Leveraging programmed cell death signature to predict clinical outcome and immunotherapy benefits in postoperative bladder cancer

Abstract

Bladder cancer is the fourth most common malignancy in men with poor prognosis. Programmed cell death (PCD) exerts crucial functions in many biological processes and immunotherapy responses of cancers. Cell death signature (CDS) is novel gene signature comprehensively considering the characteristics of 15 patterns of programmed cell death, which could affect the prognosis and immunotherapy benefits of cancer patients. Integrative machine learning procedure including 10 algorithms was conducted to construct a prognostic CDS using TCGA, GSE13507, GSE31684, GSE32984 and GSE48276 datasets. Immunophenoscore, intratumor heterogeneity (ITH), tumor immune dysfunction and exclusion (TIDE) score and five immunotherapy cohorts were used to evaluate the predictive value of CDS in immunotherapy response. The prognostic CDS constructed by StepCox[backward] + Ridge algorithms was regarded as the optimal prognostic model. The CDS had a stable and powerful performance in predicting overall survival of bladder cancer patients with the AUCs at 3-year, 5-year, and 7-year ROC of 0.740, 0.763 and 0.820 in TCGA cohort. Moreover, CDS score acted as an independent risk factor for overall survival rate of bladder cancer patients. Low CDS score had a higher abundance of immuno-activated cells, higher PD1&CTLA4 immunophenoscore, higher TMB score, lower TIDE score, lower immune escape score, lower ITH score, lower cancer-related hallmarks score in bladder cancer. The CDS score was higher in non-responders in pan-cancer patients receiving immunotherapy. Our study constructed a novel prognostic CDS, which could serve as an indicator for predicting the prognosis in postoperative bladder cancer cases and immunotherapy benefits in pan-cancer. Low CDS score indicated a better prognosis and immunotherapy benefits.

Keywords: Bladder cancer; Immunotherapy; Machine learning; Prognostic signature; Programmed cell death.

Fig. 1

Workflow of our study.

Fig. 2

Development of a prognostic CDS by integrative machine learning procedure. (A) The C-index of 89 kinds prognostic models of TCGA and GEO cohorts. The survival curve of different CDS score groups and their corresponding ROC curve in TCGA (B), GSE13507 (C), GSE31684 (D), GSE32984 (E) and GSE48276 (F) cohorts.

Fig. 3

Evaluation of the performance of CDS. (A,B) Univariate and multivariate cox regression suggested CDS as a risk factor for the prognosis of bladder cancer patients. (C) The C-index comparing the performance of CDS, age, gender, tumor grade, pT stage, pN stage, pM stage, and clinical stage in predicting the prognosis of bladder cancer patients in TCGA and GEO cohorts. (D) The C-index comparing the performance of CDS and other 52 established signatures in predicting the prognosis of bladder cancer patients in TCGA dataset.

Fig. 4

The tumor immune-microenvironment landscape of bladder cancer with different CDS score. (A) The correlation landscape between CDS score and immune cell infiltration evaluated by seven algorithms. (B–E) The correlation between CDS score and the abundance of macrophage M1, CD8⁺ T cells, CD4⁺ T cells, NK cells, Dendritic cells. (F,G) The score of immune cells and immune-related function in bladder cancer with different CDS score. (H–J) The stromal score, immune score, and ESTIMAE score in in bladder cancer with different CDS score. *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. 5

CDS score acted as an indicator for predicting the immunotherapy benefits in bladder cancer. (A,B) The level of immune checkpoints and HLA-related genes in bladder cancer with different CDS score. (C) The immune landscape in different CDS score groups. (D) The PD-1 and CLTA-4 IPS in bladder cancer with different CDS score. (E–I) The TMB score, TIDE score, T cell dysfunction score, immune escape score and ITH score in bladder cancer with different CDS score. (J,K) The overall rate and immunotherapy response rate in different CDS score group in GSE91061 and IMvigor210 cohort. *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. 6

The CDS score in responders and non-responders in patients receiving immunotherapy. The CDS score of different immune cells in responders and non-responders in bladder cancer (A), squamous cell carcinoma (B), skin cutaneous melanoma (C) patients receiving immunotherapy. *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. 7

The IC50 value of drugs in bladder cancer patients with different CDS score. (A,B) The IC50 value of camptothecin, cisplatin, docetaxel, gemcitabine, epirubicin, dasatinib, foretinib, nilotinib, sapitinib, and osimertinib were lower in low CDS score group versus high CDS score group in bladder cancer.

Fig. 8

The correlation between cancer related hallmarks and CDS score in bladder cancer. (A,B) The enrichment pathways in bladder cancer patients with high and low CDS score based on the dataset from Kyoto Encyclopedia of Genes and Genomes database. (C) The gene set score correlated with angiogenesis, DNA repair, EMT signaling, G2M checkpoint, glycolysis, hypoxia, IL2-STAT5 signaling, IL6-JAK-STAT3 signaling, mTORC1 signaling, NOTCH signaling, P13K-AKT-mTOR signaling and P53 pathway were higher in bladder cancer patients with high CDS score.

Fig. 9

Development of a predictive nomogram. Nomogram developed based on CDS, age, gender, tumor grade and clinical stage and calibration evaluated the role of nomogram in predicting the clinical outcome of bladder cancer patients in TCGA (A), GSE13508 (B), GSE31684 (C), GSE32984 (D) cohorts.