We first analyzed NMRI to predict the prognosis of gastric cancer patients by comparing the differences in survival between high and low NMRI in overall survival (OS), disease-specific survival (DSS), disease-free interval (DFI), and progression-free interval (PFI), respectively, and the results of the survival curves showed that, compared with low NMRI, the high NMRI group had a OS, DSS, DFI, and PFI poorer prognosis (p < 0.05) (Fig. 3A−D). Using the GSE84437 cohort as a validation cohort, we assessed the prognostic predictive validity of NMRI. The findings also indicated that patients in the high NMRI group had a significantly worse prognosis than those in the low NMRI group (p < 0.05), indicating that NMRI may be a more reliable predictor of patient prognosis for gastric cancer (Fig. 3E). NMRI was found to be a risk factor regardless of other clinical features based on the outcomes of univariate and multivariate regression analysis (Fig. 3F−G). Additionally, we calculated the differences in NMRI scores for other common clinical features. The results indicated substantial differences (p < 0.05) in T, N, M, and grade differences (Fig. 3H).

Fig. 3

Association of nucleotide metabolism-related indices with clinical traits and constructed column line graphs. survival curves of the TCGA cohort for overall survival (OS) (A), disease-specific survival (DSS) (B), disease-free interval (DFI) (C) and progression-free interval (PFI) (D). E Survival curves for the GSE84437 cohort. F-G Univariate and multivariate regression analysis. H Differences in NMRI scores between different clinical features (T, M, N and grading). I NMRI ROC curves at 1, 3 and 5 years. J AUC comparisons of 1-, 3-, and 5-year NMRI with other clinical traits. K Column line graphs on NMRI constructs. L Calibration curves for 1-, 3-, and 5-year column line plots. Note * p < 0.05, **p < 0.01, ***p < 0.001

The validity of NMRI prognostic prediction was assessed using the area under the ROC curve (AUC). The 1-year, 3-year, and 5-year AUCs of NMRI prognostic prediction were 0.636, 0.701, and 0.732, respectively. The AUC values of 1-year, 3-year, and 5-year were superior to the other clinical traits for predicting the survival status of patients, indicating that NMRI is a more accurate predictor of the survival status of both short- and long-term gastric cancer patients (Fig. 3I–J). Ultimately, we created column line plots utilizing NMRI and additional clinical characteristics (age, clinical stage, T, N, and M), with NMRI accounting for the majority of the column line plots' overall score (Fig. 3K). When compared to the reference line, the 1-, 3-, and 5-year column line plots demonstrated acceptable prediction accuracy, according to the calibration curves of the column line plots (Fig. 3L). These findings imply that NMRI is a valid and trustworthy method for predicting patients' chances of surviving stomach cancer.

Gene set enrichment analysis and correlation study of NMRI with tumor microenvironment

To explore the cancer signature pathways associated with NMRI, we performed GSEA analysis in the high and low NMRI groups, which showed that the high NMRI group was significantly enriched in ANGIOGENESIS, EPITHELIAL MESENCHYMAL TRANSITION and HYPOXIA signaling pathways, and the low NMRI group was significantly enriched in the DNA REPAIR and OXIDATIVE PHOSPHORYLATION signaling pathways (Fig. 3B). In addition, GO enrichment analysis of the high NMRI group revealed that high NMRI was enriched on multiple immune cell infiltration signaling pathways, such as B CELL MEDIATED IMMUNITY, T CELL MEDIATED IMMUNITY, REGULATION OF B CELL ACTIVATION, REGULATION OF T CELL ACTIVATION and T CELL RECEPTOR COMPLEX signaling pathways (Fig. 3B).

The TME scores (ESTIMATE score, immune score, and stroma score) of the patients in the high-NMRI group were significantly higher than those in the low-NMRI group, while the tumor purity scores were significantly lower than those in the low-NMRI group, indicating that the high-NMRI group had a higher level of immune infiltration (Fig. 4C). We used the ESTIMATE algorithm to assess the immune cell infiltration of the tumor microenvironment in gastric cancer patients. Using seven software programs, including XCELL, we examined the relationship between NMRI and immune cell infiltration. The results showed that NMRI and the majority of immune cells had a positive connection (Fig. 4D). The CIBERSORT algorithm's results demonstrated that while the amount of immunosuppressive M2-type macrophages was much lower in the high NMRI group, the level of immunostimulated CD8 T cells was significantly greater in the high NMRI group than in the low NMRI group (Fig. 4E). Furthermore, patients in the high-NMRI group had higher immune cell infiltration and immune-related activities than those in the low-NMRI group, according to the results of the ssGSEA algorithm used to assess these data (Fig. 4F). Taken together, the findings imply that patients with high NMRI values could also have significant levels of immune infiltration in stomach cancer.

Fig. 4

Correlation analysis of GSEA and tumor microenvironment. A-B GSEA analysis of patients in the high NMRI group. C Comparison of tumor purity, ESTIMATE score, immune score and stromal score of patients in the high/low NMRI group. D Seven software analyses of NMRI correlating with various immune cell infiltration levels. E The CIBERSORT algorithm compares the differences in immune cell infiltration levels between high/low NMRI groups. F The ssGSEA algorithm analyzes differences in immune cell infiltration and immune-related functions between high/low NMRI groups

Association of NMRI with immunotherapy efficacy and patient response

We compared the expression levels of common immune checkpoints (immunosuppressive and immunostimulatory genes), MHC molecules, cytokines, and cytokine receptors between the high and low NMRI groups in order to investigate the relationship between NMRI and the immune microenvironment further. The results indicated that the expression levels of the majority of the aforementioned genes were significantly higher in the high NMRI group than in the low NMRI group (Fig. 5A−E). The association between NMRI and genes that stimulate the immune system, genes that repress the immune system, MHC molecules, cytokines, and cytokine receptors was next examined. The findings indicated that the majority of the genes had a substantial and positive connection with NMRI (p < 0.05, Fig. 5F). These findings imply that immunotherapy was more effectively received by patients in the high NMRI group.

Fig. 5

Application of NMRI in immunotherapy response. Differences in the expression levels of immunosuppressive genes (A), MHC molecules (B), cytokine receptors (C), immunostimulatory genes (D), and cytokines (E) between gastric cancer patients in high/low NMRI groups. F Heatmap of correlation between NMRI and immunostimulatory genes, immunosuppressive genes, MHC molecules, cytokines and cytokine receptors. G Differences in TIDE scores between gastric cancer patients in the high/low NMRI group. H-J Differences in IPS scores between gastric cancer patients in the high/low NMRI group. K External immunotherapy dataset Imvigor210 validates VMRI for immunotherapy effect prediction

According to the present study, patients with lower TIDE scores are more likely to benefit from immunotherapy, and TIDE and IPS scores can be used to evaluate a patient's response to immunotherapy [14]. The study revealed that the high NMRI group's TIDE scores were considerably higher than the low NMRI group's, indicating that the immune checkpoint inhibitor medication was more effective in the high NMRI group (Fig. 5G). Further investigation revealed that the immunophenotypic core (IPS) scores of the high NMRI group were significantly higher (p < 0.05) than those of the low NMRI group, indicating that the patients in the high NMRI group may be more responsive to immunotherapy (Fig. 5H−J). Additionally, we examined the relationship between the NMRI group and immunogenicity to predict patients' response to immune checkpoint blockade (anti-PD1 and/or anti-CTLA4). In order to verify the NMRI prediction of immunotherapy effect, we lastly gathered an external real immunotherapy dataset (Imvigor210). The outcomes demonstrated that the patients' NMRI scores in the group responding to anti-PD-L1 immunotherapy were significantly higher than those of the patients in the non-responding group (Fig. 5K). The findings demonstrated that NMRI may predict how well stomach cancer patients will respond to immunotherapy, with higher NMRI patients seeing better immunotherapy outcomes.

NMRI correlation and molecular docking with common drug sensitivity

We examined the reactions of the high/low NMRI group to both standard gastric cancer chemotherapeutic treatments and targeted therapeutic pharmaceuticals in order to inform the clinical usage of medications in these patients. The IC50 of the drugs was shown to be inversely correlated with the patients' sensitivity to the drugs. The outcomes demonstrated that patients with low NMRI had better drug sensitivity, i.e., better therapeutic outcome, to Cisplatin, Gemcitabine, Methotrexate, Metformin, and Gefitinib, while patients with high NMRI had higher drug sensitivity to Pazopanib, Bexarotene, Dasatinib, Imatinib, and Sunitinib (Fig. 6A).

Fig. 6

Application of NMRI in drug sensitivity and molecular docking. Differences in response to common gastric cancer chemotherapeutic drugs (A) and targeted therapeutic drugs (B) between high and low NMRI groups. The Figure shows the docking poses of the SERP INE1 activity pocket with Fenugreekine (C), P ortulacaxanthin II (D), Leucovorin (E), Kuwanon J (F), Blumeatin (G) and Schizotenuin F (H)

For chemical screening, a computer technique based on structure is called molecular docking. Using the PDB database, we were able to retrieve the protein structure of SERPINE1 (ID: 7AQG) for the purpose of molecular docking with small natural molecules. Figure 6C−H displays the top six small molecules (Fenugreekine, Portulacaxanthin II, Leucovorin, Kuwanon J, Blumeatin, and Schizotenuin F) that have the highest affinity for binding to the SERPINE1 binding pocket. As an illustration, Portulacaxanthin II forms hydrogen bonds with residues Gln-123, Thr-120, Met-110, Gly-108, and Leu-105 of the amino acid sequences of SERPINE1, where Gln-123 acts as an acceptor and Thr-120, Met-110, Gly-108, and Leu-105 as donors. Asp-96, His-143, and Arg-118 of the SERPINE1 amino acid residues establish hydrogen bonds with Blumeatin; Asp-96 is a hydrogen bond donor whereas Arg-118 and His-143 are hydrogen bond acceptors.