Machine learning developed a programmed cell death signature for predicting prognosis and immunotherapy benefits in lung adenocarcinoma

Abstract

Background: Lung cancer is the leading cause of cancer-related deaths worldwide with poor prognosis. Programmed cell death (PCD) plays a crucial function in tumor progression and immunotherapy response in lung adenocarcinoma (LUAD).

Methods: Integrative machine learning procedure including 10 methods was performed to develop a prognostic cell death signature (CDS) using TCGA, GSE30129, GSE31210, GSE37745, GSE42127, GSE50081, GSE68467, GSE68571, and GSE72094 dataset. The correlation between CDS and tumor immune microenvironment was evaluated using various methods and single cell analysis. qRT-PCR and CCK-8 assay were conducted to explore the biological functions of hub gene.

Results: The prognostic CDS developed by Lasso + survivalSVM method was regarded as the optimal prognostic model. The CDS had a stable and powerful performance in predicting the clinical outcome of LUAD and served as an independent risk factor in TCGA and 8 GEO datasets. The C-index of CDS was higher than that of clinical stage and many developed signatures for LUAD. LUAD patients with low CDS score had a higher PD1&CTLA4 immunophenoscore, higher TMB score, lower TIDE score and lower tumor escape score, indicating a better immunotherapy benefit. Single cell analysis revealed a strong and frequent communication between epithelial cells and cancer-related fibroblasts by specific ligand-receptor pairs, including COL1A2-SDC4 and COL1A2-SDC1. Vitro experiment showed that SLC7A5 was upregulated in LUAD and knockdown of SLC7A5 obviously suppressed tumor cell proliferation.

Conclusion: Our study developed a novel CDS for LUAD. The CDS served as an indicator for predicting the prognosis and immunotherapy benefits of LAUD patients.

Keywords: Immunotherapy; Lung adenocarcinoma; Machine learning; Prognostic signature; Programmed cell death.

Fig. 1

Workflow of our study.

Fig. 2

Machine learning developed a prognostic CDS. (A) The C-index of 101 kinds prognostic models of TCGA, GSE30129, GSE31210, GSE37745, GSE42127, GSE50081, GSE68467, GSE68571 and GSE72094 datasets. The survival curve of different CDS score groups and their corresponding ROC curve in TCGA(B), GSE30129(C), GSE31210(D), GSE37745(E), GSE42127(F), GSE50081(G), GSE68467(H), GSE68571(I) and GSE72094(J) datasets.

Fig. 3

The role of CDS in predicting the prognosis of LUAD patients. (A-B) Univariate and multivariate cox regression analysis based on age, gender, clinical stage and CDS. (C) The C-index comparing the value of CDS, age, gender and clinical stage in predicting the prognosis of LUAD in TCGA, GSE30129, GSE31210, GSE37745, GSE42127, GSE50081, GSE68467, GSE68571 and GSE72094 datasets. (D) The C-index comparing the value of CDS and other 100 established signatures evaluated the prognosis of LUAD patients in TCGA dataset.

Fig. 4

Development of a predictive nomogram. (A) Nomogram developed based on CDS, age, gender, and clinical stage. (B-C) Calibration and ROC curve evaluated the role of nomogram in predicting the clinical outcome of LUAD patients. (D) DCA curve demonstrating the good potential of the nomogram for clinical application.

Fig. 5

The immune microenvironment landscape in different CDS score group. (A) The correlation between CDS and the abundance of immune cell based on seven algorithms. (B-F) The correlation between CDS and the abundance of CD8⁺ T cells, CD4⁺ T cells, B cells, macrophage M2 and CAFs. (G-H) ssGSEA analysis evaluating the level of immune cells and immune-related function in different CDS score groups. (I) The stromal score, immune score, and ESTIMAE score in different CDS score groups. (J) The immune landscape in different CDS score groups. *p<0.05, **p<0.01, ***p<0.001.

Fig. 6

CDS served as an indicator for immunotherapy benefits in LUAD. (A) The level of HLA-related genes and immune checkpoints in different CDS score group. (B-G) The TMB score, immunophenoscore, TIDE score, immune escape score and surveillance score in different CDS score group. (H-I) 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. 7

The IC50 value of drugs in different CDS score group. Low CDS score had a higher IC50 value of 5-Fluorouracil(A), Camptothecin(B), Cisplatin(C), Docetaxel(D), Epirubicin(E), Gemcitabine(F), Osimertinib(G), Paclitaxel(H), Oxaliplatin(I), Crizotinib(J), Erlotinib(K) and Gefitinib(L) in different CDS score group.

Fig. 8

The correlation between cancer related hallmarks and CDS in LUAD. Low CDS score indicated a lower sore of gene sets correlated with angiogenesis(A), coagulation(B), DNA repair(C), E2F targets(D), Hedgehog signaling(E), hypoxia(F), IL2-STAT5 signaling(G), mTORC1 signaling(H), NOTCH signaling(I), P53 pathway(J), P13K-AKT-mTOR signaling(K) and glycolysis(L).

Fig. 9

single cell analysis revealing cell communication in LUAD. (A) tSNE plot of 29 cell clusters and 8 major cell types from 11 LUAD samples. (B) Dotplot of average expression levels of marker genes in major cell types. (C) Cell Chat revealing cell-cell communication network via known ligand-receptor pairs in LUAD samples. (D) The contribution weight of outgoing and incoming signaling patterns to immune cell types in cell communication network system. (E) The expression of CDS score in major cell types based on AddModuleScore function. (F)The specific ligand-receptor pairs mediating communication between epithelial cells and CAFs.

Fig. 10

Validation of the potential function of SLC7A5 in LUAD by in vitro assays. (A) Immunohistochemical staining showing SLC7A5 expression in LUAD and normal tissues. (B) Comparison of SLC7A5 expressions in normal and LUAD cell lines. (C-D) CCK-8 assay showed that knockdown of SLC7A5 obviously inhibited the proliferation of A549 and HCC78 cells. *p<0.05, **p<0.01.