PDF
Citation
Print
Introduction
Acute myeloid leukemia (AML) is preceded by malignant clonal expansion of progenitor cells with impaired differentiation and is associated with the highest proportion of deaths among all types of leukemia [1, 2, 3–4]. In the United States, in 2024, there were 20,800 new cases of AML, accounting for 33.1% of the total incidence of leukemia, the 9 th and 10 th most prevalent malignancies in men and women, respectively, and 11,220 deaths, accounting for 47.4% of the total mortality rate of leukemia, ranking as the 6 th most common death due to cancer in men and the 8 th in women [5]. AML can lead to life-threatening infections, bleeding, hemolysis, and anemia [6]. Therefore, blood product inputs are clinically recommended to reverse anemia, control hemolysis, and reduce bleeding [6], thereby improving patient survival.
Although there is a strong relationship between blood product input and hemolysis [7, 8], the causes of hemolysis are more complex in patients with AML [9, 10]. Patients with leukemia develop thrombotic microangiopathy, which can cause intravascular hemolysis [11]. The therapeutic disposition to leukemia can also lead to hemolysis, possibly because of immune reconstitution and interferon release after a patient receives a bone marrow transplant [12]. There have also been case reports that gemtuzumab ozogamicin (Mylotarg™), a therapeutic agent for AML, may cause intravascular hemolysis in patients with low or normal blood potassium concentrations [13]. This shows that hemolysis is closely related to the prognosis of patients with AML; however, the reasons for the occurrence of hemolysis in patients with AML are not fully understood.
Existing research has focused on how to avoid hemolysis and whether new AML therapies cause hemolysis [14, 15, 16–17]. Moreover, determining the expression levels of specific genes is important for understanding the mechanisms of hemolysis in patients with AML, as well as for prognostic assessment. In this study, we used bioinformatics methods, such as weighted gene co-expression network analysis (WGCNA), to identify crucial genes involved in the hemolytic process in patients with AML and Mendelian randomization (MR) to infer causal relationships between exposure factors and study outcomes. Through this approach, we aimed to identify potential diagnostic biomarkers associated with hemolytic processes in patients with AML and to elucidate novel pathophysiological mechanisms underlying hemolysis in this patient population.
Materials and methods
Data source
The dataset was obtained by transcriptome microarray analysis of human peripheral blood from patients with AML (N = 65) (M0:25; M1:11; M2:10; M3:1; M4:7; M4 eo t [16;16] ou inv [16]:4; M5:6; M6:1) and healthy human volunteers (healthy controls; N = 62).
Identification of differentially expressed genes
First, we differentially analyzed data from peripheral blood samples in the GSE164191 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE164191) and GSE89565 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE89565) datasets using R software (version 4.3.1) and preprocessed them for batch correction and normalization. We then performed differentially expressed gene (DEG) analysis screening using the “limma” package. After analyzing the significance of expression level differences, the “pheatmap” and “ggplot2” R packages were used to generate volcano maps and DEG expression heat maps, respectively. By analyzing genome relationships and gene-phenotype correlations, WGCNA can identify gene lists with similar/synergistic interactions [18]. We constructed a gene co-expression network for AML using the “WGCNA” R package (version 1.72–1), and then evaluated the relevance of the different modules to the pathogenic mechanism of AML and identified the most relevant modules from WGCNA as the central genes.
Screening of target genes and signaling pathway analysis
We first searched for related genes in the GeneCards database (www.genecards.org) using “hemolysis” as the keyword, and then took the intersection of these genes with the WGCNA-screened genes and DEGs to obtain intersecting genes [19]. These intersecting genes were regarded as candidate hub genes relevant to AML and the pathogenesis of hemolysis. Subsequently, we performed Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis using the “clusterProfiler” R package to assist us in understanding the potential mechanisms of AML progression and pathogenesis [20]. The STRING database (https://string-db.org/) and Cytoscape software (https://cytoscape.org/) were used to construct protein–protein interaction (PPI) networks and search for hub genes.
Nomogram model construction and immune cell analysis
We constructed a nomogram model to predict the risk of hemolysis in patients with AML using the “rms” package [21]. The performance of the nomogram model was evaluated by calculating Harrell’s concordance index, which assesses predictive power [22]. To validate the diagnostic effectiveness of the candidate biomarkers, we used the “ROC” package to construct a receiver operator characteristic (ROC) curve. The area under the ROC curve (AUC) was used as a measure of accuracy. The criterion 0.9 ≤ AUC < 1 was used to identify excellent accuracy. Using correlation analysis, we examined the relationship between the target genes and immune cells in patients with AML to investigate their roles in hemolysis.
Mendelian randomization
All data analyzed in this study were obtained from a publicly available database. Two-sample MR was used to explore the causal association between hub genes and the risk of hemolysis in patients with AML with single nucleotide polymorphisms (SNPs) defined as instrumental variables. Hub gene data were obtained from a publicly available genome-wide association study (GWAS) data source. We selected two representative genes, STAT1 and TLR4, and a leukemia inhibitory factor, as representatives of AML for MR analysis. Data for STAT1 are available at https://gwas.mrcieu.ac.uk/datasets/?gwas_id__icontains=prot-a-2868&year__iexact=&trait__icontains=&consortium__icontains =, data on TLR4 is available at https://gwas.mrcieu.ac.uk/datasets/?gwas_id__icontains=prot-c-3647_49_4&year__iexact=&trait__icontains=&consortium__icontains =, and data for leukemia inhibitory factor are available at https://gwas.mrcieu.ac.uk/datasets/?gwas_id__icontains=prot-a-1736&year__iexact=&trait__icontains=&consortium__icontains =. MR analysis was performed using the “TwoSampleMR” package, and inverse variance weighting (IVW) was used to assess the relationship between hub gene levels and the risk of hemolysis in patients with AML. MR–Egger was used for additional sensitivity analyses [23].
Results
DEG screening
We first used the Limma package in R software (V4.0.1) to perform differential analysis between peripheral blood data from patients AML (GSE89565) and healthy individuals (GSE164191) and found a total of 7,339 DEGs, of which 3,733 were upregulated and 3,606 were downregulated (Figs. 1A, B; Supplementary Table S1).
Construction of WGCNA network and identification of an AML-related module
To identify potential gene modules that have a greater correlation with AML, we performed WGCNA on the peripheral blood datasets of patients with AML (GSE164191 and GSE89565) (Fig. 2A). By clustering the genes and patients separately, we identified four different modules, obtained positive correlation coefficients (Fig. 2B), and based on the correlations, the genes with significance greater than 0.5 and Module Membership greater than 0.8 in the turquoise module were selected for subsequent analysis (Fig. 2C and Supplementary Table S2).
Parallel signaling pathway enrichment analysis of hemolysis-related genes in patients with AML
We first searched the GeneCards database (www.genecards.org) for related genes using “hemolysis” as the keyword and obtained 1,330 genes (Supplementary Table S3). These genes, the hub genes in the WGCNA, and DEGs were screened using Venn diagrams, and 190 overlapping genes were identified as candidate hub genes that may play important roles in AML and hemolysis (Fig. 3A). We further explored the potential roles of these 190 overlapping genes using GO and KEGG analysis (Figs. 3B, C). GO enrichment analysis showed that the overlapping genes mainly affected lymphocyte-mediated immunity, leukocyte-mediated immunity, and the activation of immune responses. KEGG enrichment analysis showed that the overlapping genes mainly affected cytokine–cytokine receptor interactions, the hematopoietic cell lineage, and the PI3 K AKT pathway. These results revealed the important role of this group of genes in the immune response.
PPI network analysis of the hub genes
We used the STRING database to construct a PPI network of the overlapping hub genes (Fig. 4A). The presence of a line between the dots indicates an interaction between the two molecules, and different colors represent upregulation/downregulation. Subsequently, the top 10 highly ranked upregulated genes were visualized using Cytoscape software (Fig. 4B). Cells expressing PTPRC, FCGR3B, STAT1, TLR4, GZMB, CD2, SCARB1, APOE, TFRC, and FCGR2 A were sorted. A deeper color indicates a higher score.
Construction of a nomogram model for predicting the risk of hemolysis in patients with AML
We constructed a nomogram model to predict the risk of hemolysis in patients with AML (Fig. 5A). Our nomogram model performed well at predicting hemolysis in patients. Subsequently, we calculated the ROC curves for the five hub genes (TLR4, PTPRC, FCGR3B, STAT1, and APOE) to assess their diagnostic effect. The AUC of our nomogram differentiated between hemolysis in patients with AML and controls (Fig. 5B). The AUC values for TLR4, PTPRC, FCGR3B, STAT1, and APOE were 0.983, 1.000, 1.000, 0.965, and 1.000, respectively.
It is noteworthy that in our results, three hub genes, PTPRC, FCGR3B, and APOE, had AUC values of 1. APOE is a classical hemolysis inhibitory factor that protects erythrocytes from environmentally-induced lysis in a number of ways [24, 25–26]. Meanwhile, in our PPI results, APOE had the lowest level of significance among the five genes. The expression of PTPRC, also known as CD45, gradually disappears during bone marrow erythropoiesis and is commonly used for subpopulation analysis, immunoassays of myeloid cells, and as a serum marker for some hemolytic diseases [27, 28–29]. FCGR3B is a low-affinity immunoglobulin gamma Fc receptor III, also known as CD16, and a surface marker of immature erythrocytes. It can be used as a marker of nocturnal paroxysmal hemoglobinuria [30, 31] and can also indicate the occurrence of hematologic malignancy [32]. Therefore, we chose TLR4 for further analyses.
Analysis of TLR4 correlations with immune cells
KEGG pathway analysis and GO enrichment showed that these hub genes were primarily associated with cytokine–cytokine receptor interactions. A positive correlation was observed between monocytes, neutrophils, macrophages (M0), CD4 + memory-activated T cells, and TLR4 expression levels, whereas activated natural killer (NK) cells, dendritic cells, memory B cells, helper T cells, plasma cells, resting memory T cells, resting NK cells, CD8-positive T cells, and TLR4 expression levels were negatively correlated (Figs. 6A, B). These results demonstrated a direct relationship between immune cells and the hub gene TLR4 in AML.
TLR4 is causally associated with the risk of hemolysis in patients with AML
The SNP characteristics of TLR4 and the risk of hemolysis in patients with AML are shown in Supplementary Table S4. No SNP was a weak instrumental variable. Figures 7A and B present the causal effects of each genetic variant on the risk of hemolysis in patients with AML. We assessed the causal association between TLR4 levels and risk of hemolysis in patients with AML. Using the IVW method, we found that TLR4 levels were associated with the risk of hemolysis in patients with AML, with an odds ratio of 1.013 (95% confidence interval = 1.001–1.045, p = 0.031216). No horizontal pleiotropy was observed at the intercept of the MR-Egger regression (p = 0.376), indicating that pleiotropy did not influence the causal effect of the funnel plot (Fig. 7C). Despite these consistent results, all SNPs were found to be significant for causality. Patients with AML with high TLR4 levels and hemolysis risk did not have a dominant SNP, and the previous MR results were valid.
Discussion
AML is a hematological malignancy characterized by hemorrhage, fatigue, infection, and the failure of myeloblasts or pregranulocytes to differentiate normally [2, 33]. Hemolysis occurs in patients with AML for various reasons, including thrombocytopenia [34], TAM [11], hemolysis after bone marrow transplantation [12], and therapy-related hemolysis [13]. Hemolytic manifestations are common in pediatric patients (≤ 15 years of age) at first diagnosis, mostly manifesting as hepatosplenomegaly and abnormal laboratory tests [35]. In adults, hemolysis most often manifests as post-treatment hemolysis [36], which is one of the criteria for evaluating treatment efficacy and the patient’s prognosis [37, 38]. Classical cytotoxic treatment of AML (a combination of 7 days of cytarabine and 3 days of anthracycline) also leads to hemolysis, thrombocytopenic hemorrhage, and granulocytopenia, which often imply a poor prognosis [39, 40]. Here, we explored the specific biological mechanisms that produce hemolysis in patients AML using bioinformatics methods, clarified the causal relationship between hemolysis and hub genes in patients with AML, and provided new perspectives on the specific course of AML treatment and progression.
We first utilized data from the Gene Expression Omnibus database for differential analysis, identified DEGs, and then intersected them with hub genes in WGCNA and hemolysis-related genes in the GeneCards database to obtain marker genes for hemolysis in patients with AML. We then used PPI network analysis and Cytoscape software to screen hub genes, and obtained 10 hub genes. Five genes, TLR4, PTPRC, FCGR3B, STAT1, and APOE that were significantly associated with patient prognosis were identified by establishing a nomogram model of prognostic risk.
PTPRC (CD45) and FCGR3B (CD16) are expressed in immature erythrocytes and leukocytes (T cells, NK cells, and B cells) and are commonly used for blood cell subpopulation analysis and immune status analysis [27, 28]. Consistent with our findings, both CD45 and CD16 are markers of certain hemolytic disorders (paroxysmal sleep hemoglobinuria) [29, 30–31], and CD16 is a marker of certain hematologic neoplasms (myelodysplastic syndromes) [32]. However, the roles of CD45 and CD16 in the development of hemolysis are not clear. The role of STAT1 in AML is also controversial. Nitulescu et al. experimentally demonstrated that STAT1 activates the JAK-STAT pathway through phosphorylation at Ser727 to promote AML growth and malignant biological behaviors [41], whereas Gao et al. mediated the phosphorylation of the 727 site using diptoindonesin G to promote AML cell differentiation through the ERK pathway [42]. However, similar to CD45 and CD16, the role of STAT1 in hemolysis in patients is unclear.
Apolipoprotein E (APOE) is involved in the conversion and metabolism of lipoproteins and regulates many biological functions. APOE/TLR4 values have been reported to be critical for lipid metabolism [43]. By downloading AML-related data from The Cancer Genome Atlas database and performing differential analysis using edgeR, Huang et al. found APOE to be the hub gene in patients with a mutation of TP53, which is the driving gene of AML [44]. However, they used only the R package edgeR for differential analysis and did not validate or evaluate the effectiveness of the candidate biomarkers for diagnostic use through mathematical modeling. More robust and comprehensive designs should be considered when using these bioinformatic approaches to study diseases.
TLR4 is a pattern-recognizing receptor [45]. It recognizes intracellular and extracellular ligands, mediates inflammatory suppression, and maintains an environment conducive to AML cells [45]. In an in vitro study, Hu et al. found that TLR4 regulates the inflammatory response of macrophages through the AP- 1 signaling pathway in AML [46]. Moreover, Baakhlagh et al. found that AML cell proliferation decreases after inhibiting TLR4 activity [47]. In addition, Chakraborty et al. reported that the TLR4/NLR signaling pathway promotes ineffective hematopoiesis through the activation of damage-associated molecular patterns, which contribute to the development of AML [48]. We found that TLR4 was involved in the development of AML and in the occurrence of hemolysis in patients with AML; thus, it is clear that TLR4 still has broad research prospects in the field of AML.
This is the first study to explore the causal association between TLR4 levels and AML risk using a two-sample MR analysis based on a large amount of GWAS data on TLR4 (exposure) and AML (outcome). The findings of this MR study suggest that serum TLR4 levels are causally associated with an increased risk of hemolysis in patients with AML.
Despite the significance of our study, it has some shortcomings. First, only one AML dataset was used, because it was difficult to find a healthy control population cohort from a relevant database. These results would have been more convincing if more AML datasets from different populations were combined. Second, this study only used bioinformatics to analyze hub genes and their potential functions associated with the occurrence of hemolysis in patients with AML, and more biological experiments are needed to confirm the specific mechanisms of the identified hub genes.
In conclusion, we constructed a WGCNA-based co-expression network and identified hemolysis-associated AML genes. This may help in the development of pre-symptomatic diagnostics and provide further insights into the molecular mechanisms underlying hemolysis-associated AML risk genes.
XML Download