Integrated bioinformatic analysis of mitochondrial metabolism-related genes in acute myeloid leukemia

Abstract

Background: Acute myeloid leukemia (AML) is a common hematologic malignancy characterized by poor prognoses and high recurrence rates. Mitochondrial metabolism has been increasingly recognized to be crucial in tumor progression and treatment resistance. The purpose of this study was to examined the role of mitochondrial metabolism in the immune regulation and prognosis of AML.

Methods: In this study, mutation status of 31 mitochondrial metabolism-related genes (MMRGs) in AML were analyzed. Based on the expression of 31 MMRGs, mitochondrial metabolism scores (MMs) were calculated by single sample gene set enrichment analysis. Differential analysis and weighted co-expression network analysis were performed to identify module MMRGs. Next, univariate Cox regression and the least absolute and selection operator regression were used to select prognosis-associated MMRGs. A prognosis model was then constructed using multivariate Cox regression to calculate risk score. We validated the expression of key MMRGs in clinical specimens using immunohistochemistry (IHC). Then differential analysis was performed to identify differentially expressed genes (DEGs) between high- and low-risk groups. Functional enrichment, interaction networks, drug sensitivity, immune microenvironment, and immunotherapy analyses were also performed to explore the characteristic of DEGs.

Results: Given the association of MMs with prognosis of AML patients, a prognosis model was constructed based on 5 MMRGs, which could accurately distinguish high-risk patients from low-risk patients in both training and validation datasets. IHC results showed that MMRGs were highly expressed in AML samples compared to normal samples. Additionally, the 38 DEGs were mainly related to mitochondrial metabolism, immune signaling, and multiple drug resistance pathways. In addition, high-risk patients with more immune-cell infiltration had higher Tumor Immune Dysfunction and Exclusion scores, indicating poor immunotherapy response. mRNA-drug interactions and drug sensitivity analyses were performed to explore potential druggable hub genes. Furthermore, we combined risk score with age and gender to construct a prognosis model, which could predict the prognosis of AML patients.

Conclusion: Our study provided a prognostic predictor for AML patients and revealed that mitochondrial metabolism is associated with immune regulation and drug resistant in AML, providing vital clues for immunotherapies.

Keywords: acute myeloid leukemia; drug sensitivity; mitochondrial metabolism; prognostic model; tumor microenvironment.

Figure 1

Flow diagram.

Figure 2

Mutational analysis of MMRGs in AML patients and construction of MMs. (A) The mutational landscape of MMRGs in AML. (B) Localization of MMRGs on chromosomes. (C-E) KM curves for the TCGA-LAML dataset (C), GSE12417 dataset (D), and GSE37642 dataset (E) grouped by high and low MMs. (F-H) Comparison of MMs between high- and low-MMs groups in the TCGA-LAML dataset (F), GSE12417 dataset (G), and GSE37642 dataset (H).

Figure 3

Identification of co-expression modules by WGCNA. (A) Volcano plot of DEGs between high- and low-MMs groups in the TCGA-LAML dataset. (B) Heatmap of MMRGs in the TCGA-LAML dataset. (C) The mean connectivity and scale free topology module fit for different soft-threshold powers. (D) A cluster dendrogram of module eigengenes. (E) Relationships between the module eigengenes and MMs. (F-H) Venn diagrams of MMRGs with MEbrown (F), MEpurple (G), and MEcyan (H).

Figure 4

Identification of AML subtypes based on 17 module MMRGs. (A) The consensus clustering matrix of the TCGA-LAML cohort for K = 2. (B) Principal component analysis for the expression profiles of two AML clusters in the TCGA-LAML dataset. (C, D). Relative changes in the area under cumulative distribution function (CDF) curve (C), and consensus clustering CDF for k = 2-8 (D). (E) KM curves of two AML clusters in the TCGA-LAML cohort. (F) Heatmap of 17 module MMRGs in the two AML clusters.

Figure 5

Construction of a prognostic risk signature based on MMRGs. (A) The forest plot showing the univariate Cox regression results of 17 module MMRGs in the TCGA-LAML dataset. (B, C). Partial likelihood deviance of different numbers of variables (B), and coefficient profiles (C) for the LASSO regression model. (D) Distribution of risk scores and survival status, and expression of five MMRGs. (E) A nomogram for multivariate Cox regression analysis of prognostic MMRGs. (F-H). Calibration curves of the MMRGs prognostic model for 1-, 2-, and 3-year outcomes.

Figure 6

Prognostic performance of the MMRGs prognosis model. A-C. DCA curves of MMRGs prognosis model for the 1-year (A), 2-year (B), and 3-year (C). D-E. Prognostic KM curves (D) and group comparison (E) in TCGA-LAML dataset. (F, G). Prognostic KM curves (F) and group comparison (G) in GSE12417 cohort. (H, I). Volcano plot (H) and heatmap (I) of DEGs between high- and low-risk groups in TCGA-LAML cohort.

Figure 7

Validation of the expression levels of key MMRGs by IHC. A-B. H&E and ECHS1 (A) and NDUFS2 (B) IHC staining of BM samples from nonneoplastic patients and AML patients. Original magnification, 400X. (C-D). Statistics of ECHS1 (C) and NDUFS2 (D) mean IOD from IHC images of nonneoplastic patients (n = 10) and AML patients (n = 10). ****P-value< 0.0001.

Figure 8

Functional annotation of DEGs between high- and low-risk groups. (A) Bubble chart showing the GO or KEGG analysis results. B-D. GO analysis for DEGs in BP (B), CC (C), and MF (D) terms. (E) Bubble chart displaying the GO analysis results of DEGs combined with logFC. The ordinate in the bubble chart (A) is the GO terms, and the length of the bubble from Y-axis stands for the GeneRatio value of GO terms. In network diagrams (B–D), the orange dots represent specific genes, and the lavender circles represent specific pathways. In the bubble chart (E), the lilac circles represent the BP pathway; the orange circles represent the CC pathway; and the green circles represent the MF pathway.

Figure 9

Construction of the PPI network. (A) PPI network of DEGs. (B-D). Interaction networks of the top 10 DEGs in the PPI network obtained by the MCC (B), MNC (C), and Degree (D) algorithms. The color of the dots in the figure changes from yellow to red, indicating gradual increases in scores. (E) The Venn diagram of the top 10 DEGs obtained by the three algorithms. (F, G). Functional similarity analysis results (F) and PPI network (G) of hub genes.

Figure 10

Immune infiltration estimation and correlation analysis. (A-C). Comparison of StromaScore (A), ImmuneScore (B), and ESTIMATEScore (C) between groups. (D-F). Scatterplots of the correlations between risk score and StromalScore (D), ImmuneScore (E), and ESTIMATEScore (F). (G-I). Comparison of Tumor Purity (G), TIDE prediction score (H), and MMs (I) between groups. (J–L). Scatterplots of the correlations between risk score and Tumor Purity (J), TIDE prediction score (K), and MMs (L).

Figure 11

Construction of mRNA-RBP and mRNA-drugs interaction network and drug sensitivity analysis. (A, B). mRNA-RBP (A) and mRNA-drugs (B) interaction networks for the identified hub genes. (C–E). Drug sensitivity analysis results of hub genes in the GDSC database (C), CCLE database (D), and CellMiner database (E). In the mRNA-RBP interaction network (A), the blue circular block is mRNA; the orange triangle block is RBP. In the mRNA-drugs interaction network (B), the blue circular block is mRNA; the light green diamond block is drug.

Figure 12

Prognostic performance of the MMRGs prognostic model. (A, B). Forest plot (A), nomogram (B) of multivariate Cox regression model with risk score and clinical variable. (C-E). Calibration curves of multivariate Cox regression model nomogram for the 1-year (C), 2-year (D), and 3-year (E) outcomes. F-H. DCA plots of multivariate Cox regression model for the 1-year (F), 2-year (G), and 3-year (H) outcomes. (I-K). Comparison of OS events (I), prognostic KM curve (J), and time-dependent ROC curve (K) of the risk score + age + gender prognosis model.