 .

Interactions of O-GlcNAcylation-related genes and their prognostic value and identification of O-GlcNAcylation-associated clusters

(A) Interactions of O-GlcNAcylation regulators in STAD. The circle diameter reflects the significance level of the P values obtained from the log-rank test. Risk factors are indicated in purple, whereas favorable factors for overall survival are shown in green. The connecting lines depict relationships between O-GlcNAcylation regulators determined by Spearman correlation analysis, with pink representing a positive correlation and blue representing a negative correlation. Only the results of univariate regression with a significance level less than 0.05 and genes with a correlation P value less than 0.0001 between them are displayed. (B) The heatmap displays the consensus matrix obtained through consensus clustering with k = 2. 1 represents pattern (B) and 2 represents pattern (A). (C) Survival between the two O-GlcNAcylation-related patterns is depicted by Kaplan-Meier curves. (D) Differential expression of O-GlcNAcylation-related genes in the two clusters. (E) Heatmap of genetic modification patterns. *P < 0.05, ** P < 0.01, ***P < 0.001.

On the basis of the results of the univariate regression analysis, a total of 34 genes were identified to have a significant level of P​<0.01 for clustering. To explore the expression patterns of these 34 O-GlcNAcylation-related genes in GC, we employed unsupervised clustering techniques to categorize all GC patients into two distinct patterns ( Figure 1B): pattern A, with 486 cases, and pattern B, with 189 cases. Further analysis revealed that patients in Cluster A had a more favorable prognosis than those in Cluster B ( Figure 1C). We also examined the differences in the expression levels of the 34 selected O-GlcNAcylation genes between Clusters A and B and generated a heatmap of gene expression data containing comprehensive clinicopathological features of the samples for these 34 genes ( Figure 1D,E). Almost all O-GlcNAcylation-associated genes were differentially expressed between Cluster A and Cluster B, and most of them were upregulated in Cluster B, indicating that there was a relatively active O-GlcNAcylation modification in Cluster B.

To delve deeper into the functional attributes, we performed GSVA enrichment analysis using five extensive pathway databases, namely, BioCarta, Reactome, Hallmark, Wiki Pathways, and KEGG, with the aim of identifying differential signaling pathways in the two clusters. As depicted in Supplementary Figure S2, Cluster A was enriched predominantly in pathways related to DNA repair and base excision repair. In contrast, Cluster B was implicated in a wide range of biological pathways for tumorigenesis and progression, including Hedgehog signaling, transforming growth factor (TGF)-beta signaling, phosphatidylinositol 3-kinase (PI3K)-protein kinase B (AKT) signaling, and mitogen-activated protein kinase (MAPK) signaling, among others. This finding was consistent with our preceding research, where Cluster A had a more promising survival rate than did Cluster B.

TME characteristics among O-GlcNAcylation-associated subgroups and functional enrichment analysis of DEGs

Principal component analysis (PCA) revealed distinct patterns, as shown in Figure 2A. The ESTIMATE algorithm was employed to evaluate the differences in the tumor microenvironment between Cluster A and Cluster B ( Figure 2B). The results indicated that patients in Cluster B had increased stromal scores, immune scores, and ESTIMATE scores, indicating that the immune activity of individuals in Cluster B was increased and that the tumor purity was decreased. Furthermore, we utilized the ssGSEA function from the R package GSVA to calculate immune cell infiltration scores and compare the differences in immune cell infiltration among the subtypes. As demonstrated in Figure 2C, the infiltration levels of activated CD4 T cells, CD56dim natural killer cells, neutrophils, and type 17 T helper cells were greater in Cluster A than in Cluster B. Conversely, activated B cells, activated CD8 cells, eosinophils, immature B cells, immature dendritic cells, myeloid-derived suppressor cells (MDSCs), macrophages, mast cells, natural killer T cells, natural killer cells, plasmacytoid dendritic cells, regulatory T cells, T follicular helper cells, and type 1 T helper cells exhibited significantly lower infiltration in Cluster A than in Cluster B. The above findings revealed differences in O-GlcNAcylation levels among different GC subtypes and demonstrated the significant influence of O-GlcNAcylation-related genes on cell infiltration within the tumor microenvironment (TME).

Figure 2 .

Differences in the tumor microenvironment between subgroups and functional enrichment analysis

(A) PCA revealed different distributions between the two subgroups. (B) Stromal score, immune score and ESTIMATE score analyses between the two subgroups. (C) Abundances of 23 infiltrating immune cell types in the two different subgroups. (D) Volcano plot showing differentially expressed genes in the two clusters. (E,F) GO and KEGG enrichment analyses of DEGs between the two subgroups. *P < 0.05, **P < 0.01, ***P < 0.001.

To further investigate the potential biological functions of each O-GlcNAcylation pattern, we utilized the limma package to identify 230 differentially expressed genes (DEGs) associated with the O-GlcNAcylation phenotype. A volcano plot was generated ( Figure 2D). The DEGs were subjected to GO and KEGG enrichment analysis using the clusterProfiler package. The GO enrichment analysis of O-GlcNAcylation cluster-related DEGs revealed their primary association with biological processes involving the extracellular matrix ( Figure 2E). Additionally, the KEGG analysis demonstrated a significant association of the DEGs with metabolic and tumor-related signaling pathways, as depicted in Figure 2F. These findings further highlighted the significant role of O-GlcNAcylation-related genes in tumor initiation and progression, which was consistent with our previous GSVA results.

Identification of O-GlcNAcylation-related phenotypes and O-GlcNAcylation scores

The 230 DEGs associated with O-GlcNAcylation were subsequently subjected to univariate Cox regression analysis to identify genes associated with overall survival (OS) in patients with GC ( Supplementary Table S3). Thirteen genes were significantly correlated with the prognosis of GC patients ( P < 0.00000001) ( Figure 3A). To explore the potential mechanisms involving these prognosis-related DEGs in GC, unsupervised consensus clustering analysis was performed on the basis of the expression levels of these 13 genes. This analysis successfully classified the GC patients into two distinct gene clusters, namely, gene cluster A and gene cluster B, which comprised 524 and 151 patients, respectively ( Figure 3B). Survival analysis revealed that the survival rate of patients in gene cluster A was greater than that of patients in gene cluster B ( Figure 3C). Notably, significant variations in the expression of O-GlcNAcylation regulators were observed in these genetic subtypes, which is consistent with the anticipated patterns of O-GlcNAcylation. Additionally, there was a significant upregulation of O-GlcNAcylation-related DEGs in gene cluster B ( Figure 3 D,E).

Figure 3 .

Construction of gene clusters and O-GlcNAcylation scores

(A) Genes with significant prognostic correlations identified through univariate Cox regression analysis. (B) The consensus matrix for k = 2 generated by means of consensus clustering via the heatmap. 1 represents gene cluster B and 2 represents gene cluster A. (C) KM curves of survival analysis for two gene clusters. (D) Heatmap of clinicopathologic characteristics and gene expression among the two groups. (E) Differential expression of 13 genes in the two clusters. (F) KM curves of survival analysis for the high- and low-score groups. (G) Sankey diagram of clusters, gene clusters, O-GlcNAcylation scores, and survival statuses. (H) Relationships between immune cells and O-GlcNAcylation scores, with red denoting a positive correlation and blue denoting a negative correlation. *P < 0.05, **P < 0.01, ***P < 0.001.

Previous analyses were based only on population-level data and may not accurately predict individual O-GlcNAcylation modification patterns. To account for the individual heterogeneity of O-GlcNAcylation modifications, we developed a scoring system called the O-GlcNAcylation score, which quantifies the modification patterns in GC patients via principal component analysis (PCA) algorithms. By utilizing this score, we categorized all the GC patients into two groups: the high group and the low group. Survival analysis revealed that patients with lower O-GlcNAcylation scores had significant survival benefits ( Figure 3F). Mulberry plots were generated to illustrate the associations between these subtypes ( Figure 3G). Correlation analysis with immune cell infiltration revealed that the score was positively associated with various immune cell subsets, including activated B cells, activated CD8+ T cells, and natural killer cells. Notably, the score was also positively correlated with certain immunosuppressive cells, such as MDSCs, macrophages, and regulatory T cells ( Figure 3H). This finding may explain the lower survival rate previously reported in the high-score group.

Prognostic value of the O-GlcNAcylation score in GC patients

We aimed to investigate the relationships between the O-GlcNAcylation score and patient prognosis or clinical characteristics. The results revealed that patients who experienced mortality, stage III-IV disease, or disease recurrence scored higher. Additionally, patients with higher scores also had higher rates of death, late-stage disease and recurrence ( Figure 4A–C).

Figure 4 .

Assessment of the clinical prognosis of O-GlcNAcylation scores

(A) Comparison of the O-GlcNAcylation scores among individuals in the alive and dead groups and the proportions of living and dead individuals in the high- and low-O-GlcNAcylation score groups. (B) Comparison of the O-GlcNAcylation scores among individuals in the Stage I–II and Stage III–IV groups and the proportions of Stage I–II and Stage III–IV individuals in the high- and low-O-GlcNAcylation score groups. (C) Comparison of the O-GlcNAcylation scores among individuals in the recurrence and no recurrence groups and the proportions of individuals in the recurrence and no recurrence groups in the high- and low-O-GlcNAcylation score groups. (D) Heatmap illustrating the relationships among scores, chemokines, interleukins, interferons, their receptors, and other cytokines. (E) Correlations between scores and 50 hallmark pathways. *P < 0.05, **P < 0.01, ***P < 0.001.

In the tumor microenvironment (TME), cytokines play a critical role in facilitating interactions between immune and nonimmune cells. Therefore, this study aimed to reveal the correlations between the scores and various components, including chemokines, interleukins, interferons, their receptors, and other cytokines ( Figure 4D). The results demonstrated heterogeneity in the relationships between scores and different chemokines, interleukins, interferons, and their receptors. Notably, high scores were associated with the upregulation of chemoresistance-related and procancer factors, such as C-C chemokine ligand 5 (CCL5), C-C chemokine receptor type 5 (CCR5), C-C chemokine receptor type 2 (CCR2), interleukin 33 (IL33), interleukin-4 receptor (IL4R), interleukin 6 (IL6), interleukin 17B (IL17B), and interleukin 34 (IL34), while causing the downregulation of certain antitumor factors, such as interleukin 18 (IL18). Additionally, we conducted gene set variation analysis (GSVA) to explore the correlation between the scores and hallmark pathway activity. The GSVA of the hallmark pathways revealed enrichment of various tumor-related signaling pathways, including epithelial mesenchymal transition, hedgehog signaling, Kirsten rat sarcoma viral oncogene homolog (KRAS) signaling, and TGF-beta signaling, in the high-score group ( Figure 4E). These findings suggest that patients in the high-score group may have a higher incidence of tumor occurrence and metastasis.

Genetic variation landscape of O-GlcNAcylation-related DEGs in GC

We conducted genetic variant analysis of the 12 screened O-GlcNAcylation-related DEGs in stomach adenocarcinomas (STADs), including copy number variations (CNVs) and single-nucleotide variants (SNVs), via the GSCA database. The CNV pie plots in Figure 5A and B revealed that almost all 12 genes screened had both heterozygous and homozygous amplification and deletion in STAD. Among these genes, myosin light chain 9 ( MYL9) had the highest percentage of CNV (63.95%), followed by apolipoprotein D ( APOD, 34.24%), protein phosphatase 1 regulatory inhibitor subunit 14A ( PPP1R14A, 22.68%), matrix-remodeling associated 7 ( MXRA7, 21.09%), A-kinase anchoring protein 12 ( AKAP12, 17.46%), ACTA2 (11.56%), cofilin-2 ( CFL2, 11.56%), hes related family bHLH transcription factor with YRPW motif like ( HEYL, 11.56%), RNA binding protein, mRNA processing factor 2 ( RBPMS2, 9.98%), FERM domain containing kindlin 2 ( FERMT2, 9.52%), slit guidance ligand 2 ( SLIT2, 6.35%), and insulin like growth factor binding protein 7 ( IGFBP7, 6.12%). Correlation analysis between CNV and mRNA expression revealed that the mRNA expression of ACTA2 and MXRA7 in GC was positively correlated with their corresponding CNVs ( Figure 5C). Furthermore, the epigenetic methylation levels of the selected O-GlcNAcylation-related DEGs were examined through methylation analysis. This examination revealed that the 12 selected O-GlcNAcylation-related DEGs in STAD exhibited hypomethylation, as depicted in Figure 5D. We also explored the SNV percentage of the 12 examined O-GlcNAcylation-related DEGs shown in Figure 5E, among which SLIT2 presented the highest SNV percentage, followed by AKAP12, FERMT2 , ACTA2 , APOD, IGFBP7, PPP1R14A, RBPMS2, MYL9, HEYL and CFL2 .

Figure 5 .

Genetic variation landscape of differentially expressed genes (DEGs) related to O-GlcNAcylation

(A) CNV percentage in GC. (B) The percentage of heterozygous and homozygous copy number variations (CNVs), as well as the amplification and deletion percentages of both heterozygous and homozygous CNVs for each gene in GC. (C,D) Relationships of CNV and methylation levels with mRNA expression levels. The size of the circles represents the P value, with red indicating a positive correlation and blue indicating a negative correlation. (E) Heatmap demonstrating the SNV frequencies of 12 genes in GC. The number of samples with the corresponding mutated genes is indicated by numbers.

Considering the pivotal role of the tumor mutational burden (TMB) in guiding immunotherapy strategies for patients with STAD, the intrinsic relationship between the TMB and O-GlcNAcylation score was investigated, taking into account the clinical significance of the TMB. We analyzed somatic mutations in the high- and low-score groups. Interestingly, patients in the low-score group presented significantly greater frequencies of somatic mutations than did those in the high-score group, particularly in titin (TTN, 51% vs 16%), TP53 (36% vs 19%), mucin 16, cell surface-associated (MUC16, 31% vs 14%), and low-density lipoprotein receptor-associated protein 1B (LRP1B, 27% vs 10%) ( Figure 6A,B). The analysis of gene mutation differences between the high- and low-score groups revealed the following: TTN, spectrin repeat containing nuclear envelope protein 1 (SYNE1), obscurin, cytoskeletal calmodulin and titin-interacting RhoGEF (OBSCN), FAT atypical cadherin 2 (FAT2), ring finger protein 213 (RNF213), ryanodine receptor 1 (RYR1), CUB and sushi multiple domains 1 (CSMD1), dynein axonemal heavy chain 5 (DNAH5), ankyrin 3 (ANK3), reelin (RELN), and RNA, ribosomal 45S cluster 3 (RNR3) had more mutations in the low-score groups, whereas cadherin 1 (CDH1) had more mutations in the high-score groups ( Figure 6C). Additionally, we explored the relationships between the stemness indices and the scores. The results revealed a negative correlation between the stemness indices (mDNAsi and mRNAsi) and the scores, with the low-score group demonstrating significantly greater stemness index values than the high-score group ( Figure 6D,E). These findings suggest that O-GlcNAcylation-related genes are likely to impact tumor development through genetic and epigenetic modifications.

Figure 6 .

Somatic mutation analysis and the relationship between stemness indices (mRNAsi and mDNAsi) and O-GlcNAcylation scores

(A,B) Waterfall plot showing the top 20 genes with mutation frequencies in the (A) high- and (B) low-score groups. Each column represents an individual patient. The upper histogram shows the total TMB, and the numbers on the right represent the mutation frequencies of each gene. The bar graph on the right is the proportion of each mutation type. (C) Illustration of the top 12 genes with significant differences in mutation rates between the high and low groups, according to ascending order of P values. (D) Relationships between mDNAsi and O-GlcNAcylation scores, differences in mDNAsi between the high and low groups, and the distribution of mDNAsi based on high vs low groups and patient survival status. (E) Relationships between mRNAsi and O-GlcNAcylation scores, differences in mRNAsi between the high and low groups, and the distribution of mRNAsi based on high vs low groups and patient survival status. *P < 0.05, **P < 0.01, ***P < 0.001.

Prediction of immunotherapy efficacy and antitumor drug sensitivity in high- and low-score groups

These studies demonstrated that the expression levels of O-GlcNAcylation-related molecules may play a significant role in defining tumor characteristics and promoting innovative therapeutic strategies. To confirm the accuracy of our risk score in predicting the efficiency of immunotherapy, we utilized multiple independent immunotherapy cohorts from the literature to validate immunotherapy efficacy and patient prognosis. The following cases were used: late-stage nonsquamous non-small cell lung cancer treated with bevacizumab and erlotinib ( Figure 7A), melanoma treated with an anti-PD-1 antibody ( Figure 7B), advanced non-small cell lung cancer treated with an anti-PD-1/PD-L1 antibody ( Figure 7C), advanced urothelial cancer treated with atezolizumab or an anti-PD-L1 antibody ( Figure 7D), and clear cell renal cell carcinoma treated with nivolumab (anti-PD-1) versus the mechanistic target of rapamycin (mTOR) inhibitor everolimus ( Figure 7E). This validation confirmed the predictive capacity of the score with respect to therapy prognosis and efficacy. Notably, patients in the high-score group had a significant survival benefit compared with those in the high-risk group ( Figure 7A–E). Presumably, patients with high scores were more likely to benefit from immunotherapy ( Figure 7A–E).

Figure 7 .

Prediction of immune therapy prognosis in the high- and low-score groups

(A) Kaplan-Meier curve analysis of the high- and low-score groups of patients receiving immunotherapy and the proportions of responders and non-responders in the high- and low-score groups. (B,D,E) Proportions of patients with PD/SD and PR/CR in each group. (C) Proportions of patients with progression and without progression.

To further determine the potential clinical implications of our O-GlcNAcylation score, we analyzed the discrepancies in the half maximal inhibitory concentration (IC ₅₀) values of antitumor drugs in the high-score and low-score cohorts ( Supplementary Figure S3). The results indicated that, compared with high-score patients, low-score GC patients were more sensitive to A.443654, ABT.888, AICAR, the AKT inhibitor VIII, AZD6244, and BI.2536. Conversely, we observed that GC patients in the low-score group were less sensitive to A.770041, ABT.263, AG.014699, AMG.706, AP.24534, and AS601245 than those in the high-score group.

Fundamentally, our research underscores a definitive and significant correlation between the pattern of O-GlcNAcylation modification and the response to immunotherapy in GC patients. Additionally, the O-GlcNAcylation profile established in this study has potential for predicting the responsiveness of these patients to immunotherapy and various anticancer drugs.

The expression and functional phenotype of ACTA2 in GC cells

Among the 12 O-GlcNAcylation-related DEGs, most of them have been verified to be involved in the occurrence and development of GC [ 14- 23]. However, the expression and role of ACTA2 in GC are not clear. We explored this further experimentally. Normal gastric tissues and GC tissues were stained with haematoxylin and eosin. Compared with normal gastric tissues, GC tissues showed obvious necrosis ( Figure 8A left). In addition, we conducted immunohistochemistry on both normal gastric tissues and GC samples. As shown in Figure 8A right, the expression level of ACTA2 in the GC samples was significantly greater than that in the normal gastric mucosa samples. Figure 8B further shows that ACTA2 expression in the GC cell lines was greater than that in the GES-1 cells. Since the expression level of ACTA2 in both HGC-27 and AGS cells was significantly greater than that in other GC cells, we selected HGC-27 and AGS cells for subsequent experiments. To investigate the biological function of ACTA2 in GC, two small interfering RNA (siRNA) molecules targeting ACTA2 were transfected into HGC-27 and AGS cells to downregulate ACTA2 expression. Moreover, a plasmid overexpressing ACTA2 was transfected into HGC-27 and AGS cells to increase the expression of ACTA2. Through qRT‒PCR and western blot analyses, we found that ACTA2 expression was significantly elevated in HGC-27 and AGS cells after ACTA2-OE transfection, and ACTA2 expression was significantly reduced in HGC-27 and AGS cells after si-ACTA2-1/2 transfection ( Figure 8C and Supplementary Figure S4A). Among the two siRNAs, si-ACTA2-2 had a more pronounced inhibitory effect on ACTA2 expression in HGC-27 and AGS cells; thus, we selected si-ACTA2-2 for subsequent loss-of-function assays. Through colony formation and EdU experiments, we observed that overexpression of ACTA2 promoted cell proliferation, whereas the downregulation of ACTA2 significantly inhibited cell proliferation ( Figure 8D,E and Supplementary Figure S4B). Additionally, flow cytometry analysis revealed that ACTA2 overexpression markedly decreased the apoptosis rate of HGC-27 and AGS cells, whereas ACTA2 knockdown increased the apoptosis rate of HGC-27 and AGS cells ( Figure 8F and Supplementary Figure S4C). Through these experiments, we conclude that ACTA2 enhances the proliferation and inhibits the apoptosis of GC cells.

Figure 8 .

Relative mRNA level and biological function of ACTA2

(A) Left: Haematoxylin and eosin staining of GC tissues and paracarcinoma tissues; scale bar: 100 μm. Right: IHC staining for ACTA2 in GC tissues and paracarcinoma tissues; scale bar: 100 μm. N represents paracarcinoma tissues, and T represents GC tissues. (B) Comparison of the relative mRNA and protein expression levels of ACTA2 between normal gastric mucosal epithelial cells and GC cells. (C) qRT-PCR and western blot analysis confirmed the ACTA2 expression level in HGC-27 cells. (D) Colony formation assay. (E) EdU assay. Scale bar: 100 μm. (F) Flow cytometry for apoptosis detection. *P < 0.05, **P < 0.01, ***P < 0.001.

Discussion

Current studies have consistently demonstrated the presence of hyper-O-GlcNAcylation in various types of cancer [24]. Additionally, O-GlcNAcylation plays a crucial role in the progression, metastasis, and invasion of numerous tumors, including cervical cancer [25], bladder cancer [26], hepatocellular carcinoma [ 27, 28], colorectal cancer [29] and GC [30]. Furthermore, O-GlcNAcylation significantly influences early screening, prognosis, chemoresistance [31], and immune evasion of tumors [8]. On the basis of these findings, the objective of this study was to investigate the substantial impacts of O-GlcNAcylation on the tumor microenvironment (TME), prognosis, chemotherapy, and immune therapy sensitivity, specifically in patients with GC.

In this study, univariate Cox regression analysis was used to identify 34 O-GlcNAcylation-related genes closely associated with GC prognosis. On the basis of their expression levels, all GC patients were divided into two distinct clusters, with significant differences in prognosis, enriched pathways, and the tumor microenvironment. Notably, Cluster A presented a significantly greater survival rate than Cluster B did. Analysis of differences in gene expression levels revealed that the majority of these genes presented higher expression in Cluster B than in Cluster A did, suggesting relatively more active O-GlcNAcylation modifications in this cluster. Additionally, the analysis of immune cell infiltration revealed a greater level of infiltration in Cluster B, particularly in macrophages. These findings are consistent with those of previous studies, suggesting that O-GlcNAcylation promotes the activity of various immune cells [32] and that O-GlcNAc transferase facilitates tumor metastasis and chemoresistance by increasing tissue protein B levels in the TME, especially in M2 macrophages [31]. GSVA revealed that Cluster B was enriched in various tumor-related pathways, including hedgehog signaling, TGF-beta signaling, PI3K-AKT signaling, and MAPK signaling. Previous studies have demonstrated that O-GlcNAcylation modifies the transcription factors of glioma-associated oncogene homolog 1 (GLI1) and glioma-associated oncogene homolog 2 (GLI2) in the hedgehog pathway, thereby increasing their activity and causing drug resistance in breast cancer [33]. Furthermore, O-GlcNAcylated MORC family CW-type zinc finger 2 (MORC2) has been found to activate the TGF-β1 target genes connective tissue growth factor ( CTGF) and snail family transcriptional repressor 1 ( SNAIL), thereby affecting the TGF-β signaling pathway. This pathway plays a partial role in breast cancer progression by regulating genes associated with tumor metastasis [34]. In anaplastic thyroid cancer cells, O-GlcNAcylation has been shown to enhance invasion partly through the PI3K/Akt signaling pathway [35]. Moreover, elevated O-GlcNAcylation in the tumor microenvironment (TME) has been shown to reduce the production of inflammatory cytokines. This reduction occurs through the downregulation of p38 MAPK activity and the subsequent upregulation of the extracellular signal-regulated kinase 1 and 2 (ERK1/2) signaling pathway, ultimately promoting cancer progression [36]. Hence, we hypothesize that O-GlcNAcylation plays a crucial role in GC progression through these pathways. Further research is needed to elucidate the mechanisms underlying GC development and explore novel therapeutic approaches.

The utilization of O-GlcNAcylation-related genes for GC subtyping has great promise and clinical significance. To assess the clinical relevance of this subtyping approach, we meticulously identified 13 genes that exhibited differential expression patterns in two clusters that were highly correlated with GC prognosis. By using these selected genes for unsupervised clustering, we succeeded in separating GC patients into two distinct gene clusters with widely varying survival outcomes. Using the expression levels of these 13 genes, we devised an O-GlcNAcylation score that could serve not only as a valuable tool to guide immunotherapy but also to mitigate the influence of individual heterogeneity. Our study is the first to clarify the associations between the scoring system and immune-related cytokines, such as chemokines, interleukins, interferons, and their receptors. Our analysis revealed elevated expression levels of C-X-C chemokine ligand 12 (CXCL12), C-X-C chemokine receptor type 4 (CXCR4), IL33, IL6, and TGF-β in the high-score group. Previous investigations have demonstrated that the CXCL12-CXCR4 axis contributes to ICI resistance by blocking T-cell infiltration into tumors [37]. Additionally, in an otherwise ICI-resistant subclone of the B16F10 melanoma model, the administration of a specific IL-33-blocking antibody has been shown to restore the effectiveness of anti-PD1 therapy [38]. Furthermore, inhibiting IL-6 may enhance the efficacy of immune checkpoint blockade [39]. Notably, high levels of TGF-β are often correlated with a reduced response to PD-1/PD-L1 therapy [40]. Collectively, these findings suggest that patients in the low-score group may benefit more from immunotherapy. TMB has been recognized as a promising biomarker and an independent predictor of ICI response across different cancer types [41]. Studies have demonstrated a positive association between TMB and the efficacy of ICI treatment, especially in patients with GC [42]. In view of the close relationship between TMB and the somatic mutation rate, further analysis revealed that the mutation rate of the low-score group was greater than that of the high-score group. These findings may provide valuable insights into predicting the effectiveness of immune therapy in GC patients. To further validate the predictive effect of O-GlcNAcylation scoring on immunotherapy efficacy, we incorporated data from five different datasets of patients who received or did not receive immunotherapy. However, the results obtained in other tumor types deviated from our previous speculation about its predictive role in GC immunotherapy. This difference highlights the diverse factors influencing the effectiveness of immunotherapy in cancer patients. These factors include tumor immunogenicity, T cell exclusion [43], rearrangements of the anaplastic lymphoma kinase (ALK) [44], immunosuppressive factors present in the tumor microenvironment [45], impaired processing or presentation of neoantigens [46], dysbiosis of the gut microbiome [47], and, of course, tumor heterogeneity. Despite these complexities, the scoring system we established clearly has clinical value in predicting immunotherapy efficacy in patients with tumors. Finally, utilizing the scoring system we established, we predicted that patients with higher scores would be more likely to respond to anti-tumor drugs. In general, the scoring system we established has significant implications for improving the selection of chemotherapy agents and predicting the effectiveness of immunotherapy in GC patients.

Numerous studies have indicated that O-GlcNAcylation-related hub genes play a significant role in the occurrence and development of various tumors. In particular, the interaction between phosphoglycerate mutase 1 (PGAM1) and ACTA2 promotes cancer cell migration [48]. Additionally, high expression levels of ACTA2 are closely associated with poor prognosis and a lack of response to ICIs in GC patients [49]. In this study, immunohistochemistry confirmed that ACTA2 was overexpressed in GC tissues, and qRT-PCR and western blot analysis confirmed that ACTA2 was upregulated in GC cell lines. Functional assays, including the clonogenic assay, the EdU assay, and flow cytometric analysis, demonstrated that ACTA2 enhanced the proliferation and inhibited the apoptosis of GC cells, which was consistent with our previous results. Hence, ACTA2 is expected to serve as a biomarker for the assessment of the prognosis and efficacy of GC immunotherapy. Whether ACTA2 affects the progression of GC through O-GlcNAcylation needs to be further studied.

Nevertheless, our research has several limitations. This study is a retrospective study utilizing bioinformatics methods. To validate the reliability of our model and address potential biases associated with retrospective study design, incorporating prospective real-world data is essential. Furthermore, while we have established the significance of the scoring system in predicting the efficacy of immunotherapy in cancer patients, more data are needed to elucidate its specific relevance in GC patients receiving immunotherapy.

In conclusion, we identified 34 key genes linked to O-GlcNAcylation. Utilizing the expression profiles of these genes, we successfully divided GC patients into two clusters with significant differences in prognosis, tumor microenvironment complexity, and tumor mutational load. Moreover, a prognostic scoring system developed on the basis of the expression patterns of 12 differential O-GlcNAcylation-related genes demonstrated efficacy in predicting patient prognosis and responsiveness to immunotherapy, demonstrating its potential for clinical application.

Supplementary Data

Supplementary data is available at Acta Biochimica et Biophysica Sinica online.

COMPETING INTERESTS

The authors declare that they have no conflict of interest.

Funding Statement

This work was supported by the grants from the National Natural Science Foundation of China (No. 81871942), the Cancer Fundamental Project from Bethune Charitable Foundation (BCF-NH-ZL-20201119-004), the Joint Open Research Fund of Suzhou Institute of Nanotechnology and Nano-bionics & Jiangsu Province Hospital, and the Medical Engineering Translational Fund of Jiangsu Province Hospital (No. NM202404).