Comprehensive analysis of T-cell regulatory factors and tumor immune microenvironment in stomach adenocarcinoma

Among the included 37 T cell-related genes, the copy number variations (CNV) were investigated (Fig. 1a). Notably, significant increases in CNV were observed in genes such as LIG3, ZNF830, ATF6B, CLIC1, and DUPD1, while decreases were most obvious in HOMER1, DCLRE1B, B2M, MRPL51, and LTBR. The distribution profile of CNV modification in the 37 T cell-related genes is illustrated in Fig. 1b. To investigate the association between genes and their interactions in STAD studies, we generated a T cell-related co-expression visualization protein-protein interaction (PPI) network (Fig. 1c).

The proportions of each T-cell subset were not significantly correlated, as shown in Fig. 1d. The population quantities with remarkably positive relevance were CD8 + T cells and CD4 + memory-activated T cells (0.5), activated neutrophils and mast cells (0.45), and resting mast cells and activated natural killer cells (0.36). As shown in Fig. 1e, the levels of macrophage M1, T-cell follicular helper, macrophage M0, and T-cell CD4 memory activation were comparatively high in the tumor samples contained in the heatmap. In the PPI network, key gene pathways such as B2M and HLA have multiple functions and play a central role in co-expression and physical interactions.

The intercept value was set to 76 to detect outliers, the outlier samples were removed, and the remaining samples were included in the analysis (Additional figure S1a). As shown in Additional figure S1b, when power = 5, the scale independence was 0.9, and the mean connectivity was relatively high. Power = 5 was set to build the co-expression module and obtain the result of the preliminary module division. Different modules were represented in different colors in line with the results of WGCNA (Additional figure S1c). To detect outliers, a tree was built using the eigenvalues of the module and very close distance module merging, with the intercept value set to 0.5 (Additonal figure S1d). As shown in Additional figure S1e, a co-expression module was constructed, and the results were obtained after merging similar modules. According to the characteristic value of each sample in each module, correlation analysis was carried out to find out two modules related to specific traits (infiltrated immune cells). Among the seven modules, the green module was highly correlated to T cells CD8 (CD8 + T cells) (R2 = 0.3; p = 2e-08) and activated T cells CD4 memory (R2 = 0.24; p = 5e-06). Furthermore, the green-yellow module showed a higher correlation with activated T cells CD4 memory (R2 = 0.34; p = 8e-11) and T cells CD8 (CD8 + T cells) (R2 = 0.29; p = 6e-08; Fig. 1f).

From the chart, we identified five genes (IFNL2, IL12B, B2M, HLA-A, and CD19) that were selected from the intersection of the T-cell regulatory factor and the WGCNA gene (Fig. 2a), with only four (IL12B, B2M, HLA-A, and CD19) expressed in selected samples. Among the T-cell regulatory factors, normal and tumor genes showed prominently distinct Inter-cell Interference Coordination (ICIC), with tumor gene expression higher than normal in RAN, CHK1, and CDK2 cells (Fig. 2b). Prognostic analysis indicated a significant survival advantage for CD19 and IL12B (Fig. 2c, d). The distribution of genes encoding T-cell regulatory factors is shown in Fig. 2e. As shown in Fig. 2f, B2M and HLA − A exhibited a high correlation, as did IL12B and CD19.

In the union of the GSE38749, GSE84437, GSE34942, GSE15459, and TCGA-STAD cohorts, the STAD samples were grouped into diverse molecular subtypes using NMF analysis (Fig. 3a and b). We designated these two clusters as T cell-related genes: Clusters C1 and C2. Cluster C1 displayed marked indigenous survival advantages (Fig. 3c). Furthermore, these two clusters also displayed notably diverse enrichment characteristics of the KEGG pathway in GSVA enrichment analysis. For instance, in KEGG intestinal immune network for IgA production, the KEGG T-cell receptor signaling pathway and KEGG B cell receptor signaling pathway had high-level gene expression in C1 and low-level gene expression in C2, while KEGG basal transcription factors had high expression levels in C2 and low expression levels in C1 (Fig. 3d). These two T-cell clusters also displayed marked distinctions in ICIC (Fig. 3e). Activated CD8 T-cells, activated B cells, and macrophages were amplified in cluster C1, suggesting that they could promote inflammation, while activated CD4 T-cells, type 17 T-helper cells, and neutrophils were abundant in another cluster. The results indicated that the expression of major histocompatibility complex type 1 (MHC-1) in the C1 cluster was higher than that in the other clusters, which may be one of the reasons why the survival rate of CI was higher than that of the other clusters. As shown in Fig. 3f, the genes were divided into two clusters, C1 and C2, proving that the two groups clustered well.

A total of 178 genes with an absolute value of LogFC greater than 1.5 and p < 0.05 were obtained by analyzing the differences between the C1 and C2 groups. To investigate the association between T cell-related genes and other illnesses, DO analysis was carried out, and the results showed that all genes were strongly correlated with bacterial infectious diseases, leukocyte diseases, and systemic mastocytosis (Fig. 4a). For each T cell-related gene cluster, a deeper understanding of the characteristics and GO functional enrichment analysis was performed using the cluster package. These genes showed abnormal enrichment related to biological processes, molecular functions, and cellular components, which could also partially account for the high incidence and recurrence rates of malignant gastric cancer (Fig. 4b). Accumulation of the biological processes was studied by carrying out KEGG analysis, confirming T-cellular receptor signaling transduction pathway, Th1, Th2, and Th17 cell transdifferentiation (Fig. 4c, d), all of which are related to T-cells and the immune microenvironment.

LASSO and multivariate Cox regression analyses were employed, and 13 genes were selected from 178 genes by LASSO to construct signatures; six core genes (CD5, ABCA8, SERPINE2, ESM1, SERPINA5, and NMU) were screened from the DEGs to build the risk signature (Fig. 5a, b). Patients with STAD (n = 1043) were stochastically separated into a training group (n = 522) and a test group (n = 521) using the caret R package, in which the training group was employed to construct signatures. After the multivariate Cox regression analysis, the process of constructing the risk score was calculated as follows:

Risk score = (-0.3081*expression of CD5) + (0.18156*expression of ABCA8) +.

(0.2309*SERPINE2) + (0.2112*expression of ESM1) + (0.0970*SERPINA5) + (0.1116*NMU).

The outcomes of the risk-scoring system applied to all patients showed significant diversity. Patients with STAD and risk scores below the average standard in the gene cluster were assigned to the low-risk group (n = 549), whereas those exceeding the average standard were assigned to the high-risk group (n = 494). Overall, the risk score of patients with STAD in the T cell C2 group was higher than that in the other groups (Fig. 5c).

A heatmap was created, highlighting that tumor, node, and metastasis staging was elevated in the high-risk group (Fig. 5d). This conclusion also confirmed that C1 had a significantly better prognosis than C2. Gene expression in the two groups was visually analyzed to thoroughly investigate the association between the risk score and other parameters (Fig. 5e). When TCGA and GEO cohorts were merged, log2 was used for data with large values to eliminate batch effects (Additional figure S1f). Survival analysis was carried out in the training and test groups and the initial merged cohort to guarantee that the outcome of the obtained risk signature’s predictive ability could be confirmed. Similar survival advantage results were observed across all analyzed groups (Fig. 6a-c). In the low-risk group, the expression of T cell-related genes was relatively high, whereas in the other groups, the expression was low. The distribution of risk scores demonstrated that a lower risk score was associated with a higher survival probability and that the survival rate decreased with an increase in the risk score. From the distribution risk score curve, affordable T cell-related genes such as CD5 were higher in the low-risk group, whereas the inverse outcome was observed in the high-risk group. ABCA8, SERPINE2, ESM1, SERPINE2, and NMU expression levels were higher in the high-risk group, whereas an inverse outcome was noted in the low-risk group. The six core genes were distributed in the low- and high-risk groups within the training, testing, and total samples using heat maps. The probability of survival decreased as the risk score increased (Fig. 6d-f). To verify the veracity of the risk signature, we generated ROC curves (Fig. 6g-i). Univariate and multivariate Cox regression analyses of the combined cohort also verified that the risk characteristics could be independently applied as a prognostic signature for STAD (Fig. 6j, k). The 1-, 3-, and 5-year survival rates of patients could also be predicted using a contingency map containing risk scores and clinicopathological parameters (Fig. 6l). Based on the calibration chart, the previous line chart had characteristics similar to those of the calibration plot (Fig. 6m). We analyzed the predictive effects of univariate and multivariate Cox regression analyses on individual cohorts (GSE15459, GSE34942 + GSE38749, GSE84437, and TCGA) (Fig. 6n). A survival analysis was performed to test the predictive effect of the signature on individual cohorts. In all cohorts, survival decreased with increasing risk scores (Fig. 6o).

The correlation between the clusters and risk was visualized using a scatter diagram and heatmap (Fig. 7a). The results showed mast cells, monocytes, and type II interferon responses were positively linked to the risk score. T-cell inhibition, activated CD8 + T cells, activated B cells, and other immune cells were significantly negatively correlated. The matrix and estimate scores in the low-risk group were lower; however, there was no significant difference between the two groups (Fig. 7b). The interstitial and estimate scores showed marked variation between the low- and high-risk subgroups (Fig. 7c). The survival ratio and degree of purity of tumor cells in the high-risk group were significantly higher than those in the low-risk group. In STAD, the comprehensive risk score and cancer stem cell (CSC) index values were used to comprehensively assess the link between risk markers and CSC (Fig. 7d). The study displayed that risk score was positively associated with the CSC index (R = 0.28; p < 2.2e-16), demonstrating that the higher the risk score, the more obvious the differentiation grade of the stem cells. Subsequent analysis to explore the immune infiltration condition of T cell-related core genes was performed by studying the distribution disparity of somatic cell mutations between the low- and high-risk score groups using the maftools package (Fig. 8a, b). We performed a survival analysis of TMB and risk scores in different groups to forecast the prognosis of patients with STAD, and the outcome illustrated that the low-risk score and high-TMB groups had the highest survival rates (Fig. 8c, d). We also assessed the degree of association between TMB and the risk score, as well as between T-cell clusters and gene clusters (Fig. 8e). The results further demonstrated that low-risk scores correlated with high TMB. In addition, the TMB of clusters C2 and B were higher than those of the other two clusters. The low-risk group, compared to the high-risk group with a higher mutation load, was more extensive, which was consistent with the above findings and confirmed by subsequent TMB quantitative analysis (Fig. 8f). We also analyzed microsatellite instability (MSI) in the low- and high-risk subgroups to further assess the capacity of risk markers to predict ICB responses in patients (Fig. 8g). The rate of MSI was higher in the low-risk group than in the high-risk group, suggesting that immunotherapy and clinical therapy were more effective in this subgroup. The association between the risk characteristics and MSI was further confirmed (Fig. 8h).

The high-risk group exhibited a high immune escape rate and poor immunotherapeutic efficacy (Fig. 9a). Through the analysis of immune cell differentiation, it can be seen that the immune expression of the PDCD1 and CD274 high-risk group was low. Based on the data obtained, PD1 and PD-L1 were not very effective in the immune treatment (Fig. 9b). According to the characteristic value of the sample in each module and the characteristics of the sample, correlation analysis was performed to identify two modules related to specific traits: KEGG T-cell receptor signaling pathway and KEGG B-cell receptor signaling pathway were positively correlated in CD8 T cells (Fig. 9c). The treatment effect was observed using survival analysis in each group. The CR/PR group exhibited the highest probability of survival (Fig. 9d). Estimate analysis revealed a significant gap between the matrix and estimate score between the low- and high-risk subgroups (Fig. 9e-h). The results indicated that the survival rate of tumor cells in the high-risk group was significantly higher than that in the low-risk group, accompanied by poorer immune efficacy. The half-maximal inhibitory concentrations (IC50) of paclitaxel, gemcitabine, 5-fluorouracil, and doxorubicin were all higher in the high-risk groups, indicating potential drug resistance (Fig. 9i-l).

Survival analyses were performed on the external IMV210 and GSE62254 cohorts to confirm the predictive power of this risk marker in validating the survival advantage of the low-risk group. Similar results were observed in all pooled analyses (Fig. 10a, b). The binary response system showed that the SD/PD risk scores were significantly higher than those of CR/PR (Fig. 10c).

We compared our T cell-related gene signature with other published signatures [7, 26‐33], which yielded a C-index of 0.613, which was higher than other signatures published within the last 3 years (Fig. 10d). According to restricted mean survival, it also achieved a higher accuracy of survival estimation in the validation datasets (HR: 2.361; p < 0.001; Fig. 10e). ROC curves and survival analyses were generated for comparison (Additional figure S2).

After quality control, 54,274 cells were labeled to show their distribution (Fig. 10f). Enrichment analysis suggested that the T-cell content in tumors was significantly higher compared to normal tissues (Fig. 10g). Furthermore, the T-cell subpopulations were divided into gamma delta, CD4+, NK, and CD8 + T cells. SERPINE2 was highly expressed in tumors, mainly in CD4 + T cells and NK cells (Fig. 10h, i). The signature score was higher for CD4 + T cells and NK cells of tumors and lower for gamma delta T cells (Fig. 10j).

Based on immunohistochemistry, the Youjiang cohort was divided into Low SERPINE2 (n = 44) and High SERPINE2 (n = 49) groups, and samples with lower SERPINE2 expression exhibited better OS (HR = 3.197; p = 0.007; Fig. 11a-d). As shown in Fig. 11e, the SERPIN2 and CXCL12 levels were significantly correlated.

To investigate the effects of CXCL12 in vitro, we first validated the downregulation of CXCL12 expression in AGS and BCG-823 cells (Fig. 11f). According to the results of the CCK-8 assay, the knockdown of CXCL12 significantly reduced the proliferative viability of AGS and BCG-823 cells compared to the control group (Fig. 11g). Furthermore, the transwell migration assay indicated that the knockdown of CXCL12 expression markedly suppressed the metastatic ability of AGS and BCG-823 cells (Fig. 11h). Overall, the downregulation of CXCL12 significantly inhibited the proliferation and migration of AGS and BCG-823 cells.

To determine the effect of CXCL12 on patient survival, we analyzed the survival of patients with different CXCL12 expression levels. Survival analysis showed that patients with higher CXCL12 expression had worse OS than those with lower CXCL12 expression (Fig. 11i). These data suggest that the analysis of CXCL12 expression yields different subtypes of GCs. Specifically, our results suggest that lower CXCL12 expression levels are associated with improved survival in patients with GC.