Bacterial lipopolysaccharide related genes signature as potential biomarker for prognosis and immune treatment in gastric cancer – Scientific Reports

LPS-related genes and function analysis

In differential expression analysis, the DEGs were screened (Fig. 2A). The volcano plot showed the 853 DEGs. During the comparison between tumor samples and normal tissue samples, we found that 617 genes were up-regulated and 236 genes were down regulated (Fig. 2B). Figure 2C, D displayed the GO terms and KEGG pathways. Through functional enrichment analysis, we could see that the most relevant signal pathway of LPS-related genes was “cytokine cytokine receptor interaction”. The most enriched term in biological process²⁸ was molecular function¹⁶, and the enriched terms in cellular component (CC) were “extracellular matrix organization”, “receptor ligand activity”, and “collagen-containing extracellular matrix”, respectively.

Through WGCNA analysis of candidate genes, we extracted LPS related-genes. Through the application of scale-free network, we get that the optimal soft-thresholding power is 5 (Fig. 3A). We established a dendrogram of 853 co-expressed genes identified by DEG in the module (Fig. 3B). On this basis, we identified seven modules according to the average linkage hierarchical clustering and the optimal soft-thresholding ability (Fig. 3C). By studying the Pearson correlation coefficient between each module and the sample characteristics, we concluded that the blue and Turquoise modules were closely related to GC, and selected one of them for further analysis and research. Multivariate Cox regression analysis of OS in modular genes allows us to select independent prognostic genes for follow-up study (Fig. 3D). According to the PPI network (Fig. 3E), we chose LPS related hub genes (LRHG) by Degree (Fig. 3F), DMNC (Fig. 3G), EPC (Fig. 3H), MCC (Fig. 3I) and MNC (Fig. 3J), and united these genes (Fig. 3K).

Analysis of LRHG modules by WGCNA. (A) Analysis of the scale-independence of various soft-thresholding powers. (B) Identification of co-expression modules. The branches of the tree diagram correspond to the seven different gene modules. (C) Correlation of gene modules with tissue type correlation scores. (D) Univariate Cox analysis of LRHG. (E) PPI network analysis of each gene in the blue and turquoise module. (F–J) Screen the top 10 genes by Degree, DMNC (G), EPC (H), MCC (I), MNC (J). (K) Veen map of Degree, DMNC, EPC, MCC and MNC.

Survival outcomes in different LRHG groups

Firstly, we explored the association and found that DUSP1 with BRIP1, IGFBP1, NR4A3, CLSPN, PDK4 and ANGPT2 (P < 0.05); ADRB3 with ANKRD1, CLSPN and EZH2 (P < 0.01); PDK4 and BRIP1 (P < 0.05); AGT and CDC25A (P < 0.01); IGFBP1 and BRIP1 (P < 0.05); as well as CLSPN with ANGPT2 and BRIP1 (P < 0.05) had close connections (Fig. 4A). The mutant frequency of alteration between these genes displayed in Fig. 4B with DANTS1 exhibited the biggest mutation (3%), followed by ANGPT2, BRIP1, IGFBP1 and NR4A3 (2%), indicating important roles of these genes in GC development. To select independent prognostic genes, we used lasso Cox analysis to construct a prognostic index for all cancer samples (Fig. 4C), and established a risk score signature with the following formula:

Identification of LRHG hub genes. (A) Correlation the LRHG genes. (B) Waterfall plot showing the genes mutation information. (C) The coefficient profile of prognostic genes by Lasso regression analysis. (D) PCA analysis with TCGA-STAD cohort. (E) Survival analysis by K–M curve for OS of TCGA-STAD patients. (F–G) Validation analysis in GSE84437dataset.

In addition, we artificially divided people into two groups. PCA revealed clear boundaries between the two groups (Fig. 4D). Kaplan–Meier analysis further disclosed that the higher the risk score of TCGA GC patients, the shorter the survival time of patients (Fig. 4E). In the validation set GSE84437 similar phenomenon was also seen (Fig. 4F, G). These results revealed that constructed signature by prognostic genes of LRHG could stratify GC patients into two groups with different survival state. The inherent differences between the two groups deserve further exploration.

Molecular characteristics of different LRHG groups

After Cox regression analysis, based on the univariate in Fig. 5A and the multivariate in Fig. 5B, we constructed the correlation between LRHG characteristics and clinical characteristics of GC patients. From ROC curve analysis, it can be concluded that LRHG signature and clinical features are able to predict the risk, age, gender, grade and Stage with AUCs of 0.690, 0.609, 0.559, 0.548 and 0.606 (Fig. 5C). ROC curve can also show that LRHG signature to predict the 1-, 3-, and 5-year with AUCs of 0.675, 0.655, and 0.690 (Fig. 5D). Through the analysis of the predictive nomogram, the overall survival rate of the whole cohort can be predicted relatively well compared with the ideal model (Fig. 6A, B). The ROC curve showed that LRHG signature nomogram and clinical feature to predict the risk, nomogram, age, gender, grade and Stage with AUCs of 0.683, 0.753, 0.602, 0.579, 0.535 and 0.617 (Fig. 6C). In order to construct the correlation between the LRHG characteristic nomogram results and clinical characteristics of GC patients, we used the univariate Cox regression analysis in Fig. 6D and the multivariate Cox regression analysis in Fig. 6E.

Establishing the association between the LRHG and clinical features. (A) Univariate and (B) multivariate analyses of the clinical features and risk score. (C) The risk score concordance indexes with clinical features in GC patients. (D) ROC curves in GC patients in 1-, 3-, and 5-year.

Establishing the nomogram. (A) The nomogram predicts the OS probability. (B) The calibration plot predicts the OS probability of the 1-, 3-, and 5-year. (C) ROC curves in GC patients with clinical features in GC patients. (D) Univariate and (E) multivariate analyses of the clinical features and nomogram.

Immune characteristics of different LRHG groups

Recent studies have reported that LPS induces innate immune activation^29,30. Therefore, we explored the composition of immune cells in different LRHG prognostic indicator groups, and also used Wilcoxon test for correlation analysis. This test compared the proportion of immune cells in different LRHG prognostic index groups (Fig. 7A). We found that T cells CD8, T cells CD4 memory activated, T cells follicular helper, monocytes (P < 0.001), T cells gamma delta, macrophages M1 and eosinophils (P < 0.05) suggested differences between different LRHG prognostic indicator groups (Fig. 7B). Then, some LRHG characteristics are used to define immune and molecular functions (Fig. 7C). Through the survival analysis, GC patients which have a lower score had a better outcome, with APC co inhibition (Fig. 7D), check point (Fig. 7E), cytolytic activity (Fig. 7F), inflammation promoting (Fig. 7G), MHC class I (Fig. 7H), T cell co inhibition (Fig. 7I), Type II IFN Response (Fig. 7J).

Immune cells infiltration between high-risk groups and low-risk groups. (A) LRHG signature analysis in different immune subtypes. (B) CIBERSORT showed the correlation between different groups. (C) Comparison of the enrichment scores of 13 immune-related pathways in LRHG group. (D–J) Kaplan–Meier survival analysis of the correlation of immune cell abundance ratios in the LRHG group.

Benefits of immune treatment in different LRHG prognostic index groups

The TIDE algorithm can be used to evaluate the potential efficacy of immunotherapy in different groups of LRHG prognostic indicators. The higher the tide score, the higher the possibility of immune escape, which indicates that the patients are less likely to benefit from immunotherapy³¹. In our study, we found that there were differences in TIDE score dysfunction between the two groups, which confirmed that the group with low LRHG prognosis index may be more beneficial to immunotherapy (Fig. 8A). In addition, we found that the microsatellite instability (MSI) (Fig. 8B) and tumor mutation burden (TMB) (Fig. 8C) in the low LRHG prognostic index group were lower than those in the high LRHG prognostic index group. This finding suggests that patients with low LRHG prognostic index may benefit more from immunotherapy compared to patients with high LRHG prognostic index. In recent years, KRAS mutant subpopulations might also contribute to immune therapy failure³². For this purpose, we explore the LRHG prognostic index-low group had less KRAS mutant compared with the LRHG prognostic index-high group (Fig. 8D). Above all, LRHG signature shows the better result, which may benefit from TMB, MSI and KRAS mutant.

Immune response to immune therapy and the prognostic value. (A) TIDE score for different LRHG group. (B) TMB analysis, (C) MSI between the LRHG group. (D) KRAS mutant between in the LRHG high- and low-group.