The landscape of genetic variation of DENRGs in GC

A total of 48 DENRGs were identified using limma package for further analysis (p<0.05, Fig.1A). Out of 433 GC samples, 147 (33.95%) were showed regulatory mutations associated with necroptosis (Fig.1B) and ATRX (5%) was the highest frequency mutated gene. As loss or gain of function is commonly achieved through DNA mutation or amplification/deletion, we considered both somatic mutation and somatic copy number changes in our analysis. We first summarized the incidence of copy number variations and somatic mutations of 48 DENRGs in GC. The frequency of CNV alterations and found that all 48 DENRGs showed prevalent CNV alterations (Fig.1C). The rates of amplification or deletion for most of DENRGs were relatively low. The altered position of CNVs of DENRGs on chromosome were also scanned and illustrated with visual figure (Fig.1D). In addition, most of the DENRGs were significant increase in tumor tissues (Fig.1E).

The landscape of genetic alterations of DENRGs in GC. (A) Heatmap of DENRGs expression between the normal and tumor samples. Blue represents normal gastric tissue, pink represents tumor tissue; upregulated genes were defined as red, and downregulated genes as blue. (B) Mutation characteristics of DENRGs in the TCGA-GC cohort. The TMB is presented in the barplot at the top of the image; the mutation frequency of each DENRGs is indicated on the barplot right. The barplot on the right represents different mutation types proportions. (C) CNV variants frequency of the DENRGs in the TCGA-GC cohort. Red: amplification frequency. Green: loss frequency. The column represented the alteration frequency. (D) The locations of CNV alteration of DENRGs on 23 chromosomes. (E) Expression of DENRGs between normal gastric tissue and tumor tissue. Blue: normal gastric tissue. Red: tumor tissue. DENRGs, differentially expressed necroptosis-related genes. (*p<0.05; **p<0.01; ***p<0.001).

To further explore the interactions of these DENRGs, we conducted a PPI analysis, and the PPI network was shown in Fig. S1A. In addition, the correlation network containing all DENRGs was presented in Fig. S1B. The network above indicated that there was a very strong correlation among DENRGs. GO-term analysis showed that DENRGs were associated with necrotic cell death, programmed necrotic cell death, necroptotic process and apoptotic signaling pathway (Fig. S2A). KEGG pathway analysis displayed that these DENRGs were involved in multiple tumor-related signaling pathway including necroptosis, apoptosis, TNF signaling pathway, IL-17 signaling pathway, and Toll-like receptor signaling pathway (Fig. S2B).

According to Consensus clustering analysis, when the clustering variable was set to the optimal value (K=2), the intragroup correlations were the highest, and the intergroup correlations were the lowest, indicating that all GC patients could be classified into two molecular subtypes (Figs.2A, S3A and S3B), which were termed as C1 (n=208) and C2 (n=163). The heatmap demonstrated a significant difference between cluster C1 and C2 in clinical factors including tumor grade and T stage (Fig.2B). Result of KaplanMeier curve analysis revealed that the patients in C2 cluster had a poorer prognosis (Fig.2C). The results above indicated that the necroptosis subtypes classified by consensus clustering analysis do well in distinguishing prognosis of those GC patients.

Tumor molecular subtypes related by differentially expressed necroptosis-related genes. (A) Consensus clustering of GC patients for k=2 in the meta-cohort (TCGA-GC and GSE84437). (B) Unsupervised clustering heatmap of top 100 DEGs in GC. Clusters, age, gender, grade and stage were used as patient annotations. Red represents high DEGs expression and blue low DEGs expression. *p<0.05; **p<0.01; ***p<0.001. (C) KaplanMeier curves (Log-rank test, P=0.004) for OS of two necroptosis-related molecular subtypes. Blue line represents cluster C1 (n=208), yellow line represents cluster C2 (n=163). DEGs, differentially expressed genes between various molecular subtypes; OS, overall survival.

Given the clear importance of the TME in tumorigenesis, we further investigated whether the two subtypes showed differential characteristics of immune microenvironment and the main results presented in Fig.3AH. The abundance of immune infiltrating cells, including resting Dendritic cells, resting Mast cells, T cells regulatory (Tregs), Monocytes and M2 macrophages, were found significantly higher in the C2 subtype. And M1 macrophages, T cells follicular helper and activated T cells CD4 memory in C1 subtypes showed greater infiltration. These results suggested that the two molecular subtypes associated with necroptosis had distinct TME infiltration characteristics and prognoses.

TME immune cell infiltration levels between two molecular subtypes. The abundance of Monocytes (A), resting Mast cells (B), M2 macrophages (C), M1 macrophages (D), resting Dendritic cells (E), T cells regulatory (Tregs) (F), T cells follicular helper (G) and activated T cells CD4 memory (H) between the two subtypes (all p<0.05). Blue represents cluster C1, red represents cluster C2. The median value is represented as the thick line, and the interquartile range is represented as the box bottom and top. Scattered dots represent outliers.

To better understand the mechanisms responsible for the prognosis differences in the two above molecular subtypes, we further investigate the functional and pathway and 1101 DEGs associated with necroptosis phenotypes were identified by the limma package. GO analysis showed an enrichment of GO terms for these DEGs, including extracellular matrix organization, collagen containing and extracellular matrix binding (Fig.4A). KEGG pathway analysis for the DEGs showed that genes involved in immune-related pathways were enriched, including ECM-receptor interaction, Focal adhesion, and TGF-beta signaling pathway (Fig.4B). These results reconfirmed a pivotal role of necroptosis in regulating the immune microenvironment.

Functional enrichment analysis of the DEGs. (A) Top 10 enriched GO terms of the DEGs (B) Top 10 enriched KEGG pathways of the DEGs. The box color represents the number of enriched genes. Red represents a large number of genes enriched; blue is the opposite. DEGs differentially expressed genes, BP biological process, CC cellular component, MF molecular function. (all adjusted p<0.05).

Although our results identify a role of necroptosis molecular subtypes in prognosis and regulation of immune infiltration, these analyses are based only on patient groups and cannot be used to predict the necroptosis characteristics in individual GC patients. For this, we next constructed an multigenic prognostic signature associated with prognosis and response to treatment in each GC patient based on differential genes of molecular subtypes. We performed univariate Cox regression analysis on all DEGs and resulted in 84 genes as candidate genes (all P<0.005; Fig.5A). Most of the candidate genes were risk factors for the prognosis of GC except for MYB and RNF43. We then subjected the candidate genes to LASSO Cox regression analysis by narrowing the number of genes for the establishment of the NRGsig (Fig.5B and C). In total, 11 optimal genes (CYTL1, PLCL1, CGB5, ADRA1B, APOD, RGS2, CST6, MATN3, RNF43, SLC7A2 and SERPINE1) were screened (Table 2) and most of the optimal genes were significant differentialexpression between the normal tissue and tumor tissue (Fig. S4). The formula of the risk score was calculated as follow:

$$begin{gathered} Risk; score = CYTL1 {text{exp}}.; times ;0.05351 ; + ; PLCL1 {text{exp}}.; times ;0.06101 ; + ; CGB5 {text{exp}}.; times ;0.1605 hfill \ quad quad quad quad , + ; ADRA1B {text{exp}}.; times ;0.07886; + ;APOD {text{exp}}.; times ;0.03166; + ;RGS2 {text{exp}}.; times ,0.04199, + ;CST6 {text{exp}}. hfill \ quad quad quad quad , times ;0.00119 ; + ;MATN3 {text{exp}}.; times ;0.13379 ; + ;RNF43 {text{exp}}.; times ; - 0.09577; + , SLC7A2 {text{exp}}. times ;0.07123. hfill \ quad quad quad quad , + ;SERPINE1 {text{exp}}.; times ;0.12925 hfill \ end{gathered}$$

The development of NRGsig in the TCGA-GC cohort. (A) The prognostic-related genes determined by univariate Cox-regression analysis. Red represents risk genes; green represents protective genes. (B) LASSO regression of prognostic-related genes. (C) Crossvalidation for tuning the parameter selection.

All GC patients were divided into high- and low-risk score group according to the median risk score value. Next, we investigated whether the prognostic signature could distinguish different risk groups of patients clearly. A clearly discernable dimensions between the two risk groups of patients was observed according to the results of PCA and t-SNE analysis (Fig.6A and B). KaplanMeier curves analysis revealed high-risk group patients had a worse prognosis. (Fig.6C). The time-dependent ROC curves were performed to evaluate the prediction performance of the NRGsig and the areas under the curve for 5-year was 0.743 in the TCGA-GC cohort (Fig.6D). Results above demonstrated NRGsigs advantage as robust tool for prognosis.

Prognosis value of necroptosis-related prognostic signature in the TCGA-GC cohort. (A) Principal component analysis plot. (B) T-distributed neighbor embedding plot. (C) KaplanMeier curves (Log-rank test, P<0.001) for OS of high- and low-risk groups. (D) The AUC of the prediction of 1, 3, 5year survival rate of GC. OS, overall survival.

We externally validated the NRGsig using the GSE84437 dataset, an independent validation dataset, and found a similar prediction performance. Patients were then classified as being high or low risk according to the calculated NRGsig risk score. A clearly two directions between the two risk groups of patients was also observed according to the results of PCA and t-SNE analysis (Fig.7A and B). KaplanMeier curves analysis indicated high-risk group patients had a worse outcome (Fig.7C). This independent validation dataset yielded a prediction performance AUC of 0.623 at 5-year (Fig.7D). As a whole, these results showed a satisfactory prediction performance of the NRGsig in external data.

Validation of the necroptosis-related prognostic signature in the GSE84437 cohort. (A) Principal component analysis plot. (B) T-distributed neighbor embedding plot. (C) KaplanMeier curves (Log-rank test, P=0.005) for OS of high- and low-risk groups. (D) The AUC of the prediction of 1, 3, 5year survival rate of GC. OS, overall survival.

The independence of NRGsig were evaluated by univariate and multivariate Cox regression analysis and the result revealed the NRGsig was an independent prognostic factor of GC (Fig.8A and B). Above analysis were repeated in the GSE84437 cohort and similar results were observed (Fig.8C and D). Furthermore, the clinical features in the different risk groups for TCGA-GC cohort we depicted as a heatmap (Fig.8E). To verify the clinical implications of our NRGsig risk score, we examined the correlation of the risk score with the available clinical features in TCGA-GC cohort. The KaplanMeier curves indicated that risk score remained its independent predictive performance regardless of other clinical features, including age (60 or>60years), sex (female or male), grade (G1-2 and G3), T-stage (T3-4), N-stage (N0 and N1-3), and M-stage (M0) (Fig. S5AL). Survival analysis demonstrated that these 11 optimal genes were all correlation with the OS of GC patients (Fig. S6AK). All the results above illustrated that NRGsig was a satisfactory and reliable prognostic tool and could be as an independent risk factor for GC.

Independent prognosis analysis. (A, B) Univariate Cox regression analysis in the TCGA-GC cohort. (C, D) Multivariate Cox regression analysis in the GSE84437 cohort. (E) Heatmap depicting the clinicopathological characteristics and optimal genes expression between the high- and low-risk groups. Risk, age, gender, grade and stage were used as patient annotations. Red represents high expression and blue low expression. *p<0.05; **p<0.01; ***p<0.001.

After categorizing cases of TCGA-GC cohort into two risk score groups by the median risk score value, we further performed GSEA analysis towards them. The results of GSEA suggested that the KEGG_COMPLEMENT_AND_COAGULATION_CASCADES, KEGG_ECM_RECEPTOR_INTERACTION, KEGG_FOCAL_ADHESION, KEGG_HYPERTROPHIC_CARDIOMYOPATHY_HCM, and KEGG_NEUROACTIVE_LIGAND_RECEPTOR_INTERACTION were the top five most enriched pathways in the high-risk group, while the KEGG_CELL_CYCLE, KEGG_DNA_REPLICATION, KEGG_BASE_ EXCISION_REPAIR, KEGG_RIBOSOME, and KEGG_SPLICEOSOME pathways were most enriched in the low-risk group (Figs. S7A and B).

To make the prognosis tool more convenient and quantitative, we integrated risk score with other clinical features including Age and TNM stage to establish a nomogram followed by a series of performance testing (Fig.9A). The net benefit of nomogram was better than other clinical factors, a clinical value was observed as our expectations (Fig.9B). The ROC curve analysis revealed that nomogram had an advantage over other single predictors. In addition, an excellent consistency with ideal model could be observed in the subsequent calibration plot of nomogram for OS predicting (Fig.9C and D). Furthermore, to evaluate the prediction performance of the NRGsig for clinical applications in the TCGA-GC cohort, we compared our prognostic signature with other GC signatures reported in 2020 (Dai signature, Guan signature, Liu signature and Shao signature, respectively). We adopted similar risk score-estimated method described above towards these four signatures to generate risk score for samples from TCGA-GC cohort. The time-independent ROC curves illustrated that Liu signature, Shao signature and Guan signature exhibited lower AUC values for 1-, 3- and 5-year survival rates than NRGsig. The Dai signature presented similar AUC values with our signature (Fig. S8AE). Similar to our signature, these four signatures could also predict the OS of GC patients except for Liu signature and shao signature (Fig. S8GJ). Moreover, the C-index of the NRGsig was the higher than other four signatures (Fig. S8K). NRGsig evidenced its advantage in long-term survival predicting and risk stratification compared with other four prognostic signatures.

The construction and assessment of nomogram. (A) Nomogram integrating clinical factors and risk score for predicting 1-, 3-, and 5-year OS in TCGA-GC cohort (B) Decision curves of risk score, nomogram, and single clinical factors including T stage, N stage and age. (C) The time-dependent ROC curves of risk score, nomogram and single clinical factors including T stage, N stage and age. (D) The calibration curves for 1-, 3-, and 5-year OS. OS, overall survival.

In line with our aim to increase the response to immunotherapy, we investigated the potential correlates between immune infiltration of tumors and NRGsig risk score. After calculating the infiltrating score of 16 immune cells and 13 immune-related pathways by using ssGSEA, we observed significantly increased antigen presenting function including aDCs, DCs and APC co-stimulation score in the high-risk group, while the activity of APC co-inhibition and MHC class I showed the opposite variation (all adjusted P<0.05). Besides, contents of Treg cells, TIL cells and T helper cells were relatively higher in high-risk group, while the activity of Th2 cells had exactly the reverse results. Those results suggested significant difference in T cell regulation between the two subgroups. Moreover, CCR, mast cells, B cells, macrophages, neutrophils, parainflammation, type I IFN response and type II IFN response were observed to have increasing activities in samples from high-risk group (Fig.10A and B). Similar observational results existed for in the GSE84437 cohort (Fig.10C and D). Taken together, the findings of this study demonstrated that different risk groups have different immune landscape, which affected the prognosis of GC patients.

ssGSEA scores in the high- and low-risk group in the TCGA-GC and GSE84437 cohort. (A, B) TCGA cohort, (C, D) GSE84437 cohort. The scores of 16 immune cells (A, C) and 13 immune-related functions (B, D) are displayed in boxplots.

We next explored potential expression changes of immune checkpoints between high- and low-risk groups. Results showed clear differences between the two patient groups, such as BTLA, CD86, CD200, CD27, and other immune checkpoints (Fig. S9). These results highlighted NRGsig as a therapeutic potential for combination strategies with immune checkpoint blockade (ICB) therapy in GC patients. Beyond ICB therapy, we also investigated sensitivity of chemotherapeutic and targeted therapeutics agents between high- and low-risk score groups in TCGA-GC cohort. Results indicated that IC50 toward eleven chemotherapeutics including A.770041, AS601245, AZ628, Axitinib, Luminespib, Navitoclax, Motesanib, Ponatinib, Rucaparib and Saracatinib, of samples in low-risk group were higher than those of high-risk group except for Veliparib (P<0.05), suggesting that samples in low-risk group were more responsive to those medicine (Fig.11AK). As mentioned already, GSEA analysis revealed that a drug-resistant pathway like KEGG_BASE_EXCISION REPAIR was highly enriched in the low-risk score group, which could partially explain the above results. Drugs sensitivity analysis suggested that high-risk score patients might be more suitable for chemotherapy better response to chemotherapy.

Drugs sensitivity analysis in patients from different risk score groups. The sensitivity to chemotherapeutic drugs was represented by the half-maximal inhibitory concentration (IC50) of chemotherapeutic drugs. (AK) Comparisons of IC50 for chemotherapeutics drugs between two subgroups revealed that the high-risk group was more likely to benefit from the treatments (KruskalWallis test, all p<0.01).

Evidence is growing that high TMB is a feature associated with response to immunotherapy in a variety of tumors, and high TMB levels lead to an increase in tumor neoantigens, which may trigger the immune system to attack the tumor40,41. Thus, we assessed the correlation of risk score with TMB in the TCGA-GC cohort. A negative relationship was observed between them, and the TMB score of the two risk groups were evaluated and significant disparity could be observed. The results illustrated that low-risk group patients had a significantly higher TMB than high-risk group (Fig.12A). The combination of high TMB and low-risk score had the best OS in GC by KaplanMeier curves (Fig.12B).

Correlation of risk score with TMB and predictive value of risk score for immunotherapy response. (A) TMB differences between the high- and low-risk score groups and the scatter plot depicted a positive correlation between risk score and TMB. (B) KaplanMeier curves for patients stratified by risk score and TMB in the TCGA-GC cohort. (CE) Immunophenscore (IPS) between high- and low-risk score groups. Blue represents the low-score group and red the high-score group. The thick line within the violin plot represents the median value. The inner box between the top and bottom represents the interquartile range. (C) IPS score when PD-1 positive; (D) IPS score when CTLA4 positive; (E) IPS score when both PD-1 and CTLA4 positives. TMB, tumor mutation burden; IPS, Immunophenscore. (F) TIDE score differences between the high- and low-risk score groups and the scatter plot depicted a positive correlation between risk score and TIDE and lower risk score may be more likely to benefit from the immunotherapy (Spearman text, p<0.001).

Furthermore, we explored the potential of risk score as predictor for immunotherapy response. We applied two mature algorithms, including IPS and TIDE, to predict the response of GC samples with different risk score to immunotherapy. The result evidenced that the IPS value for CTLA4 or PD1 therapy response was more sensitive in the low-risk group and suggested that the NRGsig has high potentiality for predicting CTLA4 and PD1 blockade therapy (Fig.12CE). On the other hand, the TIDE score was higher in the low-risk group and was also positively correlated with risk score, which indicated the lower risk score might benefit more from immunotherapy (Fig.12F and G). Two distinct algorithms drew consistent results. The results above implied that NRGsig may effectively help predict the response to immunotherapy.

Go here to see the original:
Construction of a novel signature and prediction of the immune landscape in gastric cancer based on necroptosis-related genes | Scientific Reports -...

Comments are closed.