Header
Header
Article

Metabolism reprogramming signature associated with stromal cells abundance in tumor microenvironment improve prognostic risk classification for gastric cancer | BMC Gastroenterology


GC patients with a higher StromalScore had a poor prognosis

In terms of clinicopathological features, GC patients who were younger (age no more than 65 years old), had died, or had advanced stage disease (T3–4 and N1–3) exhibited elevated stromal scores (Fig. 1A). The high stromal score group had significantly lower overall survival (OS) than the low stromal score group in the TCGA and GSE84437 cohorts (Fig. 1B, C). We obtained 553 intersection genes by identifying the DEMRGs between the low- and high stromal score groups of the two independent cohorts (Fig. 1D).

Fig. 1

Identification of metabolism-related genes associated with the StromalScore. A Association between the StromalScore and clinicopathological features. B Kaplan–Meier survival analysis based on the StromalScore and OS in the TCGA cohort and the heatmap of DEMRGs between the high and low StromalScores in the TCGA cohort. C Kaplan–Meier survival analysis regarding the StromalScore and OS in the GSE84437 cohort and the heatmap of DEMRGs between the high and low StromalScores in the GSE84437 cohort. D Venn plot of intersection genes

The prognosis of GC was associated with the metabolic gene cluster

Using the K-means clustering algorithm, the CDF plot showed that the optimal number of clusters was 2 (Fig. 2A) when the consensus matrix (k) varied from 2 to 9. At this point, the intragroup showed the highest correlation, and the intergroup showed the lowest correlation (Fig. 2B). Therefore, the training cohort was clustered into two subtypes based on the 553 intersecting metabolic genes (Fig. 2C). The OS of subtype A was obviously worse than that of subtype B (Fig. 2D), indicating that these 553 intersection metabolic genes had a significant correlation with the OS of GC.

Fig. 2
figure 2

Metabolic gene cluster analysis. A CDF plot. B Clustering heatmap. C Heatmap of metabolic gene clusters. D Kaplan–Meier survival analysis regarding metabolic gene clusters and OS in the training cohort

A 24-metabolic gene prognostic signature established in the training cohort

Using a p value < 0.05 as a screening criterion, 179 genes were predicted to be prognostic candidate biomarkers by univariate Cox regression analysis (Additional file 3: Table S1). Thirty-seven genes with nonzero LASSO regression coefficients were retained for multivariate Cox regression analysis (Additional file 2: Fig S2A). A model was chosen based on the Akaike information criterion (AIC) [15] using a stepwise algorithm, and 24 genes from the formula were used to calculate the risk score (Additional file 2: Fig S2B, Table 2). A total of 402 patients with a risk score greater than the median value (0.995) were assigned to the high-risk group, which showed significantly reduced overall survival (OS) compared to the 402 patients with a risk score less than the median value (0.995) (Fig. 3A). The area under the curve (AUC) values for the risk score predicting OS at 1, 3 and 5 years were 0.694, 0.711 and 0.743, respectively (Fig. 3B). As the risk score increased, the patients’ risk of death increased (Fig. 3C–E).

Table 2 The gene name and coef
Fig. 3
figure 3

Prognostic assessment of the risk score in the training cohort. A Kaplan–Meier survival analysis regarding risk score and OS in the training cohort. B Time-dependent ROC analysis of the risk score predicting the OS of patients in the training cohort. CE Heatmap, risk score distribution and survival status of patients in the training cohort

The risk score was an independent prognostic indicator for GC

By analyzing 756 cases with complete clinical data in the training cohort, we found that risk score, T stage and N stage were independent prognostic indicators in univariate and multivariate Cox regression analysis (Fig. 4A, B). Next, the patients were divided into 11 subgroups for verification according to age, sex, T stage and N stage. We found that the OS of the high-risk group was significantly lower than that of the low-risk group in each subgroup (Fig. 4C), indicating that the risk score has universal applicability to the prognosis classification of GC patients.

Fig. 4
figure 4

Clinical subgroup validation of the prognostic risk score. A The forest plot of the univariate Cox analysis. B The forest plot of the multivariate Cox analysis. C Clinical subgroup survival analysis

Internal validation of the prognostic signature in the TCGA and GSE84437 cohorts

In the TCGA and GSE84437 cohorts, the high-risk patients showed significantly worse OS than the low-risk patients (Fig. 5A, C). The AUC values of the risk score predicting the OS of GC patients in the TCGA cohort at 1, 3 and 5 years were 0.668, 0.717 and 0.709, respectively (Fig. 5B). The AUC values of the risk score predicting the OS of GC patients in the GSE84437 cohort at 1, 3 and 5 years were 0.732, 0.719 and 0.761, respectively (Fig. 5D).

Fig. 5
figure 5

Internal validation of the risk score in the TCGA and GSE84437 cohorts. A, B Kaplan–Meier survival analysis and time-dependent ROC analysis of the signature for predicting the OS of patients in the TCGA cohort. C, D Kaplan–Meier survival analysis and time-dependent ROC analysis of the signature predicting the OS of patients in the GSE84437 cohort

External validation of the prognostic signature in the GSE15459 and GSE62254 cohorts

Compared to the high-risk group, the low-risk group had better clinical outcomes in the GSE15459 and GSE62254 cohorts (Fig. 6A, D). As suggested by the univariate and multivariate Cox regression analyses, the risk score independently predicted the OS of GC patients (Fig. 6B, C, E, F).

Fig. 6
figure 6

External validation of the risk score in the GSE15459 and GSE62254 cohorts. AC Kaplan–Meier survival analysis, the forest plot of the univariate Cox analysis and the multivariate Cox analysis in the GSE15459 cohort. DF Kaplan–Meier survival analysis, the forest plot of the univariate Cox analysis and the multivariate Cox analysis in the GSE62254 cohort

Exploration of the relationship between immune cell infiltration characterization and the prognostic signature

To determine the underlying mechanism of the prognostic signature, we identified differentially expressed genes (DEGs) between the high- and low-risk groups (Fig. 7A). GO term annotation showed that these DEGs were mainly involved in immune response-related biological processes (Fig. 7B). Thus, we hypothesized that the metabolic reprogramming signature may reshape the immune microenvironment of GC and attempted to investigate the association between the signature and immune cell infiltration (ICI). Patients were clustered into four subtypes according to the quantification of 22 kinds of ICIs. In terms of immune infiltration characteristics, the infiltration proportion of naïve B cells, resting memory CD4 T cells, and resting mast cells in ICI-A was the highest; ICI-B was accompanied by a large number of regulatory T cells (Tregs) and infiltrating M0 macrophages. The proportion of M1 macrophages and CD8 T cells in ICI-C was significantly higher than that in the other three subtypes. ICI-D had the highest infiltration of plasma cells and the lowest infiltration of Tregs (Fig. 7C). Interestingly, the expression levels of signature-related genes varied among different ICI subtypes (Fig. 7D), indicating that the metabolic pattern of different ICI subtypes also changed. Four ICI subtypes were distributed in both the high-risk and low-risk groups (Fig. 7E). Among them, the prognosis of ICI-C was significantly better than that of the other three subtypes (Fig. 7F). Furthermore, the risk score of ICI-C was significantly lower than that of the other three subtypes (Fig. 7G). It is worth mentioning that the ICI-C subtype accounted for 24% of the low-risk group, which was twice as high as that in the high-risk group (Fig. 7H).

Fig. 7
figure 7

Relationship between the risk score and immune cell infiltration (ICI) characterization. A Heatmap of DEGs between high- and low-risk groups. B GO term annotation of DEGs. C Boxplot of the four ICI clusters. D Distribution of four ICI clusters in high- and low-risk groups. E Distribution of ICI clusters in different risk groups F Kaplan–Meier survival analysis of different ICI clusters. G Boxplot of the risk score difference among the four ICI clusters. H Bar plot of proportions of the four ICI clusters in high- and low-risk groups



Source link

Back to top button