Landscape of the genetic variation of NMRGs in GC

Figure1 presents the workflow of the study. Herein, 97 NMRGs were evaluated to explore their roles in GC. First, 97 NMRGs in GC were examined for copy number variations (CNVs) and somatic mutations (Supplementary Fig.1A), with mutations identified in 161 of the 433 samples (37.18%). DPYD and XDH showed the highest mutation rate (5%) followed by CAD, AMPD3 and AK9 (4%). Furthermore, ENTPD8, ENTPD2, DNPH1, UCK1AK8 and AK1 exhibited higher frequencies of CNV amplification, whereas DCTD, IMPDH1, CDA, DPYD and AK6 exhibited higher probabilities of CNV deletions (Supplementary Fig.1B). Supplementary Fig.1C shows the chromosomal positions of the aforementioned CNVs. To determine the relationship between genetic variation and NMRG expression, we also compared the expression levels of 97 NMRGs between normal and tumour samples. A total of 77 genes were differentially expressed (Supplementary Fig.1D).

The flow chart of the study design.

To further explore the potential association between NM and GC, fresh serum samples, consisting of 33 patients with GC and 27 healthy volunteers, were collected for metabolomic analysis. Using the limma package in R, a total of 18 differentially expressed nucleotide metabolites were identified. Among them, 1-Methyladenosine, 1-Methylguanosine, 7-Methylguanine, Allantoic acid, Cytidine, Dihydrothymine, Inosine, N2, N2-Dimethylguanosine, Pseudouridine, Uracil, Ureidopropionic acid, Uric acid and Xanthine were downregulated, whereas 5-Methylthioadenosine, 5-Methyluridine (Ribothymidine), Allantoin, N6-Methyladenosine and Uridine were upregulated in GC samples. These findings highlighted the metabolic reprogramming of NM in patients with GC (Supplementary Fig.2).

To construct an NM prognostic model, we first performed a univariate Cox survival analysis on 77 differentially expressed NMRGs, of which six were statistically significant (Supplementary Table S3). Additionally, the prognostic significance of the six genes was validated using KM analysis (Supplementary Fig.3A). Furthermore, a heatmap of the expression of the six genes in tumour and normal tissues was also drawn (Fig.2A). We then subjected the six genes to multivariate cox analysis (Fig.2B) and correlation coefficients were calculated to construct a model (Supplementary Table S4). The NM score was calculated for each patient, and the patients were classified into high and low score groups based on the median value. The KM curve showed that the high-risk patients had a worse prognosis (Fig.2C). Regarding the TME prognostic model, a high infiltration of activated CD4 memory-activated T cells, CD8 T cells and activated dendritic cells (DCs) were observed to be associated with a better prognosis for patients with GC (Supplementary Fig.3B). Similarly, these cells were subjected to multivariate cox analysis (Fig.2D) and correlation coefficients were calculated to construct a model (Supplementary Table S5). The KM curve showed that high-TME score samples had a better survival prognosis than those with low-TME scores (Fig.2E). GSEA revealed that the high NM score group was mainly enriched in cancer-related and classical oncogenic pathways, while the high TME score group was mainly enriched in immune-related pathways. (Supplementary Fig.3C,D).

Construction of the NM- and TME-related prognostic model. (A) Expression levels of the six model genes. (B) Multivariate cox regression analysis of NM model genes. (C) KaplanMeier (KM) curves of NM-related prognostic model. (D) Multivariate cox regression analysis of three TME cells. (E) KM curves of TME-related prognostic model. NM nucleotide metabolism, TME tumour microenvironment.

First, we investigated the correlation between the six NM model genes and the three TME cells. We found that T cells CD8 were negatively correlated with UPP1, ENTPD2, NT5E and positively correlated with DPYS and AK1; T cells CD4 memory activated were negatively correlated with AK5, ENTPD2, NT5E, DPYS, AK1; dendritic cells were negatively correlated with AK5, ENTPD2, DPYS, and positively correlated with UPP1 (Fig.3A). To further explore their association, we downloaded single-cell data from the GEO database, comprising 10 GC samples. The clustering and annotated results are presented in Fig.3B. Subsequently, we calculated the NM scores in different cell types and found that the NM scores were significantly higher in monocytes and endothelial cells than in B cells, T cells, CD8+ T cells, epithelial, macrophages, Tregs and mast cells (Fig.3C,D). Based on the NM score, monocytes and endothelial cells were divided into low NM score, medium NM score and high NM score monocytes and endothelial cells for cell communication analysis. The monocytes and endothelial cells with low NM scores had more abundant communication with other immune cells (Fig.3EH). Therefore, low NM scores could have a synergistic effect with high TME scores and combining the NM model with the TME model may be a feasible method.

Correlation between the NM scores and TME cells. (A) The correlation between NMRGs and TME cells. (B) t-SNE plot of 10 gastric cancer samples. (C,D) Distribution of NM scores in different cell types. (E,F) The inferred signalling networks between different cell clusters. The significantly related ligandreceptor interactions of (G) NMlowMonocytes and (H) NMlowEndothelial cells. NM nucleotide metabolism, TME tumour microenvironment, NMRGs nucleotide metabolism-related genes.

Next, we constructed the NM-TME classifier by combining the NM and TME scores. It divided patients with GC into four categories: NMhigh/TMEhigh, NMhigh/TMElow, NMlow/TMEhigh and NMlow/TMElow. Survival analysis revealed that the NMhigh/TMElow group had a poorer prognosis while the NMlow/TMEhigh group had a better prognosis among the groups (Fig.4A). Patients in the NMhigh/TME high and NMlow/TME low subgroups showed less divergent prognoses. As a result, we combined them to form a mixed subgroup (Fig.4B). Additionally, the area under the curve (AUC) values of the NM-TME classifier were 0.732, 0.708, 0.702 and 0.807 for 1, 3, 5 and 7years, respectively (Fig.4C), indicating that the NM-TME classifier plays a significant role in the survival prediction of patients with GC.

Construction of the NM-TME classifier and functional enrichment analysis. (A) Survival analysis of the four subgroups was obtained based on the NM-TME classifier. (B) Survival analysis after merging the NMlow/TMElow and NMhigh/TMEhigh subgroups. (C) Receiver operating characteristic (ROC) curve of the NM-TME classifier. (D) Functional enrichment analysis of the three subgroups was obtained based on the NM-TME classifier. NM nucleotide metabolism, TME tumour microenvironment.

Furthermore, we also verified the prognostic significance of the NM-TME classifier in the GEO cohort, which revealed significant prognostic differences between the groups (Supplementary Fig.4A). Moreover, the evaluation of the predictive performance of the classifier under different clinical features in the TCGA cohort revealed good predictive performance (Supplementary Fig.4B).

Functional enrichment of the three groups revealed that the NMhigh/TMElow group was mainly enriched in the regulation of the olefinic compound metabolic process, endothelial cell differentiation and stem cell proliferation, while the NMlow/TMEhigh was majorly positively associated with the positive regulation of T cell migration and negatively associated with the canonical Wnt signalling pathway (Fig.4D).

Furthermore, WGCNA identified four modules (Fig.5A,B). Among them, the turquoise module was most relevant and opposite to each other for the NMlow/TMEhigh and NMhigh/TMElow groups. Therefore, the turquoise module gene could be associated with significantly different prognoses between the NMlow/TMEhigh and NMhigh/TMElow groups. Using the Metascape database, enrichment analysis of these genes revealed that they were mainly enriched in vasculature development, NABA core matrisome and extracellular matrix organization (Fig.5C).

Exploring key module eigengenes associated with the NMlow/TMEhigh and NMlow/TMElow groups using weighted gene co-expression network analysis. (A) Evaluation of the scale-free fit index for differing soft-thresholding powers () and examination of the connectivity of various soft-thresholding powers. (B) A heatmap depicts the association between module eigengenes and various subgroups. (C) Functional enrichment analysis of key module eigengenes. NM nucleotide metabolism, TME tumour microenvironment.

First, we compared the abundance of immune cell infiltration between the different groups. The immune cell infiltration was more abundant in the NMlow/TMEhigh group, especially CD8 T cells, Th1 cells, NK cells, CD4 T cells and macrophages (Fig.6A). Notably, the better prognosis in the NMlow/TMEhigh group could be attributed to the abundant immune cell infiltration. Meanwhile, we also explored whether the expression of common ICGs differed between the groups. Most ICGs were differentially expressed between the groups, with high expression observed in the NMlow/TMEhigh group (Fig.6B). These differentially expressed ICGs could be potential therapeutic targets. Additionally, it also suggests that NMlow/TMEhigh patients may benefit more from immune checkpoint blockade (ICB) therapy. HLA is a polygenic and polymorphic complex involved in antigen presentation43. Figure6C shows that HLA-B, HLA-C, HLA-F and HLA-DOB were expressed the highest in the NMlow/TMEhigh group.

Immune status of different subgroups based on the NM-TME classifier. (A) Differences in immune cell infiltration. (B) Differences in ICGs. (C) Differences in antigen presentation-related genes in different subgroups. NM nucleotide metabolism, TME tumour microenvironment, ICG immune checkpoint gene.

Numerous studies have demonstrated the association between somatic mutations in tumour genomes and the response to immunotherapy44. We therefore examined the TMB distributions among the various groups based on the NM-TME classifier. The NMlow/TMEhigh group had a higher TMB, while the NMhigh/TMElow group had a lower TMB, indicating that the NMlow/TMEhigh group may benefit more from immunotherapy (Fig.7A). Additionally, the NMhigh/TMElow/TMBhigh group had a lower prognosis than patients in the other groups (Fig.7B). Figure7C,D display the top 20 genes with high mutation frequencies in the NMlow/TMEhigh and NMhigh/TMElow groups.

TMB analysis. (A) Comparison of TMB among the defined subgroups. (B) Survival analysis based on the NM-TME classifier and TMB. The top 20 mutation genes of the (C) NMhigh/TMElow and (D) NMlow/TMEhigh groups. NM nucleotide metabolism, TME tumour microenvironment, TMB tumour mutation burden.

Considering that drugs targeting PD-1 and CTLA-4 have recently received approval for the treatment of several cancers, we evaluated whether the NM-TME classifier could predict patients reactions to immunotherapy. The patients in the NMlow/TMEhigh group were observed to have a better response rate to immunotherapy than the other two groups (Fig.8A). Microsatellite instability-high (MSI-H) is a potential predictor of immunotherapy response targeting PD-1 or its ligand PD-L145. Accordingly, the proportion of MSI-H in the NMlow/TMEhigh group was higher than that in the other two groups (Fig.8B). Additionally, we investigated the relationship between the NM-TME classifier and IPS in patients with GC to predict the response to ICIs. Figure8CF presents the differences in the results of CTLA-4/PD-1 inhibitor treatment between the NMlow/TMEhigh and Nmhigh/TMElow groups. The NMlow/TMEhigh group has higher IPS scores, implying more immunogenicity in the NMlow/TMEhigh group. Furthermore, we performed a difference analysis between the immunotherapy-responsive and non-responsive groups and also the NMlow/TMEhigh and NMhigh/TMElow groups. DEGs were then analysed using the Proteomaps 2.0 database46. Notably, the pattern of proteomap in the NMlow/TMEhigh group and immunotherapy-responsive groups were similar (Fig.8G,H). These findings suggest that the NM-TME classifier can be used to predict patients responses to immunotherapy.

The role of NM-TME classifier in immunotherapy. (A) Proportion of response to immunotherapy in different groups. (B) Proportion of MSI in different groups. (CF) Comparison of the relative distribution of IPS across groups with high NM/low TME and low NM/high TME. Functional analysis in the NMlow/TMEhigh group (G) and responder of patients under immunotherapy (H) illustrated using Proteomaps. A little polygon represents a unique KEGG pathway. NM nucleotide metabolism, TME tumour microenvironment, MSI microsatellite instability, IPS immunophenoscore.

Given that targeted therapy is an effective approach in the treatment of GC, it has important clinical applications and prospects. We, therefore, investigated whether the NM-TME classifier could predict drug sensitivity in patients with GC. The NMhigh/TMElow group benefited more from Imatinib, Midostaurin and OSI-906 (Linsitinib), while those in the NMlow/TMEhigh group benefited more from Paclitaxel, Methotrexate and Camptothecin (Supplementary Fig.5AF).

Univariate and multivariate Cox regression analyses indicated that the NM-TME classifier was an independent predictor of prognosis with the highest hazard ratio (HR) (Fig.9A,B). Following this, the NM-TME classifier and clinical features were combined to construct a nomogram. To predict the survival of patients with GC over 1 to 5years, the values of each variable can be added to obtain the total score (Fig.9C). Moreover, the AUC values of the nomogram for 1-, 3- and 5-year OS were 0.826, 0.841 and 0.822, respectively (Fig.9D).

Construction of a nomogram. (A,B) Forest map of univariable and multivariable Cox regression in the test cohort. (C) Nomogram based on the NM-TME classifiers and clinical features. (D) Receiver operating characteristic (ROC) curves of the nomogram model in predicting the 15years survival rate. NM nucleotide metabolism, TME tumour microenvironment.

Follow this link:
The combined signatures of the tumour microenvironment and ... - Nature.com