Single-cell transcriptomic profiling reveals immune cell heterogeneity in acute myeloid leukaemia peripheral blood mononuclear cells after chemotherapy

We performed scRNA-seq analysis of PBMC samples from five AML patients (two males and three females, 33–54 years old) and six healthy humans (three males and three females, 26–37 years old) to explore their cellular composition (Fig. 1a, Supplementary Table 1). The cohort included patients with either chemotherapy or transplantation at different stages (bone marrow suppression period or immune recovery period) and was genetically heterogeneous (Supplementary Table 1). Among then, AML3 and AML3B derived from the same patient, but they are in different stages after chemotherapy. After an initial quality control assessment and doublet removal using the Seurat R package and DoubletFinder, we obtained a total of 91,772 cells, with 44,950 cells from healthy donors and 46,822 cells from AML patients, for single-cell transcriptome analysis. The final cell number ranged from 1,936 to 14,081 per sample (Supplementary Fig. 1a, Supplementary Table 2), the median unique molecular indices (UMIs) ranged from 4,001 to 10,864 per cell (Supplementary Fig. 1b), and the sequencing saturation ranged from 68.7% to 83.2% per sample (Supplementary Table 3). In addition, based on UMAP analysis, we identified 14 main clusters in parallel according to the gene profile and representative gene markers of the cell (Fig. 1b and c), CD3D, CD3G, CD3E, and CD4 for CD4-T cells; KLRF1, NCAM1, and KLRD1 for NK cells; and LYZ, CD14, FCN1, and C5AR1 for CD14⁺ monocyte cells.

The relative proportions of cell clusters and the relative expression levels of cluster-specific markers in each cluster were also confirmed (Fig. 1d, e and Supplementary Fig. 2 and 3). There was high heterogeneity in the proportion of cell clusters among AML samples compared to healthy samples that showed consistent immune cell compositions (Fig. 1e, Supplementary Fig. 4, and Supplementary Table 4). The AML1 sample who had TP53 mutant that belong to high-risk grade, had a considerably higher number of HSC-prog cells, which were rarely detected in the other AML samples. The excessive number of HSC-prog cells may be related to AML relapse, as previous studies have reported that rare leukaemia cells with stem cell characteristics such as self-renewal ability and drug resistance may contribute to the maintenance and relapse of the disease [12, 13]. Compared to healthy donors, AML2 and AML3 had higher proportions of CD14⁺ monocytes, suggesting their role in inhibiting the activation of T cells and promoting AML blasts survival [6]. Compared to AML3B, the proportion of CD14⁺ monocytes in AML3B decreased obviously. In addition, AML3B, showed an 82% increase in CD4⁺ T cells compared to healthy donors, indicating an immunostimulatory effect of chemotherapy in promoting the formation of CD4-T cells [14]. After hemiallogeneic hematopoietic stem cell transplantation, AML4 had a similar cell type composition with healthy donors, which proved stem cell transplantation was an effective method to cure AML4. Similar to AML3B, AML5 also had major T cell, which might attribute to their myelosuppression status. Collectively, the results of our study reveal heterogeneity in immune cell types and proportions in AML patients and suggest potential relationships with their clinical status under chemotherapy or transplantation that are compatible with existing AML progression and maintenance theories.

To further characterise the intrinsic structure and potential functional subtypes of the overall T cell populations, we performed spectral clustering on all T cells using UMAP [15]. As a result, we identified nine stable subclusters, including five subclusters of CD4⁺ cells and four subclusters of CD8⁺ cells, each expressing their unique signature genes. The signature gene expression patterns in different subclusters of CD4⁺ and CD8⁺ T cells are shown in Supplementary Fig. 5a and Supplementary Fig. 6a. The relative cluster-specific marker expression levels in the five cellular subclusters of CD4⁺ T have been presented in Fig. 2b, d and e. For example, cells that specifically expressed naive marker genes such as LEF1, SELL, and CCR7 represented naïve T cells (Tn) [16]. In contrast, the central memory T cell (Tcm) cluster was represented by cells expressing TCF7, ANXA1, LEF1, and SELL genes [17] and is commonly associated with central memory and tumor inhibition function [18, 19]. Other clusters include the effector memory cell (Tem) cluster that highly expressed GZMK [20]; the effector T cell (Teff) cluster that specifically expressed ‘effector’ genes such as ANXA1, GNLY, CX3CR1, and TBX21 [21]; and the regulatory T cell (Treg) cluster that expressed LEF1, SELL, and regulatory marker FOXP3. The proportion of cells expressing cluster-specific markers was shown in Fig. 2c.

We observed high variation in the number of CD4⁺ Tem, Teff, and Treg in AML patients, except for CD4⁺CCR7⁺Tn and CD4⁺TCF7⁺Tcm, which were prevalent in peripheral blood in AML and healthy samples (Fig. 2c, Supplementary Fig. 5b). The percentage of cytotoxic CD4⁺Teff varied among AML patients (Fig. 2c), sharing a low proportion in most AML patients, except in AML4 and AML5, suggesting that the enhanced functions of various cells are involved in exerting immune effects [22]. Tregs are known to contribute to the suppression of anti-leukemic activity [23], and their immunosuppressive function contributes to leukaemia progression [24]. AML3B is in myelosuppression state, we also observed a decrease in Tregs in AML3B compared to AML3, consistent with the myelosuppression effect of chemotherapy [25]. These observations suggest that CD4⁺ T cells should not be viewed as a homogeneous population in tumor samples and that the cellular constitution of CD4⁺ T cell sub-clusters are associated with AML disease conditions.

Next, we analysed the AML and healthy CD4⁺ T cells using gene set variation analysis (GSVA) and the result showed high individual heterogeneity in AML patients, significantly differing from healthy controls, as indicated in Fig. 2f. We observed downregulation of five transcription factors involved in the development of T cell immunity, including KLF13, GABPA, RBBP5, ZNF66, and IRF8, in most AML patients. The exception is AML3B, which upregulates transcription factors related to cell growth and differentiation, including STAT3, KLF7, NFKB1, STAT4, MYB, ZNF274, FOXP1, IRF9 and TCF7. In adverse, these genes were downregulated in healthy donors (Fig. 2g). In addition to transcription factors, in AML samples, we identified the upregulation of genes that were reported to accelerate leukaemogenesis, including S100A8, S100A9, and RPS26 [26, 27], and the downregulation of genes related to cancer control and defence, such as XIST, GIMAP7, NKG7, and GNLY (Supplementary Fig. 7). Among the downregulated genes, XIST is a non-coding RNA transcript on the X chromosome that acts as a major effector in the X-inactivation process [28]. NKG7 is critical for controlling cancer initiation, growth, and metastasis and regulating lymphocyte granule exocytosis and downstream inflammation [29]. GNLY is found in cytotoxic granules of cytolytic T lymphocytes and NK cells, which are essential in defence against tumors and microbes [30]. These results reveal the modulation of CD4⁺ T cells in AML, such as suppression of CD4⁺ T cell activities via downregulation of transcription factors and genes related to cancer control and defence and upregulation of genes in CD4⁺ T cells that promote leukaemogenesis and inflammatory signalling.

Next, we focused on marker gene expression and composition ratios of the four major CD8⁺ T cell clusters (Fig. 3a-d). CD8⁺ Teff cells highly express CXCR3, TBX21, PRF1, and FCGR3A genes, which are commonly associated with T cells with effector functions [31], while CD8⁺ Tn cells specifically express effector marker genes CCR7, LEF1, SELL, and TCF7 [16], together with CD8⁺ Tem cells that express high levels of GZMK, similar to their CD4⁺ counterpart [20]. Notably, we observed that a CD8⁺ T cell subcluster was composed mainly of mucosal-associated invariant T cells (MAIT), characterised by the specific expression of SLC4A10 and RORC (Fig. 3d). Unlike CD4⁺ T cells, we observed similar relative compositions of CD8⁺ T sub-clusters among healthy donors and patients with AML. In addition, we observed a significant downregulation of 29 transcription factors associated with immune cell homeostasis and cancer control across all AML samples, including ZBTB7A, NR3C1, ELF2, KLF13, KLF2, XBP1, RXRA, FOSL2, and HDAC2 (Fig. 3f and 3g). Among these, ZBTB7A determines the fate of the haematopoietic system and regulates glycolysis, and AML patients are prone to ZBTB7A mutations [32]. HDAC is an epigenetic regulator that is tightly involved in AML aetiology and mediates the immune system regulation [33]. We then conducted GSVA analysis of CD8⁺ T subclusters to compare the differences in enriched pathways between AML patients and healthy donors. We observed an upregulation of 50 hallmark pathways related to immune reactions, cancer progression, and immunotherapy in AML patients [31], such as hypoxia, MYC target v1, oxidative phosphorylation, P53 pathway, and heme metabolism (Fig. 3e) [34, 35]. Notably, the MYC target v1 pathway is reported to be an important oncogenic pathway in cancer development [32]. We also observed variations in enriched oncogenic pathways that promote T cell growth and survival in certain AML patients. For example, AML4 demonstrated significant enrichment of genes involved in hedgehog signalling and KRAS signalling [36], while AML5 showed heightened activities of MTORC1 signalling pathways [37]. Our results demonstrate that AML promotes leukaemogenesis in CD8⁺ T cells through the upregulation of multiple hallmark pathways and highlight the need for personalised medicine to target upregulated transcription factors and oncogenic signalling pathways for AML treatment.

AML is often accompanied by a reduction in NK cell activity in peripheral blood [38]. We obtained approximately 4,395 NK cells from all samples (Fig. 4a), marked with KLRF1, NCAM1, and KLRD1 (Fig. 4b). Unlike T cells, UMAP spectral clustering on NK cells only results in single functional subtype showing NK cell marker gene expression homogeneity among AML patients and healthy donors. We also observed that the number of NK cells in AML patients was lower than that in healthy donors (Fig. 1e), indicating a potential role for NK cells in AML-induced suppression [38]. We then compared the DEGs in NK cells between AML and healthy donors. We observed the downregulation of 10 transcription factors in all six AML samples, including factors involved in homeostatic NK cell proliferation and survival, such as CEBPD, KLF3, KLF2, USF2, and FOXP1. (Fig. 4d). For other genes, we observed transcriptional heterogeneity among AML samples. As indicated in Fig. 4e, AML1 exhibited lower expression levels of the GIMAP4 gene, which encodes a protein belonging to the immune-associated nucleotide subfamily and is negatively regulated by T-cell acute lymphocytic leukaemia 1 (TAL1). In contrast, AML1 exhibited higher expression levels of genes encoding ribosomal proteins, including RPS26 and RPS4Y1 [39], which may promote cancer metastasis and spread, and their presence is correlated with increased disease aggressiveness and poor clinical outcomes [40]. Similar to AML1, AML3B and AML4 also upregulated some ribosomal genes (Supplementary Fig. 8). In AML2 and AML3, the upregulated genes contained S100A8 and S100A9, which are known to be poor prognostic indicators in AML, and elevated expression of the heterodimer S100A8/S100A9 was previously reported to cause glucocorticoid resistance in MLL-rearranged infant acute lymphoid leukaemia cells [41] (Fig. 4e and Supplementary Fig. 8). AML5 displayed upregulation of several genes related to encoding immunoglobulins (IGKC, IGHA1, ERAP2, and IGLC2, Supplementary Fig. 8) that participate in indirect cytotoxicity by recognising IgG antibodies attached to target cells, which is a key mechanism of action for various clinically successful anti-tumor therapeutic monoclonal antibodies (mAbs) [42]. Based on GSVA analysis (Fig. 4c), we observed high heterogeneity in gene set enrichment among AML samples, except for AML4, which had enriched gene sets similar to those of healthy samples. AML3 and AML5 showed heightened enrichment of pathways involved in anti-tumor response and lipid metabolism, including the interferon-gamma response, oxidative phosphorylation, adipogenesis, interferon-alpha response, and fatty acid metabolism pathways. In addition, AML2 was enriched with apoptosis and MTORC1 signalling pathways, whereas AML3B was enriched with P53 signalling pathways. These findings emphasise the negative impact of AML on cell number and anti-tumor activity of NK cells, as NK cells from healthy donors are activated with higher interferon-alpha and gamma response profiles. However, our results also demonstrated the potential of AML treatment by targeting the regulation of transcription factors, ribosomal proteins, and inflammatory signalling.

Monocytes are a heterogeneous group of cells originating from myelomonocytic precursors in the bone marrow, which contribute to host defence against pathogens and maintain normal tissue structure and function [43]. DCs, derived from monocytes, are essential for initiating and regulating immune responses. We analysed the heterogeneity of CD14⁺CD16⁺ monocytes associated with inflammation and AML disease progression in AML using UMAP analysis (Fig. 5a) [44]. We identified three major monocyte subsets in human peripheral blood based on the differences in phenotype, function, and expression of markers such as LYZ, FCN1, and C5AR1 [45]. The subclusters included CD14⁺ monocytes expressing high levels of CD14, CD16⁺ monocytes expressing high levels of myeloid markers such as FCGR3A, and monocyte DCs cells characterised by high CD1C, FCER1A, and CLEC10A expression (Fig. 5b). As shown in Fig. 4c, the proportion of CD14⁺ monocytes in AML patients (AML2, and AML5) was significantly higher than that in healthy controls. The results showed that patients with AML exhibited a significant increase the number of CD14⁺ monocytes, indicating an increase in pro-inflammatory responses. Consistent with the clinical status that AML2 and AML5 were both in myelosuppression stage after chemotherapy. This is also coordinate to the characteristics of CD14⁺ monocytes in other disease processes and suggest their potential clinical significance in assessing the AML process [44]. In contrast, there was no significant difference in the number of CD16⁺ monocytes between patients with AML and healthy controls. As for DCs, the number of DCs significantly increased in AML1, indicating an abnormally elevated immune response, consistent with the clinical diagnosis of a high-risk status (Fig. 5c).

Next, pySCENIC analysis observed upregulation of six transcription factors (BATF, ZNF250, STAT5B, NFIL3, DDIT3, and ETS2) associated with apoptosis and inflammation in AML2 patient, who is a high-risk patient in myelosuppression stage after chemotherapy (Fig. 5e). BATF was reported to have a higher expression level in the high-risk group than in the low-risk group of AML [46]. BATF regulates gene expression by acting as a pioneer transcription factor in lymphocytes and supports differentiation and cytokine production by Th2, Th17, follicular T helper, and Th9 cells [47, 48]. We also observed the enrichment of specific gene sets involved in the inflammatory response, apoptosis, IL6- JAK-STAT3 signalling, and IL2-STAT5 signalling in monocytes and DCs (Fig. 5d).

To explore the clonal structure of CD14⁺ monocytes and monocyte-DCs, we applied the inferCNV algorithm to analyse the CNVs of single cells. As a result, we observed a significantly higher CNV and chromosomal instability index among AML samples compared to healthy controls (Supplementary Fig. 9). For example, compared with healthy samples, the monocyte-DCs of AML1 had an obvious change, in which the CNVs of chromosome pairs 3, 5, 7, and 9 were significantly reduced, whereas the CNVs of chromosome pairs 2, 4, 11, and 19 increased, which may be caused by the increase in malignant tumor cells. Based on the aggregated CNV results of CD14⁺ monocytes, the common chromosome copies number alterations, such as the gains in chromosome 1 and the loss of chromosomes 5, 6, and 9, might be associated with chemotherapy response [49].

These findings emphasise the impact of AML on monocytes and DCs and suggest that the number of CD14⁺ monocytes, DCs, and CNVs can be used in tandem to assess AML clinical status during chemotherapy.

Standard chemotherapy for AML targets proliferating cells and effectively induces complete remission; however, many patients eventually experience relapse [50], which is caused by leukaemia stem cells that can self-renew. The self-renewal and proliferation of normal HSCs produce pluripotent progenitor cells (stem cells) and daughter cells that differentiate into effector cells [51]. To investigate the heterogeneity of HSCs and progenitor (Prog) cells in AML after chemotherapy, we performed GSVA analysis to compare the HSC-Prog genome in AML and healthy donors. Different from other AML samples, AML1 (a high risk AML patient in immune-recovery state after chemotherapy) upregulated in ten transcription factors, including EGR1 (a zinc finger protein that is associated with AML progression [52]), and leukaemia transcriptional regulators SMAD1 and ERF (Fig. 6b). We also observed the positive regulation of pro-inflammatory signalling pathways, including the p53 pathway, apoptosis pathway, and KRAS signalling (Fig. 6a). Upregulation of the p53 pathway may affect HSCs and transform HSCs into pre-leukemic stem cells with a significant risk of developing blood cancer, which would be concordant with the clinical results that AML1 has p53 mutations (Supplementary Table 1). AML with p53 abnormality is characterised by complex chromosomal aberrations [53]. Therefore, we performed CNV analysis on HSC-Prog using expression profiles and genomic information and noted that AML1 had an obvious gene copy number variations in most of the cell types (Supplementary Fig. 9), indicating that AML1 may produce more malignant cells.

Next, to determine the differences in structural cellular immunisations and immune–immune interactions between the cells derived from AML patients and healthy donors, we applied CellPhoneDB [54] to construct a cell–cell communication network via known ligand–receptor pairs within the 12 identified main clusters (Fig. 6c). Ligand–receptor pairs (p < 0.01) represent significant interactions between the two cell types. The results showed that in healthy donors, most intercellular interactions occurred between the following cell types: CD16-mono, NK, CD8⁺ T, Pro-T, CD14-mono, and mono-DCs (Fig. 6c). In contrast, in AML patients, the interaction between NK, CD14-mono, mono-DCs, and other cells was enhanced, and the range of interacting cells was expanded (Supplementary Fig. 10). Furthermore, HSC-prog cells showed more interactions with other cells through known ligand–receptor pairs, such as leukaemia-associated ligands CXCL2, CXCL3, and FLT3, signalling to the receptors CXCR1 and CXCR2 expressed by CD14-mono and NK cells [55, 56] (Fig. 6e). We also noted the enrichment of activation of the HGF-CD44 pathway between HSC-Prog and CD14-mono, CD16-mono, and mono-DCs in AML1. This suggests the role of HSC-Prog as a central communication hub in AML patients (Fig. 6d), highlighting its important role in AML development; moreover, it indicates that cell–cell communication is strengthened in the AML patient.

These findings emphasise the intra-tumoral heterogeneity at the transcriptional and cell–cell communication levels in HSC-prog, as well as highlight the need for more precise biomarkers to enhance chemotherapy efficacy. Our results also demonstrate the potential of targeting ligand–receptor pairs in HSC-Prog, such as CXCL2 and CXCL3 and their receptors CXCR1 and CXCR2, for AML diagnosis or treatment.

Five patients were pathologically diagnosed with AML at Shenzhen Peoples’ Hospital, 2^nd clinical medicine college, Jinan University (Fig. 1a). The available clinicopathological features of these patients were summarized in Supplementary Table 1, and six PMBC samples were designated as the discovery cohort. Six PBMC samples from healthy donors were collected and processed at Shenzhen Peoples’ Hospital. All patients and healthy donors had accepted the informed consent forms, and the study was approved by the Research Ethics Committee approval (LL-KY-2021271).

Single-cell suspensions were loaded into Chromium microfluidic chips using the Chromium Single Cell 3ʹ GEM, Library & Gel Bead Kit v3 (10X Genomics) according to the manufacturer’s instructions. All the subsequent steps were performed following the standard manufacturer’s protocols. The barcoded library was sequenced on the NovaSeq6000 sequencer (Illumina) with 150 bp paired-end reads by Novogene (China).

The raw scRNA-seq data were aligned to the human reference genome (GRCh38) and a digital gene expression matrix was built using STAR algorithm in CellRanger (‘count’ option; v3.1.0; 10 × Genomics). The ‘filtered_feature_bc_matrix’ file folder generated by CellRanger was used for further analysis. Using the Seurat R package [62] (v3.1.0; https://​satijalab.​org/​seurat), cells that contained more than 5500 or fewer than 200 expressed genes, or more than 15% mitochondrial transcripts, or more than 5% hemoglobin transcripts were removed. Genes that were expressed in fewer than 3 cells were also removed. In addition, doublets identified by DoubletFinder [63] (version 2.0.2; https://​github.​com/​chris-mcginnis-ucsf/​DoubletFinder) were also dropped. Following the removal of the poor-quality cells and doublets, a total of 91,772 cells were retained for downstream analysis.

For each cell, the expression of each gene was normalized and log-transformed using ‘NormalizeData’ function (scale.factor = 10,000) of Seurat. Then, the 12 samples were integrated by ‘FindIntegrationAnchors’ and ‘IntegrateData’ with 2000 variable genes. After data scaling using ‘ScaleData’, principal component analysis and UMAP analysis were performed for dimension reduction by ‘RunPCA’ and ‘RunUMAP’, respectively. Finally, cell clusters were identified using ‘FindNeighbors’ and ‘FindClusters’ (resolution = 3). The ‘FindAllMarkers’ function in Seurat was used to find markers of each identified cluster. Then, clusters were annotated based on the expressions of canonical markers in particular cell types [15, 64].

Different expression genes (DEGs) were identified using the ‘FindMarkers’ or ‘FindAllMarkers’ function in Seurat with default parameter. DEGs were filtered using a minimum log (fold change) of 0.25 and p_val_adj < 0.05. Gene set variation analysis (GSVA) was performed using 50 hallmark gene sets obtained from the molecular signature database (http://​www.​gsea-msigdb.​org/​gsea/​msigdb) by the GSVA package (v1.34.0) [64].

Cell–cell communication was analyzed using CellPhoneDB (v2.1.1, https://​github.​com/​Teichlab/​cellphonedb) with normalized count data as input file according to the literature [54]. The significant ligand-receptor pairs were filtered with a P value of less than 0.05.

Regulons of transcription factors (TFs) and their target genes were analyzed by pySCENIC (v0.10.0, https://​github.​com/​aertslab/​pySCENIC) with raw count matrix as input [65]. Briefly, the coexpression modules were inferred by GRNBoost2, then the regulons were identified by cisTarget, and the activity of these regulons was quantified by AUCell.

Single-cell CNV analyses were estimated by the inferCNV package (v1.2.0; https://​github.​com/​broadinstitute/​inferCNV/​wiki) with a raw count matrix as input. The cells of the same type in the healthy group were used as reference. In addition, the parameters of inferCNV analysis were set as default.