Dissecting L-glutamine metabolism in acute myeloid leukemia: single-cell insights and therapeutic implications

Abstract

Background

Acute myeloid leukemia (AML) is a rapidly progressing blood cancer. The prognosis of AML can be challenging, emphasizing the need for ongoing research and innovative approaches to improve outcomes in individuals affected by this formidable hematologic malignancy.

Methods

In this study, we used single-cell RNA sequencing (scRNA-seq) from AML patients to investigate the impact of L-glutamine metabolism-related genes on disease progression.

Results

Our analysis revealed increased glutamine-related activity in CD34 + pre-B cells, suggesting a potential regulatory role in tumorigenesis and AML progression. Furthermore, intercellular communication analysis revealed a significant signaling pathway involving macrophage migration inhibitory factor signaling through CD74 + CD44 within CD34 + pre-B cells, which transmit signals to pre-dendritic cells and monocytes. Ligands for this pathway were predominantly expressed in stromal cells, naïve T cells, and CD34 + pre-B cells. CD74, the pertinent receptor, was predominantly detected in a variety of cellular components, including stromal cells, pre-dendritic cells, plasmacytoid dendritic cells, and hematopoietic progenitors. The study’s results provide insights into the possible interplay among these cell types and their collective contribution to the pathogenesis of AML. Moreover, we identified 10 genes associated with AML prognosis, including CCL5, CD52, CFD, FABP5, LGALS1, NUCB2, PSAP, S100A4, SPINK2, and VCAN. Among these, CCL5 and CD52 have been implicated in AML progression and are potential therapeutic targets.

Conclusions

This thorough examination of AML biology significantly deepens our grasp of the disease and presents pivotal information that could guide the creation of innovative treatment strategies for AML patients.

Introduction

Acute myeloid leukemia (AML) is a cancer originating in the bone marrow, characterized by rapid proliferation of abnormal myeloid cells. AML’s pathogenesis is intricate and involves multiple factors. Recent studies have highlighted the significant role that metabolic pathways play in the survival and expansion of leukemia cells [1,2,3]. Among these metabolic alterations, the glutamine (Gln) utilization pathway has emerged as a potential therapeutic target due to its significance in supporting the biosynthetic demands of rapidly dividing cancer cells [4,5,6,7]. Glutamine, prevalent in blood, is pivotal in cellular metabolism, serving as a nitrogen provider for nucleotide and protein synthesis and as an essential energy source, especially during metabolic stress [8,9,10]. Despite its importance, there is still a significant gap in understanding how these metabolic pathways contribute to the pathogenesis of AML and their potential for therapeutic intervention. While studies have suggested that targeting glutamine metabolism could be beneficial, precisely identifying and targeting these pathways to enhance treatment efficacy presents substantial challenges [11, 12]. Furthermore, the complexity of glutamine metabolism and its interactions with other biological processes in AML cells add layers of complexity to both research and treatment.

Single-cell RNA sequencing (scRNA-seq) and its analytical methods has provided new opportunities for understanding the molecular profiles of diverse immune cells [13]. Prior studies have suggested that gene expression analysis from scRNA-seq immune cells may effectively predict cancer patient outcomes and responses to immunotherapy [14, 15]. In this research, our objective was to analyze scRNA-seq data from AML patients to identify glutamine-related genes. We then used these genes to develop a risk-score model for predicting the prognosis of AML patients.

Materials and methods

Bulk transcriptome data acquisition and pre-processing

The data for this study were sourced from the databases of The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO). Comprehensive expression profiles of the entire genome for AML, clinical annotations, and data on simple nucleotide variations (SNVs) in the “transcripts per kilobase per million” format were retrospectively acquired. The estimation of SNV was facilitated by the “VarScan2 Variant Aggregation and Masking” tool. The retrieval process involved the use of R packages, specifically “TCGAbiolinks (version 2.25.0)” [16] for TCGA and “GEOquery” for GEO. Totally, 134 genes linked to glutamine metabolism were identified from the GSEA database [17, 18]. (Table S1)

Single-cell sequencing data download and processing

We used the “Seurat” R package (version 4.2.0) to process the raw single-cell data from the GSE116256 dataset, which includes a total of 41 samples [19]. The initial step in the analysis involved filtering both cells and genes using specific criteria: (1) Elimination of cells with gene expression levels below 200 genes; (2) Elimination of genes detected in less than one cell; (3) Preservation of cells with expression of 500 or more genes; (4) Preservation of cells with mitochondrial gene expression under 20%.

After normalizing the data using the “NormalizeData” function from the “Seurat” R package, the identification of highly variable genes (HVGs) was determined by correlating gene expression means with their dispersion. Principal components analysis (PCA) with significant principal components was performed to cluster. The Harmony algorithm was employed to correct for batch effects across various samples. Cluster identification utilized the “FindClusters” function with the shared nearest neighbor (SNN) modularity optimization algorithm on 25 principal components at a resolution of 0.5, resulting in 24 clusters. Uniform Manifold Approximation and Projection (UMAP) was used via “RunUMAP” to visualize clustering on the UMAP-1 and UMAP-2 dimensions. The “FindAllMarkers” function, with its default settings, was used on the normalized gene expression data to identify significant markers for each cell cluster. These clusters were further characterized using biomarkers specific to different cell types. The distribution of cell types across clusters was quantified and analyzed for a more profound comprehension of the cellular composition in AML samples.

Glutamine-related gene score

The “AUCell” R package was utilized for Gene Set Enrichment Analysis (GSEA), scoring pathways for individual cells [20]. The analysis utilized the area under the curve (AUC) from 134 glutamine-related genes obtained from the Molecular Signatures Database (MSigDB) (Table S1). This method generated rankings of gene expression specific to each cell, estimating the prevalence of highly expressed genes within the selected gene set. Cells exhibiting higher AUC values indicated a greater number of genes from the set. To identify cells with active gene sets, the “AUCell_exploreThresholds” function was utilized for setting the threshold. Subsequently, AUC values were plotted using the “ggplot2” R package.

Cell communications analysis and ligand–receptor expression

CellChat, was utilized to explore the incoming and outgoing communication patterns for each cell type [21]. A thorough review of intercellular communication within AML samples was conducted, with CellChat analysis parameters set to default and a significance threshold of P ≤ 0.05.

Gene Ontology (GO) and kyoto encyclopedia of genes and genomes (kegg) pathway enrichment analysis

GO and KEGG pathway enrichment analyses involve assessing biological processes (BP), molecular functions (MF), and cellular components (CC) in GO, and identifying significantly altered metabolic pathways in KEGG. The “clusterProfiler” R package was utilized [22] to perform enrichment analysis on glutamine-related differentially expressed genes (GRDEGs) with P < 0.05.

Construction and validation of the prognostic model

Univariate Cox analysis was used to evaluate the correlation between each gene and overall survival in tumor cohorts. The tumor dataset with clinical data was divided into a training set (n = 93) and validation set (n = 39). The LASSO Cox regression model refined candidate genes to develop the prognostic model, choosing the penalty parameter (λ) based on minimal criteria. Subsequently, risk scores were calculated using a defined formula:

$$\>{\rm{riskScore}}\> = \sum\limits_{{\rm{i}} = 1}^{\rm{n}} {{\rm{Coef}}\left( {{\rm{gen}}{{\rm{e}}_{\rm{i}}}} \right){\rm{*Expression}}\left( {{\rm{gen}}{{\rm{e}}_{\rm{i}}}} \right)}$$

(Coef (gene); coefficients, Expression (gene): gene expression level)

Patients in the training group were divided into low- and high-risk group based on the median risk score. Survival analysis was conducted using Kaplan-Meier curves, with the log-rank test assessing statistical significance. The predictive model’s efficacy was evaluated through ROC curves, where an AUC above 0.6 indicates good diagnostic performance.

GSEA

GSEA evaluated significant, consistent differences in a specified gene set between two biological states, using the “clusterProfiler” R packageon a ranked gene list by log2FC values. For reliability, 1,000 gene set permutations were performed per analysis, with the c2.cp.kegg.v7.5.1.symbols set from MSigDB as the reference [13, 23]. Enrichment significance was set at an adjusted P-value < 0.05.

Gene set variation analysis (GSVA)

The “c2.cp.kegg.v7.5.1.symbols” dataset with the “GSVA” R package (v1.42.0) was utilized to conduct GSVA, with visualization via the “pheatmap” R package (v1.0.12).

Immune infiltration analysis

ssGSEA produces distinct enrichment scores for each sample-gene set pairing and assesses each sample-gene set combination individually [24]. Using TISIDB [25], encompassing data on 28 immune cell types, we quantified relative enrichment scores for each cell type from tumor sample gene expression profiles. The “ggplot2” R package [26] was then employed to graphically depict variations in immune cell infiltration.

Construction and verification of the nomogram

Patient data from the TCGA database, including clinical information, was integrated with the risk score in the regression model. A nomogram was then created using the “RMS” R package to predict survival probabilities at 1, 2, and 3 years, with the risk score serving as a crucial prognostic factor. Calibration curves were employed to evaluate the nomogram’s accuracy in incorporating both prognostic and clinical features.

Somatic mutation analysis

Genomic variations were examined using mutation data, visualized with the “maftools” R package to depict somatic variants within different clusters. This included analysis of single nucleotide polymorphism (SNP), insertion and deletion (INDEL), tumor mutation burden (TMB), and mutation frequency [27]. In malignant tumors, primary driver genes were identified as the top 20 most frequently mutated [26].

Assessment of the drug susceptibility

The half-maximal inhibitory concentration (IC50) data were obtained from the Genomics of Drug Sensitivity in Cancer (GDSC) along with clinical gene expression data. “oncoPredict” R package was used to predict the responsiveness to potential therapeutic drugs for AML patients [28].

Statistical analysis

Data were analyzed using R. Kaplan-Meier curves and log-rank tests compared survival rates between groups. Cox regression analysis, including univariate and multivariate methods, assessed prognostic variables. Visualization was done with “ggplot2” and “survival” packages, and heatmaps were created with “Pheatmap”. For normally distributed variables, two-tailed t-tests or one-way ANOVAs identified significant differences. Data that were not normally distributed were analyzed with Wilcoxon or Kruskal-Wallis tests, setting statistical significance at P < 0.05.

Results

Identification of the reactive subgroups of glutamine-related genes

To investigate the origin of the genes exhibiting elevated expression, the cellular population of AML was examined using the single-cell sequencing dataset, GSE116256. Following an initial quality control assessment, single-cell transcriptomes were derived from a comprehensive pool of 38,410 cells. This study encompassed 41 samples, with a distinct uniformity in the distribution of cells across groups. This uniform distribution suggested the absence of discernible batch effects among the groups, thereby establishing a foundation for subsequent analytical procedures (Fig. 1A). Following the clustering analysis, all cellular entities were segregated into 24 distinct clusters (Fig. 1B). Based on the unique genetic features inherent to each cluster, specific cell types were identified using cell- type-specific biomarkers (Fig. 1F). We identified 16 unique cell types, including naïve CD8 + T cells, natural killer cells, and monocytes (Fig. 1C). We also analyzed the distribution of the diverse cell types within each group (Fig. 1D). And we concentrated on active cell subgroups to detect the expression patterns of glutamine-related genes at the single-cell level. Examination of the UMAP diagram of active cells revealed CD34 + pre-B cells as active cells within the spectrum of immune cell types, with statistical significance (Fig. 1E).

Fig. 1

Subgroup identification from the scRNA-seq dataset. (A) A UMAP plot illustrating the cellular distribution between AML and control cohorts. (B) A UMAP plot delineating the distribution of AML subgroups. (C) A UMAP plot presenting the annotation findings for AML subgroups. (D) A histogram exhibiting the distribution of cell types across AML patients and the control group. (E) A UMAP chromatic plot indicating the activity scores of cells, the brighter the color, the higher the activity. (F) Expression profiles of the marker genes in each cell type

Cellular communication patterns of the aml microenvironment

To clarify the collective function of cells, we conducted an analysis of cell-cell communication, quantifying the interactions and their intensities among various cell types (Fig. 2A and B). Further investigation into the potential signals and molecular pairs among the 16 cell types revealed that stromal cells, myeloid cells, hematopoietic progenitors, CD8 + T cells, and CD34 + pre-B cells were primary signal sources. Additionally, erythrocytes, plasma cells, naïve Th0 cells, monocytes, and CD34 + pre-B cells were identified as major signal recipients (Fig. 2C). The signaling pathways potentially involved among these cell types encompassed MIF, MK, RESISTIN, and BAFF (Fig. 2D). Subsequently, signaling pairs involving CD34 + pre-B cells were explored. The findings indicated that CD34 + pre-B cells had the most robust communication with pre-dendritic cells via the MIF-(CD74 + CD44) pathway and with monocytes through the same MIF-(CD74 + CD44) pathway (Fig. 2E and F). These findings began to clarify potential interactions among these cell types, offering insights for further study of the collective role of CD34 + pre-B cells, pre-dendritic cells, and monocytes in AML patients. Additionally, the ligand was observed to be highly expressed in stromal cells, naïve T (Th0) cells, and CD34 + pre-B cells, with the CD74 receptor present across a range of cell types (Fig. 2G).

Fig. 2

The results of cell-cell communication analysis. (A) Interactions among different cell types. (B) Intensities of interactions among different cell types. (C) Heatmap showing the possible outgoing signaling pathways among different cell types. (D) Heatmap showing the possible incoming signaling pathways among different cell types. (E) Dot plot showing the possible incoming or outgoing signaling pairs. (F) Chord diagram showing the possibility of incoming or outgoing signaling pairs. (G) Expression distribution of MIF signaling genes

Functional enrichment analysis

From the comparison of CD34 + pre-B cells and other cell types, 329 differentially expressed genes (DEGs) were identified between the two groups (P < 0.05, |Log2 Fold Change|>0.5). Among these genes, the top 36 were upregulated, coinciding with the top five downregulated DEGs within CD34 + pre-B cells (Fig. 3A). 185 DEGs were identified from the comparison of AML and normal samples. Furthermore, we identified six upregulated genes along with 35 downregulated DEGs in AML (Fig. 3B). Totally, 113 key DEGs were identified from the intersection of DEGs in CD34 + pre-B cells and those in AML (Fig. 3C).

To investigate the biological functions of key DEGs, we performed GO and KEGG enrichment analyses. The GO results showed that these genes were significantly enriched in biology process (BP) pathways, including antigen processing and presentation of peptide antigen (GO: 0048002), antigen processing and presentation of exogenous peptide antigen (GO: 0002478), antigen processing and presentation (GO: 0019882) and antigen processing and presentation of exogenous antigen (GO: 0019884) (Fig. 3D, G). These genes were also enriched in the pathways of cellular components (CC), such as endocytic vesicle membranes (GO: 0030666), MHC protein complexes (GO: 0042611), endocytic vesicles (GO: 0030139), ER to Golgi transport vesicle membrane (GO: 0012507), and integral component of lumenal side of endoplasmic reticulum membrane (GO: 0071556) (Fig. 3D, H). In the pathways of molecular function (MF), involving peptide binding (GO: 0042277), peptide antigen binding (GO: 0042605, amide binding (GO: 0033218), MHC class II protein complex binding (GO: 0023026) and MHC protein complex binding (GO: 0023023), these genes were also enriched (Fig. 3D, F)).

In addition to GO enrichment analyses, KEGG enrichment analyses indicated that the pathways significantly associated with the key DEGs were related to antigen processing and presentation (hsa04612), phagosome function (hsa04145), type I diabetes (hsa04940), cell adhesion molecules (hsa04514), viral myocarditis (hsa05416), allograft rejection (hsa05330), graft-versus-host disease (hsa05332), NK cell cytotoxicity (has04650), and Epstein-Barr virus infection (hsa05169) (Fig. 3E, I).

Fig. 3

Identification and functional enrichment of DEGs. (A) Heatmap of the top 36 up- and top 5 down-regulated DEGs in CD34 + pre-B cells compared to other cell types. (B) Heatmap of the top 6 up- and top 35 down-regulated DEGs between AML and control groups. (C) Venn diagram illustrating the overlap of DEGs in CD34 + pre-B cells and AML. (D) GO analysis for the DEGs shows the top 3 significant enrichment pathways in terms of biology process (BP), cellular components (CC), and molecular function (MF). (E) KEGG analysis for the DEGs shows the significant enrichment pathways. (F) The top 5 DEGs enriched molecular function (MF) pathways. (G) Bubble chart showing the top 4 enriched biology process (BP) pathways. (H) A circle plot showing the cellular component (CC) pathways enriched by GO. (I) KEGG enrichment analyses of the DEGs. (J) Table of corresponding GO pathway IDs and pathway names for F, G, and H

Construction and verification of prognostic risk model

We identified signature genes in CD34 + pre-B cells with P < 0.05, leading to the discovery of 33 AML prognostic genes (Table S2). LASSO regression on the training set culled redundant genes, identifying 10 prognostic genes for AML patients (Fig. 4A and B). To assess model robustness, we divided samples based on their median risk scores into high- and low-risk categories. We evaluated the distribution of risk scores, survival durations, and gene expression profiles for each group. Notably, NUCB2 was predominantly expressed in the high-risk group, as shown in Fig. 4C. Kaplan-Meier survival curves for both the training cohort (Fig. 4E) and validation cohort (Fig. 4G) demonstrated poorer outcomes for those in the high-risk category. The model’s efficacy in predicting prognosis was evaluated using ROC curves, with AUCs of 0.836, 0.808, and 0.830 for 1-, 2-, and 3-year survival in the training cohort (Fig. 4D), and 0.866, 0.807, and 0.823 in the validation cohort (Fig. 4F), respectively.

Fig. 4

Cox and LASSO regression analyses were performed on the AML dataset as follows: (A) LASSO regression plot showing the path of the independent variable, with lambda (log-scale) on the x-axis and the coefficients on the y-axis. (B) LASSO regression confidence intervals for each lambda value. (C) Distribution of risk scores, AML survival summary, and heatmap for key genes. (D) Time-dependent ROC curves for 1-, 2-, and 3-year model predictions in the training set. (E) Survival curves for high- and low-risk groups from the training set, with green indicating the high-risk group and red the low-risk group. (F) Time-dependent ROC curves for 1-, 2-, and 3-year model predictions in the validation set. (G) Survival curves for high- and low-risk groups from the validation set

GSEA and GSVA results

We conducted GSEA, focusing on the significantly enriched pathways by normalized enrichment scores (NES) to delve into the mechanisms behind the DEGs. Notably enriched pathways included leishmania infection (NES = 2.4426, adjusted P = 0.0188, FDR = 0.0124, Fig. 5A), antigen processing and presentation (NES = 2.3283, adjusted P = 0.0188, FDR = 0.0124, Fig. 5B), autoimmune thyroid disease (NES = 2.2449, adjusted P = 0.0188, FDR = 0.0124, Fig. 5C), pyrimidine metabolism (NES = 1.4124, adjusted P = 0.0438, FDR = 0.029, Fig. 5D), T cell receptor signaling (NES = 1.3881, adjusted P = 0.0438, FDR = 0.029, Fig. 5E), and the insulin signaling pathway (NES = 1.3513, adjusted P = 0.0457, FDR = 0.0302, Fig. 5F).

Additionally, GSVA identified seven pathways with significant differences including O_Glycan_Biosynthesis, Oxidative_phosphorylation, Alzheimers_disease, Huntingtons_disease, Pantothenate_and_CoA_Biosynthesis, Pathogenic_Escherichia_coli_infection, and FC_Gamma_R_mediated_phagocytosis pathways (Fig. 5G).

Fig. 5

GSEA and GSVA analyses identified significant enrichment pathways. (A) Enrichment of Leishmania infection pathway. (B) Enrichment of antigen processing and presentation pathway. (C) Enrichment of autoimmune thyroid disease pathway. (D) Enrichment of pyrimidine metabolism pathway. (E) Enrichment of T cell receptor signaling pathway. (F) Enrichment of insulin signaling pathway. (G) Enrichment of additional significantly enriched pathways as determined by GSVA

Immune infiltration analysis

We evaluated the infiltration levels of 28 immune cell types (Fig. 6A) and observed notable differences among them, including activated CD4 T cells, myeloid-derived suppressor cells, activated CD8 T cells, central memory CD4 T cells, effector memory CD4 T cells, CD56 bright NK cells, gamma delta T cells, immature dendritic cells, macrophages, mast cells, NK cells, neutrophils, plasmacytoid dendritic cells, type 1 T helper cells, and type 17 T helper cells (P < 0.05; Fig. 6B). We also found notable correlations between specific genes and immune cells (Fig. 6C and E). PSAP was notably linked with CD56bright natural killer cells (R=-0.7481, P < 0.001, Fig. 6C). Similarly, S100A4 (R=-0.6964, P < 0.001) and VCAN (R=-0.8397, P < 0.001) were significantly correlated with central memory CD4 T cells (Fig. 6D and E), respectively.

Fig. 6

Differences in immune infiltrations among the high- and low-risk groups. (A) Heatmap illustrating differences in immune cell infiltration. (B) Estimated proportions of tumor-infiltrating immune cells. (C) Correlation analysis between PSAP and CD56bright natural killer cells. (D) Correlation analysis between S100A4 and central memory CD4 T cells. (E) Correlation analysis between VCAN and central memory CD4 T cells

Validation of risk scoring could be used as an independent prognostic factor

To establish the risk score as an independent prognostic indicator, it was evaluated using univariate and multivariate Cox regression analyses that included age, sex, and survival time as clinical variables. The multivariate analysis was crucial for developing the nomogram, which effectively forecasted clinical outcomes (Figure S1A). Calibration curves for the nomogram confirmed its accuracy and reliability over 1, 2, and 3 years (Figure S1B).

Tumor mutation burden (TMB) and drug susceptibility analysis

We identified the top 20 genes most frequently mutated in AML. In the high-risk group, NPM1 was the most commonly mutated gene, followed by IDH2 and TTN (Fig. 7A). Conversely, DNMT3A exhibited the highest mutation frequency in the low-risk group, followed by NPM1, IDH2, IDH1, and KDM6A (Fig. 7B). Our analysis showed that TMB significantly differed between the low-risk and high-risk groups, with TMB being notably higher in the low-risk group.(Fig. 7C).

Furthermore, we examined the capacity of the risk score to predict the drug susceptibility of AML patients. The clinical efficacy of various treatments for AML was investigated, including 5-Fluorouracil_1073, ABT737_1910, Acetalax_1804, Alisertib_1051, AT13148_2170, AZ960_1250, AZ6102_2109, AZD1208_1449, and AZD6482_2169 (Fig. 7D-L). High-risk score patients showed greater sensitivity to 5-Fluorouracil_1073 (Fig. 7D) and AZD6482_2169 (Fig. 7G), indicating that the two agents may be viable options for this patient group. In contrast, patients with low- risk scores showed greater sensitivity to ABT737_1910 (Fig. 7E), Alisertib_1051 (Fig. 7H), AT13148_2170 (Fig. 7I), AZ960_1250 (Fig. 7J), suggesting that patients in the low- risk cohort may benefit from these agents.

Fig. 7

Differences in the TMB and drug susceptibility between the high- and low-risk groups. (A) The top 20 genes with the highest mutation frequency were in the high-risk group. (B) The top 20 genes with the highest mutation frequency were in the low-risk group. (C) Differences in the TMB. (D_L) Differences in IC50 values of 5-Fluorouracil_1073 (D), ABT737_1910 (E), Acetalax_1804 (F), AZD6482_2169 (G), Alisertib_1051 (H), AT13148_2170 (I), AZ960_1250 (J), AZ6102_2109 (K), AZD1208_1449 (L) between the high- and low-risk group

Discussion

Advancements in scRNA-seq technology provide valuable insights into the molecular characteristics of immune cells that infiltrate tumors within the tumor microenvironment (TME) [29]. In this study, we utilized an AML scRNA-seq dataset to investigate the expression and roles of genes related to glutamine (Gln) metabolism in AML. Notably, CD34 + pre-B cells showed high glutamine-related activity, implying a regulatory role for glutamine in these cells and potential effects on tumor development. Our analysis of the GSE116256 single-cell dataset provided a detailed examination of AML’s cellular makeup and the expression patterns of Gln metabolism-related genes across various cell types. After stringent quality control and analysis, we discerned 16 unique cell types within AML samples, including naïve CD8 + T cells, natural killer cells, and monocytes. This comprehensive mapping elucidated the cellular heterogeneity and complexity of AML. Further analysis identified active cellular subgroups based on Gln metabolism gene expression, with CD34 + pre-B cells as a key subgroup with significant metabolic activity in the AML microenvironment. This discovery has shifted our research focus towards understanding the role of CD34 + pre-B cells in AML pathogenesis and progression.

Moreover, our study highlights the importance of the MIF-(CD74 + CD44) signaling pathway in the communication initiated by CD34 + pre-B cells in the AML microenvironment. The pathway’s potential as a key point in the influence of glutamine metabolism on AML progression is particularly noteworthy, given its role in mediating communication from CD34 + pre-B cells to pre-dendritic cells. Prior research has recognized the MIF pathway as a significant factor in immunomodulation, with implications in a range of conditions, including autoimmune diseases and cancers [30,31,32,33,34]. Our findings build upon these studies by specifically implicating the MIF pathway in AML, suggesting a potential new avenue for therapeutic intervention. The CD74 + CD44 receptor-ligand complex has been a subject of interest in the context of immune response regulation [35,36,37]. Our data suggest that this complex could be a central mediator in the aberrant signaling observed in AML. Furthermore, the interplay between Gln metabolism and immune response in AML progression is gaining increasing recognition. The differential gene expression we observed, enriching in pathways related to antigen presentation and immune response, aligns with the notion that Gln may exert its effects on AML not only through direct metabolic support but also via modulation of immune cell functions [38, 39]. This metabolic-immunological crosstalk is of particular interest, as it points to a multifaceted role for Gln in cancer biology. Recent studies have shown that inhibiting glutaminase, a key enzyme in glutamine metabolism, may be an effective therapeutic approach for AML and other cancers. The study suggests that disrupting Gln metabolism can impair the survival and proliferation of cancer cells, which is of significant interest given the known metabolic reprogramming in AML [39]. Integrating these insights with our observations, it is conceivable that therapies targeting the MIF-(CD74 + CD44) pathway in conjunction with metabolic interventions could offer a synergistic approach to disrupt the intricate network sustaining AML pathogenesis.

Additionally, we identified 10 prognostic genes associated with AML patients, which included CCL5, CD52, CFD, FABP5, LGALS1, NUCB2, PSAP, S100A4, SPINK2, and VCAN. CCL5 is overexpressed in AML and promotes its progression [40]. A prior study showed that CD52 was significantly expressed in neoplastic stem cells among AML patients. It emerged as a novel prognostic marker and therapeutic target in a specific subset of AML patients [41]. In a recent study, the realignment of lineage maturation with proliferation was attained in AML cells that did not exhibit the acute promyelocytic leukemia subtype but expressed elevated levels of the retinoic acid carrier FABP5. This can be accomplished by inhibiting FABP5 using small molecules. It is crucial to note that this therapeutic approach is different from conventional cytotoxic chemotherapy and has the significant advantage of preserving normal hematopoiesis [42]. Although there are no previous studies on the roles of CFD, PSAP, and VCAN in AML, these genes may serve as potential new targets for AML therapy.

The enrichment analysis revealed enrichment in processes related to antigen processing and presentation, endocytic vesicle membranes, and peptide binding. Additionally, KEGG pathway analysis showed involvement in pathways such as phagosomes, and type I diabetes mellitus, highlighting their significance in immune regulation and disease pathology. DEGs were enriched in pathways for antigen processing and presentation, underscoring their role in immune surveillance and response.

Totally, 33 signature genes were identified significantly associated with AML prognosis in CD34 + pre-B cells. Distinct variations in risk score distribution, survival durations, and the expression levels of 10 genes between groups indicate their value in risk stratification. Notably, higher NUCB2 expression in the high-risk group suggests its potential as a prognostic biomarker for AML. Kaplan-Meier survival curves validated the gene signature’s prognostic significance, indicating poorer outcomes for high-risk patients across training and validation cohorts.

Our analysis of immune infiltration showed significant correlations including the S100A4 and VCAN genes with central memory CD4 T cells, the PSAP gene with CD56bright natural killer cells. CD56bright natural killer cells are notably active in AML, particularly in immune responses against AML cells. Central memory CD4 T cells, a subset of memory T cells with robust immunological memory, are crucial for initiating strong and specific immune responses [43, 44]. In AML, central memory CD4 T cells contribute to immune surveillance and defense against leukemic cells by recognizing specific antigens associated with AML [45]. These T cells are crucial in coordinating immune responses by releasing cytokines and activating other immune cells.

This study had some limitations. The construction of the risk signature, predicated on Gln metabolism genes, was contingent upon a small cohort of patients with AML sourced from TCGA database. However, more extensive prospective clinical investigations are needed to ascertain the prognostic validity of these key genes. Furthermore, the reliance on bioinformatics methodologies for the development of the Gln metabolism gene-dependent risk signature underscores the need for subsequent fundamental research to validate and strengthen these conclusions.

Conclusions

The current study has made significant strides in understanding the molecular mechanisms and immune cell dynamics within the TME in AML through the bioinformatics analysis of scRNA-seq dataset. The research has successfully identified key genes and metabolic pathways associated with AML progression, particularly focusing on the role of glutamine metabolism in CD34 + pre-B cells, which may have significant implications for tumorigenesis. Enrichment and pathway analyses have highlighted the key roles of these genes in processes like antigen processing and presentation, endocytic vesicle membranes, and peptide binding, which are vital for immune regulation and disease development. These findings are significant for developing targeted therapies, prognostic models, and immunotherapeutic strategies, potentially enhancing patient outcomes and furthering our understanding of the complex mechanisms of AML pathogenesis.