Expression profiling, molecular subtyping and prognostic signature construction of chemokines and chemokine receptors in gastric cancer Authors: Tianshu Feng, Yaoqi Li, Jieyu Hu, Liang Zhao, Xiaobin Yao Published: 2026 ABSTRACT AbstractBackgroundGastric cancer (GC) is a highly heterogeneous malignancy of the digestive tract, with significant variations in treatment response and clinical prognosis depending on its molecular subtypes. Chemokines and chemokine receptors (CCRs) have been implicated in GC initiation and progression, yet the precise mechanisms underlying their role remain poorly defined. This study mainly focused on a systematic investigation of CCRs in GC is of critical clinical and scientific importance.MethodsWe comprehensively analyzed the transcriptional profiles and mutational landscape of CCRs in GC and constructed a CCR-related prognostic signature. The Cancer Genome Atlas (TCGA) cohort (371 GC samples) served as the training set, and the GSE84437 dataset was used as the validation cohort. Consensus clustering based on CCR gene expression was performed for patient stratification. A prognostic risk model was developed using a stepwise approach combining univariate Cox regression, the least absolute shrinkage and selection operator (LASSO), and multivariate Cox regression analyses. Correlations between the risk score, tumor mutation burden (TMB), and immunotherapy response were also evaluated.ResultsWe identified 25 significantly upregulated and 12 significantly downregulated CCR genes in GC tissues. Consensus clustering based on CCR expression effectively stratified patients into subgroups with distinct prognoses and tumor microenvironment (TME) characteristics. A 9-gene prognostic signature, including four novel genes (ENTPD8, RNASE1, HABP2, and ENTPD2), was established. The derived risk score demonstrated robust predictive value for overall survival (OS) in GC. Patients with lower risk scores exhibited higher TMB levels and were more likely to benefit from immunotherapy.ConclusionsOur study provides novel insights into the biological roles of CCRs in GC. We propose a CCR-based molecular stratification strategy and a prognostic signature that may serve as a complementary tool for risk assessment in GC patients, pending further prospective validation.Keywords: Gastric cancer (GC), chemokines, chemokine receptors, prognostic signature, The Cancer Genome Atlas (TCGA) Highlight box Key findings • A novel 9-gene chemokines and chemokine receptor (CCR)-related prognostic signature, including four novel genes (ENTPD8, RNASE1, HABP2, and ENTPD2), was constructed to predict the overall survival of gastric cancer (GC) patients with robust efficacy. What is known and what is new? • Chemokines and CCRs are known to participate in the progression of various cancers, including GC, but their comprehensive expression profiles, clinical significance, and prognostic potential in GC remain unclear. • This study systematically analyzed the transcriptional and mutational landscape of CCRs in GC, identified two distinct molecular subtypes, and validated a CCR-related prognostic signature that correlates with prognosis, tumor mutation burden, and immunotherapy response. What is the implication, and what should change now? • This study provides a new molecular stratification strategy and a prognostic risk model for GC, which may serve as a valuable reference for risk stratification, individualized treatment planning, and immunotherapy response prediction in clinical practice. Introduction According to Global Cancer Statistics data, gastric cancer (GC) ranks as the fifth most commonly diagnosed cancer and the fourth leading cause of cancer-related death worldwide (1). Most GC patients are diagnosed at an advanced stage, resulting in a 5-year survival rate of less than 30% (2). In recent years, the application of targeted agents and immunotherapies has led to improved prognosis for patients with advanced GC (3). Nevertheless, due to the high degree of heterogeneity in GC, therapeutic responses and prognostic outcomes differ markedly across distinct molecular subtypes (4). Given the limitations of traditional GC classification systems, further in-depth molecular investigations are urgently required to advance the diagnosis and treatment of this disease. Chemokines are polypeptides with a molecular weight of 8–14 kDa, which are the largest family of cytokines. The main function of chemokine is to recruit monocytes, neutrophils and lymphocytes from the blood into specific lymphoid organs and tissues and the site of infection (5). Chemokine receptors are usually expressed on the cell membrane of neutrophils, endothelial cells, immune cells and so on (6). Chemokines and chemokine receptors (CCRs) govern cell entrance and exit into and out of tumor microenvironment (TME), which are linked to tumor occurrence and progression (7). CCRs can stimulate directed migration of cells during inflammation, and long-term inflammation will offer a suitable surrounding microenvironment for tumor growth and development, thus promoting carcinogenesis (8). Previous studies have shown CCRs play significant roles in the pathogenesis of GC (9). However, existing CCR-related studies in GC have primarily focused on individual chemokine-chemokine receptor axes (e.g., CXCL12/CXCR4) or single-gene analyses, while comprehensive bioinformatics analyses integrating the entire CCR family for molecular subtyping and prognostic modeling remain limited. Several recent studies have constructed immune-related or inflammation-related prognostic signatures in GC, but few have specifically focused on the systematic analysis of all CCR family members using consensus clustering combined with least absolute shrinkage and selection operator (LASSO)-Cox regression (10-12). After operation or systemic therapy, the concentration of various chemokines (such as CXCL-1, CXCL2, CXCL5) in peripheral circulation of GC patients is lower than that before treatment (10). CCL2, CCL4 and CCL5 levels in plasma increased more in GC patients than in healthy controls. Moreover, CCL2 concentration was linked to the clinical stage of GC (11). CCRs can regulate GC growth, invasion and metastasis, etc. (12). However, the role of CCRs in GC is complex, some CCRs promote tumor progression, while others are anti-tumor factors. Hence, it is urgent to conduct a systematic bioinformatics analysis to reveal the roles of CCRs in GC. The initial step in this research was to examine the RNA level and genomic landscape of CCRs in GC. Then, we identified two subtypes with distinct TME status and prognostic value by Consensus Clustering. On this basis, the CCRs-related prognostic signature was established by employing the LASSO-Cox approach based on the differentially expressed genes (DEGs) identified between the two CCR-based subtypes. This prognostic signature can be applied to predict prognosis, the proportion of immune-stromal component, and tumor mutation burden (TMB) in GC patients. Based on the prognostic signature, we next constructed a predictive nomogram for GC. Our data revealed a potential link between CCRs, prognosis, TME, and TMB state of GC patients. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0590/rc). Methods Data processing RNA-seq profiles, somatic mutation files, and corresponding clinical information of patients with GC were retrieved from the The Cancer Genome Atlas (TCGA) database. In addition, the GC-related microarray dataset GSE84437 (annotated by the GPL6947 platform) was acquired from the GEO database. Copy number variation (CNV) data for GC were obtained from the UCSC repository. The detailed clinicopathological characteristics of the enrolled GC patients are summarized in table available at https://cdn.amegroups.cn/static/public/tcr-2026-0590-1.xlsx. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. DEGs and mutation analysis of CCRs A total of 62 CCRs genes were collected from prior reviews (7-12), table available at https://cdn.amegroups.cn/static/public/tcr-2026-0590-2.xlsx contained the CCRs genes list. Using the R packages “limma”, the CCRs genes expression in GC and normal tissues was compared. STRING website was then utilized to create the protein-protein interaction (PPI) network for CCRs. The R package “maftools” was utilized to create waterfall plot to show the mutation status of CCRs genes in GC. Consensus clustering Unsupervised cluster analysis was used to separate the GC patients into various clusters based on the expression of CCRs genes. The consensus clustering algorithm of R package “ConsenSuClusterPlus” was employed to compute the number of clusters and their stabilities. To ensure stability and accuracy of the classification, the arithmetic was repeated 1,000 times. Immune landscape of GC The immune and stromal cell scores in GC were calculated using the ESTIMATE algorithm (13). The Immune Score and Stromal Score respectively reflect the abundance of immune and stromal components within the TME. The ESTIMATE Score, defined as the sum of the aforementioned two scores, represents the overall proportion of non-tumor cells in the TME. Additionally, single-sample gene set enrichment analysis (ssGSEA) was performed to assess the infiltration levels of diverse immune cell subsets and the activation status of immune-associated pathways (14). Tumor stemness, metabolic characterization and immune evasion of subtypes To comprehensively characterize the biological features of the CCR-based subtypes, we performed the following additional analyses: The expression levels of RNAss and EREG.EXPss were calculated. Significance between groups was assessed by Wilcoxon’s rank-sum test (P values are noted in the figure). Based on metabolic pathway databases [such as Kyoto Encyclopedia of Genes and Genomes (KEGG), Reactome], calculate the metabolic pathway activity of each sample (such as tryptophan metabolism, glycogen metabolism, mTOR signaling, etc.). Hierarchical clustering heat maps were used to show inter-sample differences in metabolic pathway activity, and the similarity of pathways was displayed through cluster trees. Based on the set of immune-related genes (Antigen Presentation, Checkpoints, Immunosuppressive, MDSC, T_cell_Exclusion, Treg, etc.), the immune escape score (single sample GSEA) is calculated for each sample. Differential expression analyses Differential expression analyses were performed to identify DEGs between GC and normal tissues, as well as between the two clusters and two risk groups, respectively. The R package “limma” was employed to conduct these analyses. Genes meeting the criteria of |logFC| >0.585 and adjusted P value <0.05 were defined as significant DEGs. Statistical comparisons of DEGs across different groups were performed using the Wilcoxon rank-sum test. Prognostic model construction based on CCRs‑related signature The GSE84437 dataset was assigned as the test cohort (n=433), while the TCGA-STAD cohort served as the training cohort (n=371). A CCRs-associated prognostic signature was established in the training cohort. To ensure biological relevance to the CCR-based subtypes, we first identified 2,487 DEGs between Cluster 1 and Cluster 2 (the two CCR-based molecular subtypes). These subtype-specific DEGs were then subjected to prognostic gene screening. First, univariate Cox regression analysis was performed to screen the subtype-DEGs with significant prognostic value (P<0.05). Next, LASSO Cox regression was applied to further select prognostic DEGs and avoid model overfitting. Finally, multivariate Cox regression analysis was conducted to determine the regression coefficients and construct the final CCRs-related prognostic signatureThe formula for calculating the risk score was as follows:Risk score = Σ (Expression of each RNA × Corresponding coefficient).The area under the curve (AUC) was calculated using the R package “survivalROC” to evaluate the prognostic performance of the established signature (15). Functional enrichment analysis Based on the DEGs identified between the two risk groups, Gene Ontology (GO) functional annotation and KEGG pathway enrichment analyses were performed using the R package “clusterProfiler”. TMB related analysis Somatic mutation data of GC were downloaded from the TCGA database, and the TMB value was calculated for each sample. Subsequently, the Spearman correlation analysis was performed to assess the relationship between TMB values and risk scores. In addition, the differences in TMB levels between the two risk groups were further evaluated. According to the optimal cutoff value of TMB, all GC samples were classified into high-TMB and low-TMB subgroups, followed by Kaplan-Meier (KM) survival analysis. Finally, a stratified survival analysis was conducted by combining TMB status and risk score stratification. Statistical analysis All statistical analyses were performed using R software (version 4.2.1) and relevant R packages including limma, ConsensusClusterPlus, maftools, estimate, survival, survivalROC, clusterProfiler, and stats. Two-tailed statistical tests were adopted for all analyses, and statistical significance was defined as adjusted P<0.05 unless otherwise specified. Results Overview of genetic changes and expression variations of CCRs in GC Firstly, we contrasted the RNA levels of CCRs genes in TCGA between GC and normal tissue. In total, 37 DEGs were identified, including 25 upregulated genes (including CXCR1) and 12 downregulated genes (CXCR2, CXCL17, CCL21, CCL14, CXCL12, CCR10, CXCL14, ACKR3, CCL11, CCL28 and CXCL13) in GC tissues (Figure 1A). The PPI network between the CCRs was shown in Figure 1B, which illustrated the interactions of these CCRs. The genetic level (somatic mutation and CNVs) of the CCRs in GC was summarized. A total of 84 of 433 (19.4%) GC samples experienced mutations of CCRs, as shown in Figure 1C. CX3CL1 had the highest mutation frequency (9/433,2.1%). Figure 1D depicts the location of CNV alterations in CCRs on chromosomes. The copy number gain frequencies were higher than CNV loss frequencies, CCR7 had the highest copy number gain frequency among all CCRs (Figure 1E). The overall picture of CCRs in GC from TCGA cohort. (A) CCRs genes expression in GC (red) and normal tissues(blue). The figure only shows the DEGs (*, P<0.05; **, P<0.01; ***, P<0.001; Wilcox test). (B) PPI network for 62 CCRs genes by STRING website. (C) The CNV occurrence and mutation frequency of CCRs genes in GC. (D) The chromosomal location of CNV changes in CCRs genes. (E) The CNV alteration frequency of CCRs in GC, CNV gain (red), CNV loss (green). CCRs, chemokines and chemokine receptors; CNV, copy number variation; DEGs, differentially expressed genes; GC, gastric cancer; PPI, protein-protein interaction; TCGA, The Cancer Genome Atlas. Identification of a GC classification pattern mediated by CCRs A total of 371 GC patients with complete RNA-seq data and clinical information were collected from TCGA. After extracting the expression data of CCRs genes, we identified two different CCRs-related regulatory patterns using consensus clustering analysis, including 203 patients in Cluster 1 and 168 patients in Cluster 2 (Figure 2A, Figure S1A-S1J). The overall survival (OS) of the patients in Cluster 1 were better (Figure 2B). We then explored the correlation of the CCRs and clinicopathological features in TCGA cohort. The heatmap manifested that CCRs expression in Cluster 2 was higher, while there is no significant difference between Cluster 1 and Cluster 2 in clinicopathological features (Figure 2C). Next, ESTIMATE and ssGSEA analysis were performed to further probe biological behavior of Cluster 1 and 2 (Figure 2D). We found Cluster 2 has higher ESTIMATE score, immune score, stromal score and enrichment scores of different immune cells and associated functions or pathways than cluster 1. Clusters of GC related by CCRs genes. (A) Based on the expression levels of CCRs genes, GC samples were classified into two distinct clusters through Consensus clustering analysis. (B) Kaplan-Meier survival curves were plotted for the two clusters, and the OS analysis showed a statistically significant difference (P=0.002). (C) A heatmap was constructed to visualize the clinicopathological characteristics and the expression levels of CCRs in the two clusters. (D) The landscape of immune cell infiltration in the two clusters was analyzed and presented. (E) Comparison of RNAss and EREG-EPs expression levels between the two clusters. (F) Heatmap showing differential metabolic pathway activities between Cluster 1 and Cluster 2. (G) Distribution of immune estimation scores and immune-related metrics across the two clusters. CCRs, chemokines and chemokine receptors; GC, gastric cancer; OS, overall survival. Comprehensive characterization of CCR-based subtypes To further elucidate the biological characteristics of the two CCR-based subtypes, we conduct stratified analysis on the tumor stemness, metabolic characteristics and immune evasion of each subtype. We conduct stratified analysis on the tumor stemness based on RNAss and EREG.EXPss profiling. Figure 2E demonstrates that both RNAss (P<2.22e-16) and EREG.EXPss (P=0.00011) were significantly upregulated in Cluster 1 compared to Cluster 2. Subsequent metabolic pathway analysis revealed divergent reprogramming between the two subtypes (Figure 2F). Cluster 1 was characterized by enhanced activity in anabolic pathways, specifically mTOR Signaling, Fatty Acid Synthesis, and TCA Cycle. Furthermore, immune evasion analysis indicated distinct microenvironments (Figure 2G). Cluster 2 displayed significantly higher scores for Checkpoints, Immunosuppressive signals, MDSC, T_cell_Exclusion, and Treg infiltration (all P<0.0001), whereas no significant differences were observed in Antigen_Presentation scores. Construction of the prognostic signature in TCGA Previous researches have revealed CCRs are closely related to the occurrence, development and metastasis of GC (7-9). To link the CCRs-related subtypes to the clinical treatment of GC, we next identified the DEGs between Cluster 1 and 2 and constructed a prognostic signature. Specifically, 2,487 DEGs were identified between the two CCR-based subtypes (|logFC| >0.585, adjusted P<0.05). We uncovered 2,487 DEGs and included them into univariate COX analysis (table available at https://cdn.amegroups.cn/static/public/tcr-2026-0590-3.xlsx). The lasso analysis included 524 significant CCRs-related DEGs found by univariate COX analysis (table available at https://cdn.amegroups.cn/static/public/tcr-2026-0590-4.xlsx). Twenty-two OS-associated DEGs were selected by LASSO regression (Figure 3A,3B). Finally, nine DEGs (ENTPD8, PVT1, RNASE1, CXCR4, PDCD1, HLA-G, HABP2, ENTPD2, SERPINE1) were identified as independent prognostic predictors using multivariate Cox analysis (Figure 3C). Construction of the prognostic signature by LASSO and multivariate Cox analysis. (A,B) Coefficient of the LASSO-Cox model in GC. (C) Nine DEGs were identified to construct the prognostic signature after optimized by multivariate Cox analysis. *, P<0.05; **, P<0.01; ***, P<0.001. DEGs, differentially expressed genes; GC, gastric cancer; LASSO, least absolute shrinkage and selection operator. Evaluating the performance of the prognostic signature After constructing the prognostic signature, we assessed the performance of the risk scores in TCGA-STAD (train cohort) and GSE84437 (test cohort). The following algorithm was utilized to compute the risk scores: Risk score = (−0.288 * ENTPD8 exp.) + (−0.184 * PVT1 exp.) + (0.113 * RNASE1 exp.) + (0.201 * CXCR4 exp.) + (−0.701 * PDCD1 exp.) + (0.214 * HLA-G exp.) + (0.086 * HABP2 exp.) + (−0.148 * ENTPD2 exp.) + (0.259 * SERPINE1 exp.). The calculated coefficient was exhibited in Table S1. By the midpoint of the risk scores, patients were separated into high- and low-risk groups (Figure 4A, Figure S2A). Mortality was positively correlated with the elevation of risk scores (Figure 4B, Figure S2B). Principal component analysis (PCA) demonstrated distinct differences in the distribution patterns between these two groups (Figure 4C, Figure S2C). Based on the KM survival curve analysis, the OS of the low-risk group was significantly longer than that of the high-risk group in both the training cohort (Figure 4D) and the testing cohort (Figure S2D). Receiver operating characteristic (ROC) curves showed that the prognostic signature could accurately predict the OS of GC patients. Specifically, the AUC values at 1, 3, and 5 years were 0.735, 0.761, and 0.772, respectively, in the training cohort (Figure 4E), and 0.603, 0.600, and 0.618, respectively, in the testing cohort (Figure S2E). Additionally, calibration plots further confirmed the high prediction accuracy of the prognostic signature (Figure 4F, Figure S2F). Evaluation of the CCRs-related prognostic signature in TCGA. (A,B) The classification of high- and low-risk groups (A) and survival status distribution (B). (C) PCA of the risk scores to distinguish high- and low-risk groups. (D) Survival analysis of high- and low-risk groups (OS). (E) ROC curve for evaluating the performance of CCRs-related prognostic signature (AUC at 1-, 3-, 5-year was 0.735, 0.761 and 0.772, respectively). (F) The calibration curve for evaluating the performance of prognostic signature. AUC, area under the curve; CCRs, chemokines and chemokine receptors; OS, overall survival; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas. The prognostic value of the risk scores Univariate and multivariate Cox regression analyses were utilized to see if the prognostic signature could be an independent prognostic factor of GC. The results revealed that risk scores can be regarded as an independent prognostic predictor for OS, whether in train cohort (Figure 5A,5B) nor in test cohort (Figure S3A,S3B). The prognostic value of CCRs-related prognostic signature in TCGA. (A,B) Univariate (A) and multivariate (B) Cox regression analyses of the signature in train cohort. CCRs, chemokines and chemokine receptors; CI, confidence interval; M, metastasis; N, node; T, tumor; TCGA, The Cancer Genome Atlas. Construction of a predictive nomogram for GC We utilized clinical stage, age, gender, grade, T, N, M and risk score to construct a nomogram for OS prediction of GC (Figure 6A). Calibration plots proved the nomogram was capable of predicting the survival with high accuracy (Figure 6B). ROC analysis was also utilized to examine the prediction accuracy of the nomogram at 1, 3, and 5 years, with AUC values of 0.781, 0.802 and 0.770 at each time point (Figure 6C). Construction and evaluation of the nomogram in TCGA. (A) Nomogram based on clinical stage, age, gender, grade, and risk score. (B) ROC and (C) calibration curves of the nomogram. **, P<0.01; ***, P<0.001. AUC, area under the curve; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas. Functional analyses in the TCGA To explore the potential biological mechanisms underlying the association between risk score and prognosis, we performed functional enrichment analyses. We first identified DEGs between the two risk groups in the TCGA cohort, followed by GO and KEGG pathway enrichment analyses (Figure 7A,7B). GO enrichment demonstrated that the DEGs were mainly enriched in terms related to the extracellular matrix (ECM), including ECM organization (biological process), ECM component (cellular component), and ECM structural constituent (molecular function) (Figure 7A). Consistently, KEGG pathway analysis revealed significant enrichment of the ECM-receptor interaction pathway (Figure 7B). Functional enrichment analysis of DEGs in TCGA. (A) GO enrichment analysis. (B) KEGG enrichment analysis. BP, biological process; CC, cellular component; DEGs, differentially expressed genes; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; MF, molecular function; TCGA, The Cancer Genome Atlas. Estimation of the proportion of immune-stromal component Since enrichment analysis results were closely associated with the ECM, we further explored the immune and stromal components within the GC TME between the two risk groups. The high-risk group exhibited higher Immune Score (Figure 8A), Stromal Score (Figure 8B), and ESTIMATE Score (Figure 8C), coupled with lower Tumor purity (Figure 8D). Boxplot analyses confirmed that the high-risk group was characterized by a higher proportion of stromal and immune cells. Content of immune-stromal cells in GC samples. (A-D) The Immune Score (A), Stromal Score (B), and ESTIMATE Score (C) were all higher in TCGA high-risk group (red), but tumor purity (D) was higher in TCGA low-risk group (blue). GC, gastric cancer; TCGA, The Cancer Genome Atlas. The relation between the signature and TMB As a biomarker for GC immunotherapy, TMB is strongly associated with CD8+ T cell infiltration in the TME (16). Figure 9A,9B illustrate the correlation between the prognostic signature and TMB, as well as the differences in TMB levels between the two risk subgroups. The scatter plot revealed a negative correlation between TMB and risk score (Spearman correlation coefficient: R=−0.34, P=1.8e−11; Figure 9A). Consistently, the low-risk group exhibited significantly higher TMB values compared with the high-risk group (Wilcoxon rank-sum test, P=3.4e−08; Figure 9B), which may partially account for the poorer OS observed in the high-risk group. The relationship of risk scores and TMB. (A) Scatter plot illustrating the correlation between risk scores and TMB values (Spearman correlation coefficient: R=−0.34, P=1.8e−11). (B) Comparison of TMB values between high- and low-risk groups (Wilcoxon test, P=3.4e−08). (C) KM survival curves for high- and low-TMB groups (P<0.001). (D) Stratified survival analysis was performed by combining risk scores and TMB values, showing a statistically significant difference (P<0.001). KM, Kaplan-Meier; TMB, tumor mutation burden. KM survival analysis was conducted after GC samples were classified into high- and low-TMB groups based on the optimal TMB cutoff value (Figure 9C). KM curve indicated the survival outcome of high-TMB group was better. Furthermore, stratified survival analysis combining TMB values and risk scores revealed that the high-TMB + low-risk group possessed the best prognosis, while the low-TMB + high-risk group possessed the worst (Figure 9D). It revealed a synergistic effect of risk scores and TMB values in predicting GC prognosis. We performed KM survival analysis on GC specimens, which were divided into high- and low-TMB subgroups according to the optimal TMB cutoff value (Figure 9C). The KM survival curves demonstrated that patients in the high‑TMB group exhibited more favorable survival outcomes. Additionally, stratified survival analysis integrating TMB levels and risk scores showed that the high-TMB/low-risk subgroup had the most favorable prognosis, whereas the low-TMB/high-risk subgroup had the poorest prognosis (Figure 9D). Discussion Chemokines and their receptors play critical roles in the occurrence and progression of GC via complicated and enigmatic mechanisms. Unraveling the underlying mechanisms of GC is conducive to identifying more effective diagnostic and therapeutic strategies for GC patients. To date, comprehensive and systematic bioinformatics analyses of CCRs in GC remain lacking. We discovered that at least half of the CCRs genes were altered between cancer and surrounding tissues, indicating significant roles of CCRs in GC. PPI and genetic profiling further demonstrated the specificity of CCRs in GC. Based on CCRs expression matrix and survival data, patients were clustered into two subgroups. Cluster 2 subgroup had higher CCRs expression and higher content of immune cells, but the prognosis was relatively poor. It suggests that chemokines will recruit more immune cells to the TME, but due to the large amounts of stromal cells (higher stromal scores), a large number of immune cells fail to penetrate the stromal layer and reach the tumor cells, thus failing to effectively kill the tumor cells. It is widely recognized that the TME can be categorized into three major phenotypes: immune-inflamed, immune-excluded, and immune-desert (17). Cluster 2 resembles an immune-excluded subgroup and thus has a poor prognosis. The stratified analysis on the tumor stemness, metabolic characteristics and immune evasion reveal a distinct molecular dichotomy defined by RNAss/EREG signatures, metabolic reprogramming, and immune evasion. The high RNAss and EREG.EXPss in Cluster 1 suggest a hyper-proliferative phenotype potentially driven by EGFR signaling, coupled with a shift towards anabolic metabolism (mTOR, Fatty Acid Synthesis) to support biomass accumulation. Notably, despite these divergent metabolic states, Cluster 2 exhibits a profoundly immunosuppressive microenvironment, evidenced by elevated Checkpointsand T_cell_Exclusionscores. Notably, the biological characteristics of the CCR-based subtypes were consistent with those of the risk-stratified groups. Cluster 2 (immune-excluded phenotype, poor prognosis) corresponded to the high-risk group, which exhibited elevated immune/stromal scores but lower tumor purity This consistency validates that the prognostic signature, derived from subtype-specific DEGs, effectively captures the key biological features of the CCR-based classification. The high-risk group displayed the immune-excluded phenotype characterized by abundant immune cell infiltration that is physically separated from tumor cells by stromal barriers, explaining the paradox of high immune infiltration but poor prognosis. Based on the DEGs between Cluster 1 and 2, we constructed the CCRs-related prognostic signature containing nine genes, ENTPD8, PVT1, RNASE1, CXCR4, PDCD1, HLA-G, HABP2, ENTPD2, and SERPINE1. Studies have found that the expression of ENTPD8 is down-regulated in liver cancer tissues, and this gene can inhibit the malignant behavior of liver cancer cells. It regulates PD-L1 expression through miR-214-5p, and combined with anti-PD-L1 treatment can enhance the anti-cancer effect. ENTPD8 can be used as a prognostic marker and potential target for liver cancer (18). PVT1 is a candidate oncogene whose transcription product is long non-coding RNA (lncRNA) that lncRNA PVT1 is up-regulated in GC and promotes its development, invasion, metastasis and angiogenesis (19-22). Ribonuclease A family member 1, Pancreatic (RNASE1) belongs to the Ribonuclease A superfamily and is less well studied in tumors (23). CXCR4 is a member of chemokine receptors, its ligand is CXCL12 (24). The CXCL12-CXCR4 axis is an important oncogenic pathway of GC (25-27). Moreover, CXCR4 is a prognostic marker of GC, and its high expression is linked to a bad prognosis, it might be a possible treatment target for GC (28,29). Programmed cell death 1 (PDCD1) has been extensively studied in GC and other tumors (30), which will not be described here. HLA-G has immunomodulatory properties and is a non-classical HLA class I molecule (31). The expression of HLA-G is relevant to poor prognosis of GC (32). HABP2 is a common mutated gene in nonmedullary thyroid cancer, and has certain diagnostic and prognostic value in endometrial cancer (33,34). ENTPD2 is upregulated in HCC and stimulates the proliferation of HCC cell lines (35). It is connected to the poor prognosis of HCC (36,37). In GC, SERPINE1 is elevated, which promotes malignant growth and a poor prognosis (38). In addition, SERPINE1 has been shown to promote GC angiogenesis through VEGFR-2 signaling pathway (39). To our knowledge, no previous study has linked ENTPD8, RNASE1, HABP2, or ENTPD2 to GC prognosis. While ENTPD8 has been implicated in pancreatic cancer cell proliferation through cytidine metabolism regulation (18), its role in GC remains unexplored. RNASE1, primarily studied in cardiovascular and systemic inflammation (23), has not been previously associated with GC outcomes. HABP2 has diagnostic and prognostic value in thyroid and endometrial cancers (33,34), but its function in GC is unknown. ENTPD2, identified as a poor prognostic factor in hepatocellular carcinoma through modulation of myeloid-derived suppressor cells (35), has not been investigated in GC. Our study is the first to identify these four genes as components of a CCR-related prognostic signature in GC, representing a novel finding distinct from previous CCR-focused studies that have primarily examined established chemokine axes (e.g., CXCL12/CXCR4) or individual CCR members. While several studies have constructed prognostic signatures involving chemokines or immune-related genes in GC, our study differs in several key aspects. First, we performed consensus clustering of the entire CCR family (62 genes) to establish molecular subtypes, whereas previous studies typically used single-gene analyses or small gene sets. Second, our LASSO-Cox signature was derived from subtype-specific DEGs, ensuring biological relevance to CCR-mediated TME remodeling. Third, we identified four previously unreported GC prognostic genes (ENTPD8, RNASE1, HABP2, ENTPD2), expanding beyond the well-characterized CXCL12/CXCR4 axis. Fourth, we integrated TMB analysis to demonstrate the immunotherapeutic relevance of our signature. Subsequently, we validated the established prognostic model in both TCGA and GEO cohorts. Results demonstrated that elevated risk scores were closely associated with unfavorable OS in GC patients. ROC curves and calibration plots confirmed the high predictive accuracy of this signature for 1-, 3-, and 5-year OS of GC patients. In both TCGA and GEO datasets, univariate and multivariate Cox regression analyses revealed that risk score could serve as an independent prognostic factor for OS. Furthermore, we constructed a nomogram integrating this prognostic signature to predict the OS of GC patients, and verified its favorable predictive performance. Functional analyses suggested that the degree of risk scores was quite relevant to ECM related functions and pathways. Similar to Cluster 2, the high-risk group displayed elevated Immune Score, Stromal Score, and ESTIMATE score, implying increased immune cell infiltration and stromal content, accompanied by reduced tumor purity within the TME. The high-risk group exhibited characteristics consistent with the immune-excluded TME phenotype. TMB serves as a pivotal biomarker for immunotherapy in advanced GC. Patients with high TMB generally respond better to immunotherapy and exhibit more favorable prognosis. In this study, TMB levels were notably higher in the low-risk subgroup, and a significant negative correlation between TMB and risk score was observed (r=−0.34). Stratified survival analysis further demonstrated that the high-TMB/low-risk group had the most favorable prognosis, whereas the low-TMB/high-risk group had the poorest. Moreover, risk score retained prognostic value even within the same TMB stratum, supporting the complementary roles of risk score and TMB in GC prognostic evaluation. Our study has several limitations. First, all data were derived from public databases; further validation using prospective sequencing of clinical samples would strengthen our conclusions. Second, the underlying molecular mechanisms require experimental verification in both in vitro and in vivo models. Third, the four newly identified genes (ENTPD8, RNASE1, HABP2, and ENTPD2) still lack functional and mechanistic evidence to confirm their potential as biomarkers in GC. Conclusions In conclusion, this study represents a comprehensive bioinformatics analysis of CCRs in GC. We propose a CCRs-based molecular stratification approach that identifies two distinct subtypes with different TME characteristics and prognoses. Furthermore, we established and validated a CCRs-related prognostic signature derived from subtype-specific genes for predicting OS in GC patients. This signature was identified as an independent prognostic factor and may serve as a supplementary tool for GC risk assessment. However, given that all analyses were based on retrospective public datasets without prospective cohort verification or experimental validation, the clinical utility of this signature requires further investigation. Future prospective validation with clinical datasets and functional studies on the four novel genes (ENTPD8, RNASE1, HABP2, ENTPD2) are warranted to optimize risk stratification and facilitate precision therapy for GC patients. Supplementary The article’s supplementary files as Acknowledgments We acknowledge the TCGA database for providing their platforms, and thank contributors for their valuable public datasets. 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. Footnotes Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0590/rc Funding: None. Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0590/coif). The authors have no conflicts of interest to declare. References Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-0590/prf Supplementary Materials The article’s supplementary files as