Establishment of Golgi apparatus-related genes signature to predict the prognosis and immunotherapy response in gastric cancer patients

Abstract

The Golgi apparatus plays a crucial role in intracellular protein transportation, processing, and sorting. Dysfunctions of the Golgi apparatus have been implicated in tumorigenesis and drug resistance. This study aimed to investigate the prognostic and treatment response assessment value of Golgi apparatus-related gene (GARGs) features in gastric cancer patients. Transcriptome data and clinical information of gastric cancer patients were obtained from The Cancer Genome Atlas and Gene Expression Omnibus databases. Cox regression analysis was employed to assess the prognostic significance of GARGs and construct risk features. The immune landscape, drug sensitivity, immune therapy response, gene expression patterns, and somatic mutation characteristics were analyzed between different risk groups. A nomogram model for predicting gastric cancer prognosis was developed and evaluated. Among 1643 GARGs examined, 365 showed significant associations with gastric cancer prognosis. Five independent prognostic GARGs (NGF, ABCG1, CHAC1, GBA2, PCSK7) were selected to construct risk features for gastric cancer patients. These risk features effectively stratified patients into high-risk and low-risk groups, with the former exhibiting worse prognosis than the latter. Patients in the high-risk group displayed higher levels of immune cell infiltration, while the expression levels of NGF, CHAC1, GBA2, PCSK7 were significantly correlated with immune cell infiltration. Notably, the low-risk group exhibited higher sensitivity to epothilone.B, metformin, and tipifarnib compared to the high-risk group. Moreover, patients in the low-risk group demonstrated greater responsiveness to immune therapy than those in the high-risk group. In terms of biological processes and KEGG pathways related to immunity regulation, significant suppression was observed in the high-risk group compared to the low-risk group; meanwhile cell cycle pathways exhibited significant activation in the high-risk group. Furthermore, the low-risk group exhibited a higher tumor mutation burden compared to the high-risk group. The risk features derived from GARGs, in conjunction with age, were identified as independent risk factors for gastric cancer. The nomogram incorporating these factors demonstrated improved performance in predicting gastric cancer prognosis. Our study established risk features derived from GARGs that hold potential clinical utility in prognostic assessment and immune therapy response evaluation of gastric cancer patients.

1. Introduction

Gastric cancer, a prevalent malignant tumor worldwide and a leading cause of cancer-related mortality, poses a significant public health burden. In 2020 alone, the global incidence of gastric cancer exceeded 1 million cases, with over 750,000 deaths reported.^[1] Stomach adenocarcinoma (STAD) represents approximately 95% of all gastric cancer cases. Despite notable advancements in treatment modalities, the overall 5-year survival rate for patients at an advanced stage remains dishearteningly low, hovering around 20%.^[2] Even with radical resection and perioperative chemotherapy for locally advanced patients, the 5-year overall survival rate remains below 40%.^[3–5] Accurate prognostic assessment plays a pivotal role in tailoring individualized treatment strategies for patients with gastric cancer. The current Lauren/World Health Organization classification and tumor-node-metastasis staging system constitute essential tools for selecting appropriate therapeutic interventions.^[6] Nevertheless, existing prognostic models predominantly rely on conventional clinical and pathological features while neglecting crucial aspects such as tumor molecular characteristics and individual genetic variations. Consequently, there is an urgent need to develop an accurate, reliable, and predictive prognostic risk model specific to gastric cancer to inform treatment decisions effectively. In recent years, significant progress has been made in unraveling the molecular mechanisms underlying gastric cancer due to rapid advancements in high-throughput technologies and large-scale genomic projects like The Cancer Genome Atlas (TCGA). These endeavors have provided profound insights into the intricate interplay between tumor genomic abnormalities, epigenetic alterations, and immune-related gene dysregulation that drive gastric cancer development and progression.^[7,8] Leveraging these novel biomarkers presents unprecedented opportunities for constructing robust prognostic risk models specifically tailored for gastric cancer.

The Golgi apparatus, a pivotal organelle involved in diverse cellular functions including protein synthesis, modification, and distribution,^[9] has emerged as a subject of growing interest in the context of tumorigenesis and cancer progression.^[10] Firstly, the Golgi apparatus plays a crucial role in protein synthesis and modification processes such as transport, glycosylation, phosphorylation, and folding. These intricate processes are indispensable for maintaining normal cellular homeostasis. However, dysregulation of Golgi function in tumors can disrupt the delicate balance of protein synthesis and modification,^[11] leading to aberrant accumulation or loss of proteins that profoundly impact cell growth, proliferation, and metastatic potential. Secondly, accumulating evidence underscores the intricate relationship between the Golgi apparatus and tumor immune evasion mechanisms.^[12] For instance, the Golgi apparatus exerts regulatory control over antigen presentation and major histocompatibility complex molecule expression, thereby influencing the recognition of tumor cells by the immune system.^[13] Consequently, targeting the Golgi apparatus has garnered increasing attention as a potential therapeutic strategy for tumors.^[14] Lastly, previous investigations have revealed an association between abnormal expression patterns of Golgi apparatus-related genes (GARG) and unfavorable prognosis in specific subsets of gastric cancer patients,^[15] suggesting a potential involvement of the Golgi apparatus in gastric cancer initiation and prognosis. Therefore, developing prognostic risk models based on GARG holds significant importance.

The present investigation delves into an in-depth exploration of the intricate interplay between Golgi-related gene expression and clinical pathological features in patients afflicted with gastric cancer, employing publicly accessible transcriptomic and clinical datasets. The primary objective of this study is to scrutinize the prognostic implications of GARGs’ expression, thereby augmenting the existing body of evidence regarding their potential as robust prognostic markers in gastric cancer.

2. Materials and methods

2.1. Data download and preprocessing

The transcriptomic profiles and corresponding clinical-pathological annotations of the TCGA-STAD cohort were acquired from The Cancer Genome Atlas (TCGA, https://portal.gdc.cancer.gov/) database. Rigorous inclusion criteria were applied to exclude cases with incomplete clinical and pathological information as well as prognosis data, resulting in the enrollment of a total of 302 patients diagnosed with gastric cancer. Additionally, the GSE15459 cohort, comprising 192 cases of gastric cancer, was procured from the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) database. To delineate the gene set associated with Golgi apparatus-related functions, we retrieved the GARG gene set encompassing 1643 genes from the GOCC_GOLGI_APPARATUS gene set available in the MSigDB (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp) database.

2.2. Construction of prognostic features derived from GARG

Initially, a univariate Cox regression analysis was executed to identify GARGs that exhibited significant associations with the prognosis of gastric cancer patients. Subsequently, lasso Cox regression analysis was employed to effectively reduce the number of prognostic genes and mitigate the risk of overfitting. Furthermore, a multivariate Cox regression analysis was conducted to ascertain independent prognostic GARGs and establish a comprehensive risk score utilizing the following equation: riskscore = ∑(βi * gene_expressioni), where βi denotes the coefficient corresponding to gene i and gene_expressioni represents the expression level of gene i. Subsequent to risk score computation, patients were dichotomized into high-risk and low-risk groups based on the median risk score, followed by survival analysis and receiver operating characteristic curve analysis for robust evaluation of prognostic performance.

2.3. Tumor immune infiltration analysis

Tumor immune cell infiltration was assessed using the CIBERSORT and ESTIMATE algorithms. The CIBERSORT R package was employed for the comprehensive analysis of infiltration levels of 22 distinct immune cell types. Additionally, the ESTIMATE R package was utilized to compute the ESTIMATE score, Immune score, and Stromal score, providing quantitative measurements of tumor microenvironment composition and immune cell presence.

2.4. Tumor mutation burden analysis

Somatic mutation data for the TCGA-STAD cohort were acquired from the TCGA database. Subsequently, comprehensive analysis of tumor mutation burden and visualization of mutation waterfall plots were performed using the maftools R package, which is widely recognized for its proficiency in genomic data analysis.

2.5. Drug sensitivity analysis

Drug sensitivity analysis was conducted employing the pRRophetic R package, a robust tool widely acknowledged for its proficiency in pharmacogenomic analysis. The IC50 (lowest inhibitory concentration) values of various drugs were calculated based on the TCGA-STAD patient expression matrix. Subsequently, a comparative assessment of drug sensitivity disparities between high-risk and low-risk groups was performed for a comprehensive panel of 45 distinct drugs.

2.6. Immunotherapy response assessment

The Tumor Immune Dysfunction and Exclusion (TIDE) algorithm, a widely recognized tool for evaluating immunotherapy response, was utilized in this study. The standardized transcriptome data from the TCGA-STAD cohort served as the basis for calculating TIDE scores through the official TIDE website hosted by Harvard University’s Dana-Farber Cancer Institute (http://tide.dfci.harvard.edu/). Subsequently, a comparative analysis of TIDE scores between high-risk and low-risk group patients was performed to identify potential disparities. Moreover, an in-depth investigation into immune checkpoints (ICPs) and human leukocyte antigens (HLAs) was conducted to elucidate immunological variances and evaluate the response to immunotherapy within the high-risk and low-risk groups.

2.7. Gene set enrichment analysis (GSEA)

GSEA was conducted employing the clusterProfiler R package, a robust and widely accepted tool for assessing the enrichment of biological processes and KEGG pathways in the high-risk and low-risk patient groups. To account for multiple testing, the Benjamini-Hochberg correction method was applied, with a predetermined significance threshold of P < .05. This stringent approach ensured reliable identification of significantly enriched gene sets, minimizing the likelihood of false positive results.

2.8. Nomogram construction

Prognostic factors in gastric cancer patients were assessed through rigorous univariate Cox and multivariate Cox regression analyses. To develop a comprehensive prognostic model, the rms R package was employed to construct a nomogram model, which integrated multiple variables. The performance of the nomogram model was meticulously evaluated using calibration curve analysis to assess its predictive accuracy. Additionally, decision curve analysis was conducted utilizing the rmda R package to evaluate the clinical utility and net benefit of the nomogram model, ensuring robustness and practicality in clinical decision-making.

2.9. Statistical analysis

The statistical analysis of the data was conducted using R version 4.2.2, a widely used software for statistical computing and graphics in the academic community. Between-group comparisons were examined using the Wilcoxon test, a nonparametric test suitable for analyzing differences between independent groups. Survival analysis was performed by constructing Kaplan–Meier survival curves and conducting the log-rank test, a commonly employed method for comparing survival distributions between groups. Statistical significance was determined at a P-value threshold of < 0.05, adhering to standard practices in hypothesis testing within the field of biostatistics.

3. Results

3.1. Construction of the gastric cancer risk model derived from GARG

After performing data cleansing procedures, the TCGA-STAD dataset comprised a total of 29,451 genes, while the GEO dataset included 22,867 genes. Among these, a subset of 1448 genes with known GARGs was selected for subsequent model construction and evaluation (Fig. 1A). To identify GARGs with prognostic significance, univariate Cox regression analysis was conducted, resulting in the identification of 365 GARGs significantly associated with gastric cancer prognosis (Table S1, Supplemental Digital Content, http://links.lww.com/MD/L908). Subsequently, lasso Cox regression analysis was employed to further refine this set of prognostic GARGs down to a final selection of 11 genes (Fig. 1B and C). Lastly, multivariate Cox regression analysis revealed that out of these 11 GARGs, 5 were determined to be independent prognostic factors: nerve growth factor (NGF), ATP binding cassette subfamily G member 1 (ABCG1), ChaC glutathione specific gamma-glutamylcyclotransferase 1 (CHAC1), glucosylceramidase beta 2 (GBA2), and proprotein convertase subtilisin/kexin type 7 (PCSK7). Notably, high expression levels of NGF and ABCG1 were associated with poor prognosis in gastric cancer patients, whereas low expression levels of CHAC1, GBA2, and PCSK7 were indicative of unfavorable outcomes (Fig. 1D). The chromosomal locations corresponding to these 5 prognostic GARGs are visually depicted in Figure 1E.

Figure 1.

Construction of GARG risk features in gastric cancer patients. (A) Venn diagram showing the inclusion of GARGs in various datasets. (B–C) Lasso Cox regression analysis to reduce prognosis-related GARGs. (D) Identification of 5 independent prognostic GARGs through multivariate Cox regression analysis. (E) Chromosomal localization of the 5 GARGs in the risk features.

3.2. Evaluation of the GARG signature

The risk score formula, incorporating the 5 aforementioned genes, was established as follows: riskscore = 0.5428548 * ABCG1–0.2958615 * CHAC1–0.6216817 * GBA2 + 0.5744943 * NGF – 0.8519342 * PCSK7. By employing the median risk score as a threshold, we stratified both the TCGA-STAD and GSE15459 cohorts into high-risk and low-risk groups (Fig. 2A and B). Subsequent survival analysis demonstrated that patients assigned to the high-risk group exhibited significantly poorer prognosis compared to those in the low-risk group (Fig. 2C, P < .0001). Moreover, the area under the curve (AUC) values for predicting overall survival at 1, 3, and 5 years in gastric cancer patients based on the risk score were determined as 0.72, 0.807, and 0.877 respectively (Fig. 2D). Similarly, within the GSE15459 cohort, individuals were classified into high-risk and low-risk groups (Fig. 2E and F), with notable disparity in prognostic outcomes observed between these 2 groups (Fig. 2G, P = .028). The AUC values for predicting overall survival at different time points (1, 3, and 5 years) in this cohort were calculated as follows: 0.566, 0.599, and 0.611, respectively (Fig. 2H).

Figure 2.

Evaluation of prognostic performance of GARG-derived gastric cancer risk features. (A) Risk scoring divides the TCGA-STAD cohort into high-risk and low-risk groups. (B) Scatter plots of survival time and survival outcome for gastric cancer patients in the TCGA-STAD cohort. (C) Kaplan–Meier survival curve analysis for high-risk and low-risk groups in the TCGA-STAD cohort. (D) ROC analysis of riskscore predicting prognosis of patients in the TCGA-STAD cohort. (E) Risk scoring divides the GSE15459 cohort into high-risk and low-risk groups. (F) Scatter plots of survival time and survival outcome for gastric cancer patients in the GSE15459 cohort. (G) Kaplan–Meier survival curve analysis for high-risk and low-risk groups in the GSE15459 cohort. (H) ROC analysis of riskscore predicting prognosis of patients in the GSE15459 cohort.

3.3. Immune infiltration differences between the high-risk and low-risk patients

In the TCGA-STAD cohort, we conducted a comprehensive assessment of tumor immune cell infiltration and analyzed the disparities in immune infiltration patterns between high-risk and low-risk patients. Our findings, as illustrated in Figure 3A, revealed that high-risk patients exhibited elevated levels of naive B cells, M1 and M2 macrophages, resting dendritic cells, activated NK cells, and resting mast cells compared to the low-risk group. Conversely, high-risk patients demonstrated lower levels of resting NK cells, M0 macrophages, activated dendritic cells, and activated mast cells. Notably, correlation analysis unveiled significant associations between CHAC1, GBA2, NGF, PCSK7 expression levels with specific immune cell infiltrations (Fig. 3B), indicating their potential involvement in shaping the tumor immune microenvironment. Moreover, our investigation indicated that low-risk patients displayed significantly reduced ESTIMATE_SCOREs, Stromal_SCOREs, and immune_scores when compared to the high-risk group (Fig. 3C–E).

Figure 3.

Differences in immune infiltration between high-risk and low-risk groups derived from GARG. (A) Comparison of immune cell infiltration in 22 types of tumors between high-risk and low-risk patients. (B) Relationship between 5 GARG risk features and immune cell infiltration. Comparison of ESTIMATE_SCORE (C), Stromal_SCORE (D), and immune_score (E) between high-risk and low-risk patients. *P < .05, **P < .01, ***P < .001, ****P < .0001.

3.4. Differences in chemotherapy and immunotherapy response between high-risk and low-risk groups

We conducted an extensive analysis to assess the sensitivity of high-risk and low-risk patients to a panel of 45 chemotherapy drugs. Our findings revealed that, in comparison to the high-risk group, low-risk patients exhibited enhanced sensitivity to 3 specific drugs: epothilone.B, metformin, and tipifarnib. However, no significant differences in sensitivity were observed between the high-risk and low-risk groups for the remaining 42 drugs tested (Fig. 4A). Furthermore, we performed a comprehensive analysis of the expression disparities in ICP genes between the high-risk and low-risk groups. The quantification and description of these expression differences are depicted in Figure 4B. Notably, our evaluation using the TIDE algorithm demonstrated significantly lower TIDE scores among low-risk patients compared to high-risk patients (Fig. 4C). Although immunotherapeutic approaches involving ICP inhibitors such as anti-CTLA-4 monoclonal antibodies and anti-PD-1/PD-L1 antibodies have shown clinical efficacy and improved patient prognosis, it is crucial to consider other factors that may influence immunotherapy sensitivity, including HLAs. Therefore, we investigated the expression patterns of HLA gene family members in both high-risk and low-risk groups. Intriguingly, our results indicated higher expression levels of most HLA genes within the high-risk group (Fig. 4D).

Figure 4.

Differences in chemotherapy and immunotherapy response between high-risk and low-risk groups derived from GARG. (A) Sensitivity differences to 45 drugs between high-risk and low-risk patients. (B) Differences in immune checkpoint gene expression between high-risk and low-risk patients. (C) Differences in TIDE scores between high-risk and low-risk patients. (D) Differences in HLA expression between high-risk and low-risk patients. *P < .05,**P < .01,***P < .001,****P < .0001.

3.5. Differences in biological processes, KEGG pathways, and somatic mutations between high-risk and low-risk groups

GSEA revealed that high-risk patients displayed significant suppression of various biological processes, including antigen receptor-mediated signaling pathway, humoral immune response, complement activation, and B cell receptor signaling pathway, when compared to low-risk patients. Conversely, chromosome-related biological processes exhibited notable activation in the high-risk group (Fig. 5A). Furthermore, high-risk patients exhibited remarkable inhibition of KEGG pathways such as cytokine-cytokine receptor interaction, complement and coagulation cascades, and Th17 cell differentiation. In contrast, pathways related to nucleocytoplasmic transport, homologous recombination, and cell cycle were significantly activated in comparison to the low-risk group (Fig. 5B). The top ten most frequently occurring somatic mutations in both high-risk and low-risk groups are presented in Figure 5C and D. Notably, low-risk patients demonstrated a higher tumor mutation burden (TMB) than those in the high-risk group (Fig. 5E), and a statistically significant negative correlation was observed between riskscore and TMB (Fig. 5F; R = −0.154; P = .008).

Figure 5.

Differences in biological processes, KEGG pathways, and somatic mutations between high-risk and low-risk groups derived from GARG.(A) Biological processes significantly activated or inhibited in high-risk compared to low-risk patients. (B) KEGG pathways significantly activated or inhibited in high-risk compared to low-risk patients. (C) The top 10 genes with the most frequent somatic mutations in the high-risk group. (D) The top 10 genes with the most frequent somatic mutations in the low-risk group. (E) Comparison of tumor mutational burden (TMB) between high- risk and low- risk patients. (F) Significant negative correlation between riskscore and TMB (R = -0.154,P = .008). ****P < .0001.

3.6. Associations between GARG-derived risk features and clinical pathological characteristics

The risk scores were compared across various subgroups based on clinical pathological characteristics, as depicted in Figure 6A–H. Our analysis revealed that elderly patients (age > 60) exhibited significantly higher risk scores than younger patients (age ≤ 60) (P = .021). Furthermore, the mean risk score of deceased patients was substantially elevated in comparison to surviving patients (P = 7.9e−11). Notably, patients who did not undergo radiotherapy demonstrated significantly higher risk scores than those who received radiotherapy (P = .046). In terms of pathological stage, the risk scores for stage I were significantly lower than stages II (P = .0026), III (P = .029), and IV (P = .021). Additionally, patients with stage T2 (P = 6.5e−5), T3 (P = 1.1e−5), and T4 (P = 9.3e−6) displayed markedly higher risk scores compared to those with stage T1. However, no significant differences in risk scores were observed among different stratifications of chemotherapy, N stage, and M stage.

Figure 6.

Correlation between risk features derived from GARG and clinical pathological features in different age groups (A), survival outcomes (B), radiotherapy (C), chemotherapy (D), pathological staging (E), and TNM staging subgroups (F–H).

3.7. GARG-derived gastric cancer prognostic nomogram

Univariate Cox analysis was conducted to assess the prognostic factors for gastric cancer patients, including GARG-derived risk score (P = 1.2e−14), radiotherapy (P = .0071), pathological stage (P = .016), and N stage (P = .027) (Fig. 7A). However, multivariate Cox regression analysis revealed that only GARG-derived risk score (P < .0001) and age (P = .004) remained as independent prognostic factors for gastric cancer (Fig. 7B). Consequently, a nomogram model incorporating risk score and age was developed to predict the 1-year, 3-year, and 5-year overall survival of gastric cancer patients (Fig. 7C). The calibration curve demonstrated excellent concordance between the predicted and observed values for the 1-year, 3-year, and 5-year overall survival rates (Fig. 7D). Moreover, when compared to other prognostic factors, this nomogram exhibited superior net benefit in predicting the 1-year overall survival of gastric cancer patients (Fig. 7E).

Figure 7.

Construction of a prognostic assessment nomogram model for gastric cancer patients. Univariate (A) and multivariate Cox regression analyses of risk features derived from GARG and other clinical pathological features. (B) Prognostic assessment nomogram model for gastric cancer based on risk features derived from GARG and age. (C) Calibration curves for assessing 1, 3, and 5-year overall survival of gastric cancer patients using the nomogram. (D) Decision curve analysis comparing the nomogram with other prognostic factors for assessing 1-year overall survival of gastric cancer patients.

4. Discussion

Despite recent advancements in the treatment of gastric cancer, its prognosis remains unfavorable due to tumor heterogeneity, limited therapeutic options, and low rates of early detection.^[15] Consequently, it is imperative to establish a robust classification system that can stratify patients based on different risks and prognoses. This will enable personalized treatment strategies and timely follow-up, thereby maximizing patient outcomes. The Golgi apparatus, an organelle responsible for protein synthesis, modification, and distribution, has been found to play a significant role in regulating various biological processes in tumors. Specifically, mutations in proteins involved in Golgi-related pathways have been associated with increased cancer metastasis and decreased patient survival rates.^[16] Additionally, the Golgi apparatus releases immune components that contribute to the formation of the tumor microenvironment. Alterations in the Golgi apparatus within cancer cells can lead to heightened secretion of immune factors, resulting in the creation of an immunosuppressive tumor microenvironment that promotes cancer progression.^[17,18] Furthermore, studies using front-rear polarity models have demonstrated that the orientation of the Golgi apparatus can influence cell polarity and migration.^[19] In response to chemotactic gradients or changes in extracellular matrix stiffness, the dynamic localization of the Golgi apparatus leads to localized secretion of pro-migratory factors such as matrix metalloproteinases, growth factors, and cytokines. These factors facilitate cell migration and invasion.^[20] Therefore, GARG represents a promising candidate as both a prognostic biomarker and a potential therapeutic target for various cancers.

In this study, we investigated the association between GARG and the prognosis of gastric cancer using publicly available databases. We employed a comprehensive approach to develop a prognostic classifier comprising 5 genes derived from GARG. This classifier was constructed through univariate regression, lasso analysis, and multivariate Cox regression. To validate the prognostic impact of GARG signals, we analyzed data from both TCGA and GEO cohorts. Through systematic bioinformatics analysis, we elucidated the relationship between GARG signals and various aspects of gastric cancer, including tumor immune cell infiltration, gene expression patterns, somatic mutations, drug sensitivity profiles, and response to immune therapy. These findings offer novel insights into enhancing patient prognosis and stratification by incorporating microenvironmental characteristics and transcriptomic information. Furthermore, we developed a novel nomogram integrating GARG signals and age as predictive factors, which demonstrated excellent performance in forecasting gastric cancer prognosis. This nomogram represents an innovative extension of the utility of GARG features in clinical practice.

The tumor microenvironment represents a complex ecosystem comprising noncancerous cells surrounding tumor cells, and the infiltration levels of immune cells often exhibit dynamic changes during tumor progression. In our study, patients with elevated risk scores exhibited heightened immunescores, StromalScores, and ESTIMATEScores. CIBERSORT analysis revealed significant disparities in B cells, NK cells, macrophages, dendritic cells, and mast cells between low-risk and high-risk patients. The aberrant infiltration of immune cells suggests a potential association with gastric cancer development. Specifically, the low-risk group demonstrated diminished levels of M2 macrophage infiltration in comparison to the high-risk group, implying that an immunosuppressive microenvironment may contribute to the unfavorable prognosis observed in high-risk gastric cancer patients. Combination therapy incorporating immunotherapy and chemotherapy has emerged as a primary treatment modality for gastric cancer. While drug sensitivity analysis indicated minimal distinctions between low-risk and high-risk groups, alternative assessments underscored potential variations in immunotherapy responsiveness between them. ICP inhibitors impede T cell immune function, thereby facilitating immune evasion; notably, most high-risk patients exhibited significantly higher ICP expression levels than those in the low-risk group.^[21] Studies have demonstrated that TMB correlates with cancer immunotherapy response and targeted therapy efficacy.^[22,23] In this investigation, we identified higher TMB scores in low-risk patients relative to high-risk patients; moreover, patients with elevated TMB scores displayed poorer prognosis potentially attributable to immune-related effects. Furthermore, we employed the widely utilized TIDE score^[24] to predict immunotherapy sensitivity in numerous cancer patients and found that the TIDE score was significantly elevated among high-risk patients compared to their low-risk counterparts. Collectively, these findings suggest that low-risk patients may exhibit heightened sensitivity to immunotherapy interventions.

The identified feature genes have been extensively documented for their significant involvement in various cancer types. NGF, a neurotrophic factor, exhibits antiapoptotic and angiogenic properties in breast cancer by activating distinct signaling cascades mediated by TrkA and NGFR/p75NTR receptors.^[25] In gastric cancer, NGF acts as a downstream effector of SNRPA and contributes to tumor development through aberrant cholinergic signaling.^[26,27] ABCG1 plays a crucial role in maintaining cholesterol homeostasis and modulating tumor immunity. Studies have demonstrated its promotion of migration and invasion in lung cancer cells,^[28] whereas high nuclear expression of ABCG1 serves as an unfavorable prognostic indicator in hepatocellular carcinoma.^[29] Moreover, ABCG1 is implicated in the tumor microenvironment as it regulates the function of tumor-infiltrating macrophages to suppress tumor growth.^[30] CHAC1 induces glutathione depletion, resulting in the accumulation of reactive oxygen species and TP53 somatic mutations. Concurrently, Helicobacter pylori infection upregulates CHAC1 expression to facilitate the development of gastric cancer.^[31,32] GBA2 exhibits antitumor properties by modulating endogenous sphingolipid metabolism through enhancing glucosylceramide degradation and synthesis. Additionally, it induces cellular apoptosis via unfolded protein response.^[33] PCSK7 is found to be downregulated in non-small cell lung cancer patients and can be employed alongside GPX1, BCL9L, and MAP3K7CL for prognostic assessment of non-small cell lung cancer cases.^[34] Consequently, the identified feature genes in this study offer promising targets for laboratory investigations aimed at elucidating the molecular mechanisms underlying gastric cancer.

This study has several notable limitations. Firstly, the utilization of retrospective datasets obtained from publicly available databases introduces inherent selection bias, which could potentially affect the robustness of the developed models. To comprehensively assess the clinical significance of GARG features, it is imperative to conduct further validation through prospective and multicenter studies. Additionally, additional in vivo and in vitro experiments are warranted to investigate the intricate interplay between GARG and the TME, as well as to elucidate the underlying molecular mechanisms that underpin our observed findings.

5. Conclusion

In conclusion, our constructed risk model based on GARG exhibits significant predictive capability for patient prognosis. Moreover, the observed association between this risk model and drug sensitivity as well as immune therapy response necessitates further exploration. Therefore, immediate attention should be given to conducting in vivo and in vitro experiments aimed at validating the involvement of GARG in Golgi morphology and function, while also elucidating its intricate relationship with tumor initiation and progression in gastric cancer.

Author contributions

Conceptualization: Rui Liu, Weiwei Chu.

Data curation: Rui Liu, Weiwei Chu.

Funding acquisition: Haiming Wang.

Investigation: Haiming Wang.

Methodology: Rui Liu, Haiming Wang.

Project administration: Haiming Wang.

Resources: Xiaojin Liu.

Software: Jie Hong.

Supervision: Haiming Wang.

Validation: Rui Liu.

Visualization: Rui Liu, Weiwei Chu.

Writing – original draft: Rui Liu, Weiwei Chu.

Writing – review & editing: Haiming Wang.