Optimal combination of immune checkpoint and senescence molecule predicts adverse outcomes in patients with acute myeloid leukemia

Abstract

Background: High expression of immune checkpoints (ICs) and senescence molecules (SMs) contributes to T cell dysfunction, tumor escape, and progression, but systematic evaluation of them in co-expression patterns and prognosis in acute myeloid leukemia (AML) was lacking.

Methods: Three publicly available datasets (TCGA, Beat-AML, and GSE71014) were first used to explore the effect of IC and SM combinations on prognosis and the immune microenvironment in AML, and bone marrow samples from 68 AML patients from our clinical center (GZFPH) was further used to validate the findings.

Results: High expression of CD276, Bcl2-associated athanogene 3 (BAG3), and SRC was associated with poor overall survival (OS) of AML patients. CD276/BAG3/SRC combination, standard European Leukemia Net (ELN) risk stratification, age, and French-American-British (FAB) subtype were used to construct a nomogram model. Interestingly, the new risk stratification derived from the nomogram was better than the standard ELN risk stratification in predicting the prognosis for AML. A weighted combination of CD276 and BAG3/SRC positively corrected with TP53 mutation, p53 pathway, CD8+ T cells, activated memory CD4+ T cells, T-cell senescence score, and Tumor Immune Dysfunction and Exclusion (TIDE) score estimated by T-cell dysfunction.

Conclusion: High expression of ICs and SMs was associated with poor OS of AML patients. The co-expression patterns of CD276 and BAG3/SRC might be potential biomarkers for risk stratification and designing combinational immuno-targeted therapy in AML.Key MessagesHigh expression of CD276, BAG3, and SRC was associated with poor overall survival of AML patients.The co-expression patterns of CD276 and BAG3/SRC might be potential biomarkers for risk stratification and designing combinational immuno-targeted therapy in AML.

Keywords: Prognosis; acute myeloid leukemia; immune checkpoint; risk stratification; senescence.

Figure 1.

Schematic diagram of the study design. Transcriptome sequencing data of AML patients in the Cancer Genome Atlas (TCGA) were downloaded from the University of California Santa Cruz (UCSC) database for constructing weighted correlation network analysis (WGCNA) of immune checkpoint (IC) and senescence molecules. Then, univariate and multivariate COX regression models and correlation analysis were used to select molecules with co-expression patterns to estimate risk scores and construct a prognosis model. TCGA, Beat-AML, GSE71014, and Guangzhou First People’s Hospital (GZFPH) datasets were used to analyze and validate the relationship between the co-expression of immune checkpoints and senescence molecules and prognosis. Moreover, the correlation between risk score and gene mutations, Gene set variation analysis (GSVA) pathways and tumor immune microenvironment was investigated. Finally, the risk score and clinical information were used to establish a nomogram model and risk stratification to predict the overall survival (OS) of AML patients, as well as the immune response to immunotherapy.

Figure 2.

Selection of co-expressed and prognostic ICs and selection of senescence molecules of AML patients. (A) WGCNA was constructed using IC and senescence molecules from 155 AML patients in the TCGA database. Highly correlated genes were assigned to the same color module (left panel), and the correlation between genes in the same color module was shown in a heatmap (right panel). The color scale from red to yellow represents the correlation from weak to strong. (B) Univariate Cox regression was used to analyze the IC and senescence molecules in the blue color module in the TCGA (left panel) and Beat-AML (right panel) datasets. The genes with p < 0.05 were displayed in the circular plot. (C) IC and senescence molecules were associated with poor OS in both TCGA and Beat-AML datasets. *, p < 0.1; **, p < 0.05; ***, p < 0.01; ****, p < 0.001.

Figure 3.

( A–B) CD276, BAG3, and SRC are positively correlated with age in AML patients in the TCGA (A) and GZFPH (B) datasets.

Figure 4.

(A–D) OS analysis of CD276 (left panel), BAG3 (middle panel), and SRC (right panel) in AML patients in the TCGA (A), Beat-AML (B), GSE71014 (C) and GZFPH (D) datasets.

Figure 5.

A weighted combination of CD276, BAG3, and SRC was associated with the prognosis of AML patients. (A) Correlation among CD276, BAG3, and SRC with p < 0.05 in TCGA, Beat AML, GSE71014, and GZFPH datasets. (B) The radar plot shows the contribution of CD276, BAG3, and SRC to OS in the TCGA dataset, which was determined by the coefficients β in the multivariate COX regression model. Risk score = β1* (CD276 expression) + β2* (BAG3 expression) + β3* (SRC expression). (C) Positive correlation between risk score and age in the TCGA (left panel) and GZFPH (right panel) datasets. D-F: Relationship between risk score and favorable, intermediate, and poor risk subgroups TCGA (D), Beat-AML (E), and GZFPH (F) datasets. (G) OS analysis of low- and high-risk patients based on the combination of CD276, BAG3, and SRC in the TCGA, Beat-AML, GSE71014, and GZFPH datasets.

Figure 6.

Construction of risk stratification for AML patients. (A) The risk score, risk stratification, age, and FAB subtype were used to construct the nomogram model in the TCGA dataset. B-C: Kaplan Meier curves of new (left panel) and standard (middle panel) risk stratification were plotted and C-index was used for evaluating the performance of new and standard European Leukemia Net (ELN) risk stratification (right panel) in the TCGA (B) and GZFPH (C) datasets.

Figure 7.

Correlation between risk score and gene mutation in AML patients. A-B: Mutation landscape of the top 20 genes in low- (left panel) and high-risk score (right panel) subgroups in the TCGA (A) and Beat-AML (B) datasets. C-D: High-risk score was positive correlation with TP53 mutation (C) and TP53 mutation was associated with poor OS (D) in AML patients in the TCGA (left panel) and Beat-AML (right panel) datasets.

Figure 8.

Gene set variation analysis (GSVA) of high-risk score subgroup in AML. A-B: GSVA pathways of high-risk score subgroup in the TCGA (A) and GSE71014 (B) datasets. C-D: Relationship between risk score and GSVA score of p53 pathway in the TCGA (C) and GSE71014 (D) datasets. E-F: OS analysis of low- and high-GSVA score of p53 pathway based on the optimal cut-point in the TCGA (E) and GSE71014 (F) datasets.

Figure 9.

Correlation between risk score and tumor immune microenvironment in AML. A-B: Correlation between risk score and immune score (left panel), immune cell subpopulations (middle panel), and T-cell senescence score (right panel) in the TCGA (A) and GSE71014 (B) datasets. (C) The GZFPH dataset was used to validate the relationship between risk score and T-cell senescence score. (D) Risk score had a positive correlation with Tumor Immune Dysfunction and Exclusion (TIDE) score in the TCGA (left panel) and GSE71014 (right panel) datasets.