Inside the Battle Against Acute Myeloid Leukemia: Biology, Breakthroughs, and Hope.

1. Introduction
Acute myeloid leukemia (AML) is an aggressive hematologic malignancy characterized by clonal expansion of myeloid progenitor cells with impaired differentiation and aberrant self-renewal capacity [1,2,3,4,5]. Clinically and biologically, AML exhibits remarkable heterogeneity, encompassing diverse genetic alterations, epigenetic states, cellular hierarchies, and altered interactions with the bone marrow (BM) microenvironment [6,7]. Recent studies using single-cell and multi-omics technologies have revealed that this heterogeneity is highly dynamic, shaped by clonal evolution and context-dependent cell state transition rather than traditional lineage hierarchies. This heterogeneity underlies the variable disease course and therapeutic response observed among patients and highlights the complexity of the molecular programs governing malignant myelopoiesis [4,8].
This review follows the progression from normal hematopoiesis to malignant transformation in AML. It begins with the key cellular and molecular processes underlying normal hematopoiesis, then examines the cellular origins of AML and the genetic and epigenetic changes that drive leukemogenesis, integrating recent insights into clonal architecture, mutational order, and epigenetic reprogramming. Subsequent sections explore altered signaling pathways, leukemic stem cell (LSC) biology, intratumoral heterogeneity, and the role of the BM microenvironment in disease progression. Finally, it highlights emerging technologies that are reshaping AML research, including multi-omics approaches, single-cell and spatial analyses, and computational modeling. These methodologies have enabled system-level dissection of AML heterogeneity, reconstruction of cellular hierarchies, and identification of context-dependent vulnerabilities. Together, this framework offers an integrated, mechanism-focused view of AML and provides a foundation for future therapeutic advances.
This review was conducted based on a comprehensive literature search of the PubMed database. Relevant articles published mainly between 2010 and 2026 were retrieved using combinations of keywords such as “acute myeloid leukemia”, “hematopoiesis”, and “leukemia stem cells”. In our selection process, priority was given to seminal works, highly cited articles, and studies published in authoritative journals to ensure the reliability and relevance of the data; conference abstracts and non-original research were excluded. The reference lists of selected articles were also manually screened to identify additional relevant studies.

2. Normal Hematopoiesis: A Cellular and Molecular Overview
Hematopoiesis is hierarchically organized, with multipotent stem and progenitor cells giving rise to distinct myeloid and lymphoid lineages. At the apex of this hierarchy are hematopoiesis stem cells (HSCs), characterized by their capacity to self-renew and differentiate into multiple lineages (Figure 1, left panel). Recent single-cell analyses have challenged the classical stepwise model of hematopoietic differentiation, instead supporting a continuum model in which fate commitment arises gradually through intermediate transcriptional states [9,10,11,12].
HSC differentiation is tightly regulated by a complex network of transcription factors, epigenetic regulators and signals from the BM niche. In HSCs, a core set of seven transcription factors, namely FLI1, ERG, GATA2, RUNX1, TAL1, LYL1, and LMO2, known as the “heptad”, co-occupies the genome to form a highly coordinated and tightly interconnected regulatory network [13,14]. Integrated multi-omics approaches combining ChIP-seq, Hi-C, HiChIP, and histone modification profiling have shown that at the HSC–multipotent progenitor (MPP) stage, these factors extensively bind thousands of promoters and distal enhancers, establishing an open and reprogrammable regulatory landscape that maintains multipotency.
As hematopoietic stem and progenitor cells (HSPCs) undergo lineage specification, heptad binding undergoes pronounced lineage-specific remodeling. Specifically, TAL1 and GATA2, together with LYL1 and LMO2, show enhanced occupancy in megakaryocyte–erythroid regulatory regions, whereas FLI1, ERG, and RUNX1 are more active in myeloid-specific regions [13]. These shifts reflect a system-wide reorganization of transcription factor assemblies, regulatory elements, and chromatin architecture during lineage commitment [15]. Enhancers initially reinforced by the heptad are later taken over by lineage-defining transcription factors, such as PU.1 in myeloid differentiation and GATA1 in erythroid differentiation [13,16,17]. Together, these observations support a three-layer model of transcriptional control during hematopoietic lineage specification: an initial priming phase, in which the heptad establishes a multipotent regulatory framework; a remodeling phase, characterized by lineage-specific enhancer–promoter reconfiguration; and a final activation phase driven by lineage-specific transcription factors [13,15].
Epigenetic regulation further modulates lineage bias in HSCs. For example, loss of DNA methyltransferases DMNT1 or DMNT3A induces hypomethylation, skewing differentiation toward the erythroid lineage [18,19], while loss of the ten-eleven translation enzymes (TET2) favors myelomonocytic differentiation and reduces erythroid and common lymphoid progenitors [20,21]. Beyond intrinsic regulatory, HSC maintenance and fate decision are influenced by the bone marrow (BM) niche, a complex microenvironment comprising hematopoietic and mesenchymal cells, including endothelial cells, mesenchymal stem cells (MSCs), macrophages, and megakaryocytes, which provides structural support, chemokines, growth factors, and metabolic cues that regulate HSC quiescence, self-renewal, proliferation and differentiation [22,23,24,25,26].
Notably, a 2024 study integrated scRNA-seq with spatial proteomics (CODEX multiplex imaging) and generated a high-resolution spatial map of human BM, identifying a hypoxic artero-endosteal region enriched for early myeloid progenitors while peri-adipocytic zones preferentially harbored HSPCs. This finding highlights a strong correlation between the microenvironment architecture and lineage outcomes [27].

3. Cellular Origin of AML
AML is an aggressive hematologic malignancy originating from hematopoietic stem and progenitor cells (HSPCs) that acquire somatic mutations, ultimately conferring a proliferative advantage and clonal expansion. The process by which HSPCs accumulate somatic mutations and expand clonally is referred to as clonal hematopoiesis (CH) [28]. Over the past decade, advances in next-generation sequencing (NGS) technologies, including whole-exome sequencing (WES), whole-genome sequencing (WGS), and single-cell sequencing, have enabled high-resolution characterization of clonal architecture in clonal hematopoiesis [29,30,31]. The most common mutations are epigenetic regulators, including DNMT3A, TET2 and ASXL1, which are considered driver mutations and account for approximately 80~90% of CH-associated variants.
CH encompasses diverse subtypes. A clinically significant subset, termed clonal hematopoiesis of indeterminate potential (CHIP), is defined by the presence of somatic mutations in leukemia-associated genes at a variant allele fraction (VAF) of ≥2% in individuals without overt hematologic malignancy or cytopenia [28,32,33]. HSCs harboring CHIP-associated mutations may be considered as pre-leukemia stem cells (pre-LSCs), particularly when they acquire additional leukemia-specific mutations that drive progression to AML [34,35,36]. Accordingly, leukemogenesis can be conceptualized as a two-stage process: initiation in pre-LSCs and transformation into full leukemia stem cells (LSCs).
Mutations in epigenetic regulator genes, such as DNMT3A, TET2 and ASXL1, collectively referred to as “DTA” mutations, are frequently enriched in pre-LSCs, placing these cells in a transcriptionally and functionally “primed” state. Acquisition of additional mutations in genes involved in DNA damage response (DDR; e.g., TP53, CHEK2), RNA splicing (SRSF2, SF3B1, and U2AF1), transcriptional regulation (RUNX1, GATA2), and signaling pathway genes (JAK2, KRAS, PTPN11, FLT3) facilitates the transition from pre-LSCs to LSCs, which exhibit enhanced self-renewal, proliferative capacity, and impaired differentiation, ultimately driving AML development within the patient’s bone marrow [37,38,39] (Figure 1, right panel).
Given their central role in disease propagation, LSCs represent a critical therapeutic target in AML. A precise definition of LSCs, coupled with identification of specific biomarkers, is essential for guiding the development of targeted therapies. Finally, LSCs are distinguished from bulk leukemic blasts by their long-term self-renewal capacity and ability to reconstitute the entire leukemia hierarchy in xenotransplantation models [35,40]. Phenotypically, LSCs are heterogeneous; the classical immunophenotype is CD34+CD38−, but additional markers, including CD123, TIM3, C96, CLL-1, CD25, CD32 and CD47, have been identified to distinguish LSCs from normal HSCs. Several of these markers have been validated as functional and possess potential therapeutic relevance [41,42,43,44,45,46,47].

4. Genetic Mechanisms of Leukemogenesis
Building on the framework of pre-LSCs and LSCs, the pathogenesis of AML is now recognized as a multilayered process involving coordinated alterations in genetic, epigenetic, splicing, transcriptional, and high-order chromatin regulatory programs. This expanded understanding has driven systemic efforts to classify AML-associated mutations, enabling improved disease stratification and informing diagnostic and therapeutic guidelines.
Early models of AML leukemogenesis were largely based on the “two-hit” hypothesis, in which mutations were broadly categorized into class I or class II. Class I mutations, such as activating alterations in FLT3 or RAS, primarily confer proliferative and survival advantages, resulting in hyperproliferation but remaining insufficient for complete leukemic transformation. In contrast, class II mutations, typically involving transcription factors or fusion oncogenes such as PML/RARα, AML1/ETO, CEPBA, RUNX1-RUNX1T1, and CBFB-MYH11, impair hematopoietic differentiation and cooperate with class I mutations to induce overt AML [48,49].
In 2013, a landmark study published in The New England Journal of Medicine (NEJM) performed comprehensive genomic profiling of 200 paired de novo adult AML samples and categorized recurrent drive mutations into nine functional classes (Table 1). These include transcription factor fusions (PML-RARA, MYH11-CBFB, RUNX1-RUNX1T1, PICALM-MLLT10); nucleophosmin (NPM1) mutations; tumor suppressor genes (TP53, WT1, PHF6); DNA methylation regulators (DNMT3A, DNMT3B, DNMT1, TET1, TET2, IDH1, IDH2); activated signaling pathways (FLT3, KIT, other Tyr-kinases, Ser-Thr-kinases, KRAS/NRAS, PTPs); myeloid transcription factors (RUNX1, CEBPA, other myeloid TFs); chromatin modifiers (MLL-X fusions, MLL-PTD, NUP98-NSD1, ASXL1, EZH2, KDM6A, and other modifiers); cohesion genes; and spliceosome genes [29]. This study revealed extensive patterns of mutation co-occurrence and mutual exclusivity, highlighting distinct leukemogenic trajectories (Figure 2).
One of the most striking observations was the frequent co-occurrence of FLT3, DNMT3A, and NPM1 mutations. AML cases harboring this triad formed a distinct molecular subgroup characterized by unique mRNA, miRNA, and DNA methylation signatures, suggesting the existence of a discrete AML subtype, a finding consistently validated in subsequent studies [29,50,51,52]. In contrast, core-binding factors (CBF) and PML-RARA fusion leukemias, as well as KMT2A/MLL rearrangements, were largely mutually exclusive with NPM1 and DNMT3A mutations. Similarly, RUNX1 and TP52 mutations showed mutual exclusivity with FLT3 and NPM1 alterations [53,54].
These patterns of mutual exclusivity indicate that certain transcription factor fusions are individually sufficient to drive leukemogenesis. Mechanistically, PML-RARA arising from (t(15;17)) functions as a dominant transcriptional repressor by recruiting corepressor complexes (e.g., NCOR/SMART), disrupting PML nuclear bodies and enforcing a block in granulocytic differentiation [55,56]. The RUNX1-RUNX1T1 fusion protein generated by (t(8;21)) recruits HDACs, DNMTs, and other corepressor complexes to RUNX1 target loci, resulting in widespread repression of myeloid differentiation programs and maintenance of a self-renewing progenitor state [57]. The CBFB-MYH11 fusion associated with (inv(16)) sequesters RUNX1 (the α-subunit of CBF) in cytoplasmic complexes through interactions with cytoskeletal and myosin-associated proteins, thereby abrogating nuclear core-binding factor activity and perturbing downstream differentiation pathways [54]. In KMT2A/MLL-rearranged AML, aberrant recruitment of cofactors such as Menin and DOT1L leads to ectopic H3K4 and H3K79 methylation, driving sustained activation of oncogenic transcriptional programs, including HOXA/B, MEIS1 and PBX3, imposing a stem-cell-like chromatin state associated with aggressive, differentiation-blocked leukemia [58,59].
Collectively, these observations underscore that the mutational architecture of AML reflects functional cooperation among distinct molecular lesions. A prototypical example is the synergistic interaction between DNMT3A loss-of-function mutations and FLT3-ITD signaling. In murine models, Dnmt3a deficiency induces widespread DNA hypomethylation, including enhancer demethylation at key stemness-associated loci such as MEIS1, thereby establishing a permissive epigenetic landscape that primes HSCs but is insufficient for leukemic transformation. Subsequent acquisition of FLT3-ITD provides constitutive proliferative and survival signals, promoting expansion and enhanced self-renewal of pre-LSCs, and ultimately driving overt AML development [60,61,62,63]. This cooperative mechanism is strongly supported by clinical data, as patients harboring concurrent DNMT3A and FLT3 mutations exhibit pronounced hypomethylation signatures and significantly inferior clinical outcomes [53,64]. Together, these findings highlight a paradigmatic model of synergetic leukemogenesis in AML, in which epigenetic instability mediated by DNMT3A mutations collaborates with oncogenic signaling pathways to promote malignant transformation and disease aggressiveness. Such functional cooperation explains the recurrent co-occurrence of specific mutation combinations in AML and underscores why individual lesions are rarely sufficient to induce leukemia in isolation. Understanding these cooperative genetic networks is essential for the rational design of combination therapeutic strategies.

5. Epigenetic Dysregulation in AML
Epigenetic dysregulation including DNA methylation, histone modification, chromatin remodeling, and noncoding RNA and RNA modifications constitutes a fundamental hallmark of AML, profoundly influencing leukemogenic mechanisms, leukemia classification and therapeutic development. Genomic studies have demonstrated that more than 60% of AML cases harbor mutations involved in epigenetic dysregulation [29,52]. These alterations strongly influence leukemic gene expression programs and contribute to disease initiation and progression, making epigenetic regulators attractive targets for AML therapy.

5.1. DNA Methylation and Demethylation
AML exhibits profound mutations in DNA methylation patterns. DNMT3A, a de novo DNA methyltransferase responsible for catalyzing 5′-cytosine methylation, is one of the most frequently mutated epigenetic regulators in AML [65]. Somatic mutations in DNMT3A were first reported in 2010, and are present in over 20% of AML patients, with the R882 hotspot mutation being the most relevant [66]. Later-scale population-based analyses, including statistical modeling of the UK Biobank data, have further identified DNMT3A mutations as a significant risk factor for the development of AML [67,68]. Mechanistically, the DNMT3AR882H mutation, located within the catalytic domain, exerts a dominant-negative effect by impairing the formation of enzymatically active DNMT3A tetramer. This results in reduced methyltransferase activity and locus-specific DNA hypomethylation, leading to repression of key differentiation-associated transcription factors such as CEBPA and PU.1 [69,70]. In addition, DNMT3AR882H preferentially induces hypomethylation at polycomb repressive complex 2 (PRC2) target regions, and promotes aberrant activation of self-renewal programs, including enhanced binding at MYC-associated motifs [71].
In contrast, TET2 functions as a DNA demethylation enzyme that catalyzes the oxidation of 5-methylcytosine and thereby acts as a key regulator of DNA methylation dynamics. Loss-of-function mutations in TET2 are detected in approximately 15% of patients with myeloid malignancies [72]. Mechanistically, TET2 shapes enhancer methylation landscapes and cooperates with lineage-defining TFs, such as PU.1 and RUNX1, to regulate myeloid differentiation [73,74]. Recent work using Tet2 knockout mouse models identified SOX4 as a critical downstream driver, activating NOTCH, FGF and PI3K signaling pathways and conferring selective clonal advantage contributing to Tet2-deficient hematopoietic cells [75].
Mutations of isocitrate dehydrogenase (IDH) genes were first identified in 2009 through AML genome sequencing [76]. IDH1 and IDH2 encode metabolic enzymes that link cellular metabolism with epigenetic regulation. Recurrent mutations in IDH1 (R132) and IDH2 (R140, R172) result in neomorphic enzyme activity that produces the oncometabolite 2-hydroxyglutarate (2-HG) instead of α-ketoglutarate (α-KG). Accumulation of 2-HG competitively inhibits α-KG-dependent dioxygenases, including TET family enzymes and lysine histone demethylases, leading to impaired DNA and histone demethylation and global DNA hypermethylation [77,78,79,80]. CHIP-seq analyses have further revealed that IDHmut-specific hypermethylation is enriched at active enhancers regions, including loci associated with MYC and ETV6, which directly interact with key AML oncogenic programs [81]. Moreover, 2-HG inhibits α-KG-dependent alkB homolog (ALKBH) DNA repair enzymes, resulting in increased DNA damage accumulation and contributing to genomic instability [82]. Collectively, IDH mutations disrupt epigenetic homeostasis and DNA repair mechanisms in HSCs, thereby impairing cell differentiation and promoting dysregulated proliferation that drives leukemogenesis (Figure 3, left box).

5.2. Histone Modifications and Chromatin Remodeling
Mutations in genes regulating histone modifications disrupt chromatin organization and represent key drivers of AML. Among these, ASXL1, EZH2, and KMT2A are frequently mutated and lead to profound alterations in transcriptional programs controlling hematopoietic differentiation and self-renewal.
ASXL1 mutations result in a global reduction in H3K27me3 by impairing recruitment of the polycomb repressive complex2 (PRC2) to oncogenic loci, including the HOXA cluster, thereby promoting myeloid transformation [83]. In addition, ASXL1 interacts with O-GlcNAc transferase (OGT) to regulate H3K4 methylation, and disruption of this axis impairs normal myeloid differentiation [84]. EZH2, the catalytic subunit of PRC2, mediates H3K27me3-dependent gene repression and exhibits context-dependent roles in cancer. While gain-of-function EZH2 mutations in lymphomas enhance silencing of tumor suppressor genes [85,86], loss-of-function EZH2 mutations are common in myeloid malignancies. In AML, EZH2 functions as a tumor suppressor during disease initiation by repressing fetal oncogenic programs such as PLAG1, but later supports disease maintenance by sustaining repression of differentiation-associated genes and leukemic stem cell programs [87,88]. KMT2A, also known as MLL1, encodes a histone methyltransferase that catalyzes H3K4me3 and regulates HOX gene transcription through interactions with the Menin–LEDGF complex [89,90,91]. In KMT2A-rearranged leukemias, the C-terminal SET domain is replaced by fusion partners such as AFF1, MLLT3, MLLT10, MLLT1, or ELL [91]. These fusion proteins retain chromatin-targeting activity and aberrantly recruit transcriptional elongation machinery and DOT1L to HOXA/MEIS1 loci, leading to increased H3K79 methylation, sustained oncogene expression, and leukemic self-renewal [89,92,93] (Figure 3, right box).
Dysregulation of PRCs further alter chromatin states in AML. PRC1-mediated H2AK119 ubiquitination and PRC2-mediated H3K27me3 cooperate to repress differentiation-associated genes, thereby maintaining leukemic self-renewal [93,94]. Increased activity of PRC1 components RING1A and RING1B enhances repression of differentiation programs in AML [95,96]. PRC2 consists of EZH1/2, SUZ12, EED, and RBBP4/7, and AML-associated mutations in these subunits destabilize the complex and reduce H3K27me3 levels [94]. This leads to de-repression of stemness-related genes such as HOXA9, MEIS1, and PLAG1, promoting leukemogenesis [87,88,94]. Many PRC target genes exhibit bivalent chromatin marked by both H3K4me3 and H3K27me3; disruption of this balance maintains leukemic cells in a differentiation-blocked, stem-like state [97,98].
Aberrant activation of super-enhancers also plays a central role in AML. Leukemia cells hijack super-enhancer machinery involving transcription factors, p300/CBP, BRD4, and SWI/SNF to drive oncogenes such as MYC and promote self-renewal [99,100,101,102,103]. Chromosomal rearrangements can rewire enhancer landscapes, as seen in inv(3)/t(3;3) AML, where a GATA2 enhancer activates EVI1 while causing GATA2 haploinsufficiency [104]. Additionally, TRIB1 remodels Hoxa9-associated super-enhancers by promoting C/EBPα degradation, increasing H3K27ac and oncogenic transcription [105]. The super-enhancer-associated gene CAPG promotes AML progression through NF-kB signaling and is associated with poor prognosis [101]. Although super-enhancers are clearly critical for AML pathogenesis, their therapeutic targeting and interactions with other genetic lesions remain areas of active investigation.

5.3. Noncoding RNAs and RNA Modifications
Noncoding RNAs (ncRNAs) have emerged as critical regulators of gene expression and play integral roles in the pathogenesis of AML. Based on transcript length and structural features, ncRNAs are broadly classified into microRNAs (miRNAs), long noncoding RNAs (lncRNAs) and circular RNAs (circRNAs).
miRNAs are small, ~22-nucleotide RNA molecules that post-transcriptionally repress gene expression by binding to complementary sequences within the 3′ untranslated region (3′-UTRs) of target mRNAs. Early profiling studies revealed distinct, AML subtype-specific miRNA expression signatures that are closely associated with leukemogenesis and clinical outcomes [106,107,108]. In FLT3-ITD AML, miR-155 is transcriptionally induced by p65 and STAT5, subsequently suppressing the myeloid transcription factor PU.1 and thereby impairing myeloid differentiation [109,110]. miR-196b exerts a complex- and stage-dependent function in MLL-rearranged leukemia and normal hematopoiesis by simultaneously targeting oncogenic factors (HOAX9/MEIS1) and the proapoptotic tumor suppressor FAS [111,112]. Notably, miRNAs can exhibit opposing roles depending on the AML subtype. For example, miR-9 functions as an oncogene in MLL/AF9-driven leukemia [106], whereas in pediatric AML harboring t(8;21) translocations or EVI1 overexpression, miR-9 acts as a tumor suppressor by inhibiting leukemic progress and promoting differentiation [113,114].
circRNAs are covalently closed, highly stable RNA molecules generated through back-splicing during pre-mRNA processing. Emerging evidence indicates that dysregulated circRNA expression contributes to AML pathogenesis and may serve as a prognostic biomarker or therapeutic target [115]. Bioinformatic analyses have identified circ-0004277 as one of the most significantly downregulated circRNAs in AML [116]. Fusion gene-driven -circRNAs, including PML/RARα-driven f-circPR and MLL/AF9-derived f-circM9, promote leukemia cell survival and disease progression [117]. Furthermore, exosomal circ_0006896 has been shown to interact with HDAC1, leading to reduced H3 acetylation, suppression of lipid peroxidation and inhibition of ferroptosis in AML cells [118].
lncRNAs are transcripts longer than 200 nucleotides that regulate gene expression through diverse mechanisms, including chromatin remodeling, transcriptional regulation, and modulation of RNA stability and translation. HOTAIRM1, a well-characterized lncRNA transcribed from the HOXA cluster, plays a pivotal role in myeloid differentiation [119,120]. In addition, several lncRNAs, such as UCA1, NEAT, MEG3, and MALAT1, have been shown to function either as oncogenes or tumor suppressors in AML, depending on cellular context and genetic background [121,122].
Collectively, ncRNAs constitute a complex and multilayered regulatory network that orchestrates gene expression programs for AML initiation, maintenance, and progression. Elucidation of ncRNA-mediated regulatory mechanisms not only deepens our understanding of AML biology but also offers foundation for the development of ncRNA-based diagnostic indictors and targeted therapeutic strategies.
Beyond ncRNAs, N6-methyladenosine (m6A) RNA modification has emerged as an additional critical epitranscriptomic layer of gene regulation in AML [123,124,125]. m6A modification is dynamically regulated by methyltransferase complexes (writers), such as METTL3 and METTL14; demethylases (erasers), including FTO; and m6A-binding proteins (readers), such as YTHDC1/2, YTHDF2, and IGF2BP2. METTL3 is frequently overexpressed in AML and sustains leukemic cell proliferation while blocking myeloid differentiation by enhancing m6A-dependent translation of key oncogenic transcripts, including MYC, BCL2, and SP1 [126,127]. Conversely, the m6A demethylase FTO acts as an oncogenic driver by reducing m6A levels on target mRNAs, such as ASB2 and RARA, thereby promoting leukemogenic programs [128]. Notably, Wenlong Li and colleagues developed a selective FTO degrader that suppresses AML progression by increasing m6A modifications on ribosome-biogenesis-related mRNAs, promoting their YTHDF2-mediated decay, disrupting protein translation and highlighting FTO as a promising therapeutic target in AML [129].

5.4. Chromatin Architecture and 3D Genome Organization
Chromatin 3D architecture alternation driven by AML mutations has become a critical layer of epigenetic regulation in recent years. Chromatin is highly hierarchical, organized into A/B compartments, topologically associating domains (TADs), and chromatin loops, largely sustained by architectural proteins including CTCF, cohesion, lamins and the Mediator complex [130,131,132,133]. In AML, accumulating evidence has shown that chromatin architecture is profoundly changed. Hi-C-based studies have revealed AML-specific chromatin loops and expanded, shrunken and shifted TAD boundaries in AML compared to HSPCs. WGS data also showed enhancer hijacking in genes such as HSF4, MYC, CBL and POU2F3, as well as silencer hijacking affecting AML-related genes including JAK1 and KMT2C, which regulate gene expression [132]. Mutations in chromatin architectural components also play role in remodeling the 3D genome. Cohesion component STAG2, which functions as a leukemogenic potential mutation, shapes chromatin interaction through RAD21 and CTCF binding, resulting in increased HSC self-renewal and disturbed differentiation [134].
Importantly, alterations in 3D genome organization are tightly incorporated with other epigenetic regulations, including DNA methylation, histone modification and noncoding-RNA-mediated regulation. Research has demonstrated that genes in the A compartment have lower methylation at transcription starting site (TSS) but higher methylation in gene bodies than genes in the B compartment. The impact of DNA methylation on the 3D genome structure can be reversed by DNMT3A/3B/1 triple-knockdown (TKD) in the U937 AML cell line [132].
Altogether, structural variations, fusion oncoproteins, and mutations in chromatin regulators in AML collectively reshape chromatin topology, thereby stabilizing leukemic transcriptional states, promoting leukemic transcriptional programs, and contributing to leukemogenesis.

6. Disrupted Signaling Pathways in AML
Accumulating evidence indicates that dysregulated signaling pathways are fundamental drivers of AML, governing leukemic cell survival, apoptosis, proliferation, and stemness. These pathways play pivotal roles in leukemogenesis and represent key targets for therapeutic intervention.

6.1. Proliferation and Survival Pathways
Receptor tyrosine kinases (RTKs) are central regulators of cellular proliferation, survival, apoptosis, and differentiation. Among them, c-KIT and FMS-like tyrosine kinase 3 (FLT3), members of the class III protein tyrosine kinase (PTK) family, are of particular importance in acute myeloid leukemia (AML), as their aberrant activation is frequently associated with adverse clinical outcomes. c-KIT is commonly overexpressed or mutated in AML, while FLT3 internal tandem duplication (FLT3-ITD) mutations represent one of the most prevalent genetic alterations in the disease. Constitutive activation of c-KIT and FLT3-ITD leads to sustained signaling through key oncogenic pathways, including PI3K/AKT, RAS/RAF/MEK/ERK, and JAK/STAT, thereby promoting leukemic cell proliferation, survival, and resistance to apoptosis [48,135,136,137,138]. Among downstream effectors, STAT5 signaling plays a particularly critical role in maintaining leukemia stem cell (LSC) properties. Genetic or pharmacological suppression of STAT5 in CD34+ AML cells significantly impair long-term repopulating capacity, underscoring its essential function in LSC self-renewal and disease persistence [139,140].
In addition to RTK-driven signaling, activating mutations in small GTPases of the RAS family constitutes another major oncogenic axis in AML. The canonical RAS gene family includes KRAS, NRAS, and HRAS, with KRAS and NRAS being the most frequently mutated in AML [141,142]. Oncogenic RAS mutations result in constitutive activation of downstream effector pathways, including MAPK and PI3K-AKT signaling, as well as Ral GTPase-mediated processes, collectively driving aberrant cell proliferation, survival, and metabolic adaptation [143,144].
Importantly, AML progression is not solely dictated by intrinsic genetic alterations but is profoundly influenced by the BM microenvironment, which cooperates with oncogenic signaling to support leukemic cell survival and expansion. Through ligand–receptor interactions such as CXCL12/CXCR4, VEGF/VEGFR, and integrin-mediated adhesion, the BM niche robustly activates core signaling pathways, including RAS-MAPK, PI3K-AKT, NF-kB, and mTOR cascades [15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,147]. Upregulation of E-selection in AML bone marrow further enhances leukemic blast survival by stimulating MAPK/ERK and PI3K/AKT signaling [148,149]. Notably, microenvironment-derived cues synergize with recurrent AML mutations, such as FLT3-ITD, to amplify downstream signaling output and promote leukemic cell growth and therapy resistance [150,151]. Collectively, the dynamic interplay between intrinsic genetic drivers and extrinsic BM microenvironmental signals converges on shared proliferation and survival pathways, underscoring the niche as a critical co-driver of AML pathogenesis [152,153].

6.2. Self-Renewal and Stemness Pathways
The maintenance and malignant expansion of LSCs are critically dependent on the aberrant activation of developmental signaling pathways, such as Wnt/β-catenin, NOTCH, and Hedgehog (Hh) signaling, which collectively confer stemness and self-renewal capacity. Dysregulation of these pathways enables LSCs to sustain long-term leukemic propagation and evade differentiation cues. In the canonical Wnt/β-catenin pathway, stabilization of β-catenin leads to its cytoplasmic accumulation and subsequent nuclear translocation, where it binds to TCF/LEF family transcription factors. This transcriptional complex activates downstream target genes, including MYC, CCND1 and ABCB1, thereby promoting LSCs self-renewal, proliferation, and drug resistance [154,155]. Multiple studies have identified epigenetic silencing of Wnt pathway antagonists, such as secreted frizzled-related proteins (SFRPs) and DKK1, through promoter hypermethylation as a mechanism driving constitutive Wnt pathway activation in AML [155,156,157]. Pharmacological or genetic inhibition of β-catenin signaling, including modulation of GSK3β activity, significantly reduces AML cell proliferation and impairs LSC self-renewal capacity [158].
The NOTCH signaling pathway is initiated by the interaction between Delta-like (DLL) or Jagged-like (JAG) ligands and Notch receptors, resulting in the release of the Notch intracellular domain and transcriptional activation of downstream effectors. In contrast to its oncogenic role in some hematologic malignancies, NOTCH signaling appears to be attenuated in AML. Reduced expression of canonical NOTCH target genes, including HES1 and DTX1, has been observed in AML, suggesting a state of suppressed NOTCH pathway activity [159]. Epigenetic repression, characterized by increased H3K27 trimethylation at promoters of NOTCH target genes, has been proposed as a major mechanism underlying this suppression. Functionally, NOTCH inhibition promotes expansion of multipotent progenitors, whereas enforced activation of NOTCH signaling restricts AML growth and LSC maintenance, supporting a tumor-suppressive role for NOTCH in AML [159,160].
Aberrant activation of the Hedgehog (Hh) signaling pathway represents another critical mechanism sustaining LSC survival and self-renewal in AML. Canonical Hh signaling culminates in the activation of GLI family transcription factors, which regulate genes involved in stemness, quiescence, and survival [161,162]. Hh signaling not only directly supports LSC self-renewal but also modulates interactions between LSCs and the BM microenvironment, facilitating maintenance of quiescence and protection from chemotherapy-induced cytotoxicity [163,164].
At the transcriptional level, the LSC stemness phenotype is frequently driven by the aberrant expression of key stemness-associated genes, most notably HOXA9 and MEIS1. These Homeobox transcription factors are highly expressed in MLL-rearranged AML and other high-risk AML subtypes, including NPM1-mutated AML [165,166]. The synergistic action of HOXA9 and MEIS1 is essential for maintaining LSC self-renewal, leukemia propagation, and differentiation blockade [167,168]. Therapeutic targeting of this axis has therefore emerged as a promising strategy. Menin inhibitors, such as Revumenib and Ziftomenib, disrupt the Menin–KMT2A interaction, leading to suppression of both HOXA9 and MEIS1 transcription and induction of LSC differentiation and leukemic cell apoptosis [58,98,169].

6.3. Apoptosis and Cell Cycle Dysregulation
Dysregulation of apoptosis and cell cycle control is a fundamental mechanism underlying leukemogenesis and therapeutic resistance in AML. Under physiological conditions, the tumor suppressor p53 functions as a central cellular stress sensor that preserves genomic integrity by orchestrating cell cycle arrest, DNA repair, apoptosis, or senescence in response to oncogenic or genotoxic stress [170,171]. In AML, however, disruption of the p53 pathway enables leukemic cells to evade apoptosis and bypass cell cycle checkpoints. TP53 mutations occur in approximately 5~10% of AML cases and are strongly associated with adverse clinical outcomes. AML cells circumvent p53-mediated tumor-suppressive functions primarily through two mechanisms. First, TP53 mutations result in the accumulation of transcriptionally inactive p53 protein, which correlates with poor prognosis. Second, functional inactivation of p53 frequently occurs in the absence of TP53 mutations through overexpression of its negative regulators, MDM2 and MDM4. Notably, MDM2 overexpression can phenocopy TP53 loss and is associated with similarly unfavorable outcomes. Consistent with impaired p53 signaling, a substantial proportion of AML samples exhibit loss of p21 expression, directly reflecting disruption of p53-dependent cell cycle arrest [172,173,174].
Apoptotic signaling in AML is predominantly regulated by the BCL-2 family of proteins, which governs the intrinsic mitochondrial apoptotic pathway. This family comprises anti-apoptotic members (including BCL-2, BCL-XL, MCL-1, BCL-B, BCL2A1, and BCL-W) and pro-apoptotic members, which include the multidomain effectors BAX and BAK, as well as BH3-only proteins such as PUMA, NOXA, BIM, BID, BAD, BIK, and others. The balance between these opposing factions determines cell fate. BH3-only activator proteins (e.g., BID, BIM, PUMA, and NOXA) directly activate BAX and BAK, leading to mitochondrial outer-membrane permeabilization (MOMP), cytochrome c release, caspase activation, and execution of apoptosis. In contrast, BH3-only sensitizer proteins (e.g., BAD, BIK, and HRK) promote apoptosis indirectly by sequestering anti-apoptotic BCL-2 family members, thereby liberating activator proteins to engage BAX and BAK [175,176].
Early studies demonstrated that BCL-2 is frequently overexpressed in AML and is associated with poor prognosis and chemo-resistance [177]. Therefore, pharmacological targeting of BCL2 has emerged as a highly effective therapeutic strategy. Small-molecule BCL2 inhibitors, such as venetoclax, function as BH3 mimetics that displace BAX/BAK from anti-apoptotic sequestration, thereby restoring mitochondrial apoptotic signaling and inducing leukemic cell death [178,179]. Nevertheless, both primary and acquired resistance remain significant clinical challenges, arising from compensatory mechanisms, such as upregulation of alterative anti-apoptotic proteins (MCL1 and BCL-xL), mitochondrial reprogramming, and metabolic adaptions that sustain leukemic cell survival [180,181,182].

7. Leukemic Stem Cells and Intratumoral Heterogeneity
In AML, LSCs originate from diverse oncogenic drivers and pre-LSCs populations, resulting in substantial genetic, epigenetic and cellular heterogeneity. This heterogeneity is closely associated with differential responses to targeted therapies, variable disease trajectories, adverse clinical outcomes, and frequent relapse. Consequently, a detailed characterization of the phenotypic and functional properties of LSCs is critical for defining distinct LSC subtypes and for enabling precision-based therapeutic strategies in AML [183,184].

7.1. Phenotypic Heterogeneity
Historically, AML LSCs have been operationally defined by a CD34+CD38− immunophenotype [185]. However, increasing evidence indicates that LSCs are highly heterogeneous and exhibit marked phenotypic plasticity. A broad spectrum of additional surface markers, including CD123, CD33, TIM3, CD47, CLL-1, CD25, CD44, CD96, and GPR56, has been identified across multiple studies [41,42,43,44,45,46,186,187,188]. Notably, CD123 and CD33 are highly expressed on AML blasts and LSCs, and are associated with inferior remission rates; over 90% of AML patients express at least one of these antigens [189]. Therapeutic targeting of CD33 or CD123 using T-cell engager (TCE) antibodies and chimeric antigen receptor (CAR)–T cell approaches has demonstrated clinical efficacy [190,191,192]. T-cell immunoglobulin mucin-3 (TIM-3) has emerged as a functional marker, identifying a residual LSC population that persists following hematopoietic stem cell transplantation and is strongly predictive of relapse, underscoring its prognostic and functional relevance [193]. In addition, CD47 functions as an innate immune checkpoint that is upregulated in LSCs, enabling evasion of macrophage-mediated phagocytosis and adding further complexity to LSC biology [194,195]. Collectively, this extensive immunophenotypic diversity complicates the unequivocal identification of LSCs while simultaneously providing opportunities for refined diagnostic stratification and targeted therapeutic intervention.

7.2. Functional and Metabolic Heterogeneity
Beyond phenotypic features, LSCs are distinguished by their unique functional properties. A hallmark characteristic of LSCs is their propensity to reside in a quiescent or dormant state, which confers resistance to cytotoxic chemotherapy and targeted agents. These quiescent LSCs can subsequently re-enter the cell cycle, contributing to minimal residual disease (MRD) persistence and disease relapse [185,196]. Mechanistically, LSC dormancy is governed by a multilayered regulatory network integrating metabolic control with transcriptional and epigenetic programs. LSCs exhibit pronounced metabolic heterogeneity; quiescent LSCs preferentially rely on oxidative phosphorylation (OXPHOS) rather than glycolysis, thereby minimizing energy expenditure [197]. This metabolic state is accompanied by a distinctive mitochondrial phenotype characterized by reduced mitochondrial mass, altered morphology, and dynamic remodeling, which collectively limit reactive oxygen species (ROS) production and preserve stemness [198]. Furthermore, elevated expression of the anti-apoptotic protein BCL-2 supports mitochondrial respiration in LSCs. Pharmacological inhibition of BCL-2 with venetoclax disrupts OXPHOS and selectively eliminates quiescent LSCs, highlighting mitochondrial metabolism as a critical therapeutic vulnerability in AML [199]. LSCs also exhibit a pro-inflammatory phenotype, express fatty acid transporter CD36, and induce lipolysis in BM adipocytes to fuel fatty acid oxidation (FAO) in leukemic cells [200].
Increasing evidence also underscores the pivotal role of gene regulatory networks in enforcing LSC quiescence. INKA1, also known as C3orf54, functions as a key regulator by restraining the G0 phase and delaying early repopulation while preserving long-term self-renewal capacity. Mechanistically, INKA1 impairs the nuclear localization of PAK4, leading to reduced global H4K16 acetylation and maintenance of LSCs in a primitive, quiescent state [201]. Similarly, Forkhead box M1 (FOXM1) has been shown to regulate LSC quiescence through activation of the Wnt-β-catenin signaling pathway in MLL/AF9-driven AML models [202]. MicroRNAs represent an additional layer of post-transcriptional regulation. In particular, miR-126 is highly enriched in LSCs and suppresses AKT activity and CDK3 expression; the miR-126-CDK3 regulatory axis governs G0 exit kinetics and promotes chemotherapy resistance by sustaining LSC quiescence [203]. Moreover, RNA modification enzymes play crucial roles in LSC maintenance. METTL3 and METTL14 promote cell cycle arrest in LSCs [126,204,205], whereas METTL16 drives the m6A-dependent expression of branched-chain amino acid (BCAA) transaminases BCAT1 and NCAT2, thereby reprogramming BCAA metabolism in AML [206] (Figure 4).

7.3. Microenvironmental Heterogeneity
LSC harbors a highly heterogenous bone marrow environment (BMM), where spatially and functionally distinct niches exert differential control over LSC fate. Single-cell and spatial transcriptomic analyses in recent years have revealed that AML bone marrow is composed of multiple mesenchymal stromal, endothelial, and immune subpopulations with non-redundant roles in supporting leukemic growth and persistence.
Spatially, LSCs occupy distinct niches within BM. Primitive, quiescent LSCs are preferentially enriched in endosteal and arteriolar regions, which are characterized by low oxygen tension and enriched in niche-derived quiescence signals. In contrast, more proliferative LSC subsets are localized predominantly to perivascular and sinusoidal niches, where higher oxygen availability and cytokine concentrations support metabolic activity and clonal expansion. This spatial heterogeneity generates differential exposure to extrinsic cues, resulting in coexisting LSC populations with distinct functional states within the same patient.
At the cellular level, the BM niche supporting AML LSCs is composed of diverse stromal and non-stromal cell types. Mesenchymal stromal cells (MSCs) and endothelial cells provide key factors such as CXCL12 and stem cell factor (SCF) that promote LSC survival and self-renewal [23]. Osteoblast lineage cells have been implicated in maintaining quiescent, therapy-resistant LSC pools [25], whereas adipocytes support LSCs through enhancing fatty acid oxidation (FAO) [201,207]. In addition, immune cells within the niche, including macrophages and regulatory T cells, contribute to immune evasion and protection of LSCs from cytotoxic therapies [208,209].
Importantly, niche heterogeneity is highly dynamic. AML cells actively remodel the BM microenvironment by altering stromal cell function, vascular integrity, and cytokine gradients, thereby establishing leukemia-permissive niches. This dynamic interplay between LSCs and heterogeneous BM niches underscores the critical role of microenvironmental context in sustaining LSC plasticity and driving disease persistence.

8. Bone Marrow Microenvironment and Immune Interactions
The bone marrow microenvironment (BMM) constitutes a highly specialized and dynamic niche that orchestrates normal hematopoiesis through tightly regulated interactions among vascular, stromal, and immune components. In AML, leukemic blasts and leukemia stem cells (LSCs) actively remodel this niche to establish a heterogeneous and permissive environment that supports leukemic survival, self-renewal, and therapy resistance.
Vascular remodeling represents a hallmark of AML-induced niche alterations [210]. Leukemic cells drive aberrant angiogenesis by secreting pro-angiogenic factors such as VEGF [211], adhesion molecules including VCAM-1 [212] and E-selection [213,214], and inflammatory cytokines such as TNF-α, collectively resulting in an expanded yet dysfunctional BM vasculature [210,215,216]. This dysregulated vascular network provides protective niches that shield leukemic cells from chemotherapeutic insult via activation of AKT/mTOR and NF-kB pathways [214]. Concurrently, endothelial cells exposed to AML cells undergo phenotypic changes, further reinforcing leukemic cell adhesion and quiescence [216].
Stromal components, particularly mesenchymal stromal cells (MSCs), are profoundly reprogrammed in AML. Leukemia-educated MSCs exhibit altered transcriptional reprogramming, including upregulation of E-cadherin [217], induction of connective tissue growth factor [218], reduced capacity to support normal hematopoiesis, and increased secretion of leukemogenic cytokines and chemokines such as SCF, TNF, IL6, IL-10 and M-CSF [217,219,220,221,222]. Additionally, nestin+ BMSCs enhance OXHPOS, TCA activity and GSH-mediated antioxidant defense, thereby promoting AML survival and chemotherapy resistance [223]. Furthermore, nestin+ BMSCs facilitate AML relapse by sustaining high protein translation in chemo-resistant leukemia cells through the transfer of cap-dependent translation machinery, including eIF4A, via extracellular vesicles [224]. MSCs also contribute to AML survival and drug resistance via HDAC3-driven mitochondrial ROS accumulation, which induces stromal senescence and generates a hyperinflammatory niche that activates NF-kB signaling in leukemia cells. Collectively, AML-derived signals disrupt osteoblastic differentiation, metabolic capacity and bone homeostasis, thereby eroding HSC-supportive niches and facilitating leukemic dominance [218,225,226,227,228].
Immune remodeling is another critical dimension of AML-induced niche reprogramming. AML progression is associated with impaired immune surveillance and the accumulation of immunosuppressive cell populations, including regulatory T cells (Tregs), myeloid-derived suppressor cells (MDSCs), leukemia-associated macrophages (LAMs), regulatory B cells (Bregs) and leukemia-associated neutrophils [229,230]. Increased Treg frequency in AML promotes leukemic progression and contributes to therapeutic resistance [231]. Extracellular vesicles (EVs) play a critical role in Treg expansion. Rab27a-dependent secretion of leukemic EVs promotes engraftment and selectively induces effector Treg proliferation. Tregs internalize EVs carrying the costimulatory ligand 4-1BBL (CD137L), which enhances FOXP3 expression and drives an effector Treg phenotype by activating STAT5 signaling while concurrently suppressing mTOR-S6 signaling [232]. EVs can also deliver miRNAs such as miR-21 to promote Treg gene expression [233]. The PD-L1/PD-1 axis further mediates Treg expansion; PD-L1 engagement converts naïve T cells into Tregs by inhibiting AKT-mTOR signaling, whereas PD-L1 blockade attenuates Treg generation [234,235,236]. Additionally, elevated indoleamine 2,3-dioxygenase (IDO) expression in AML, regulated via JAK-STAT1 signaling, produces immunoregulatory metabolites such as kynurenines that further drive Treg expansion [237,238,239,240,241]. Tregs exert immunosuppressive effects through both contact-dependent mechanisms mediated by molecules such as CTLA-4 and NRP-1 [242,243], and contact-independent pathways involving cytokines such as IL-10 and IL-35 [208,244,245]. Additionally, Tregs utilize perforin and granzyme B to induce apoptosis in NK cells and CD8+ T cells [246].
Leukemia-associated macrophages (LAMs) are central regulators of the AML niche, producing a spectrum of mediators that influence leukemic progression. Macrophages can polarize into anti-leukemic M1 or pro-leukemic M2 phenotypes, with M2 macrophages being enriched in AML and characterized by surface markers including CD206, CD163, and CD115 [209,247,248,249]. Cytokines such as IL-4 and IL-13 drive M2 polarization primarily through the STAT6 pathway [152], and macrophage colony-stimulating factor (M-CSF) further skews macrophages towards M2-like states [249,250,251]. Additionally, pro-M2 factors, including Gfi1 and let-7b, are upregulated in AML and reinforce the M2 phenotype [252,253]. Metabolically, the M2 phenotype preferentially relies on fatty acid oxidation (FAO) and OXPHOS rather than glycolysis, contains increased mitochondrial content, and utilizes glutamine to fuel TCA cycles [254,255,256,257]. Functionally, M2-like macrophages secret soluble factors such as CCL2, CXCL8, and TGF-β, which enhance leukemic cell survival and suppress apoptosis [209,258,259]. Moreover, immune-checkpoint molecules expressed on tumor-associated macrophages, including PD-L1, TIM-3, and CD47, inhibit anti-tumor immunity and promote leukemia cell persistence. Therefore, LAMs emerge as key contributors to immune evasion and represent promising therapeutic targets for enhancing the efficacy of checkpoint-based immunotherapies [195,229,230,260,261] (Figure 4).

9. Emerging Concepts and Future Directions
Recent advances in AML research have been propelled by the rapid development and application of integrative multi-omics technologies, encompassing genomics, epigenomics, transcriptomics, proteomics, and metabolomics. These approaches have substantially refined our understanding of leukemogenesis and disease progression. Large-scale multi-omics studies have demonstrated that integrating multiple molecular layers enables superior risk stratification and the identification of novel molecular dependencies that remain undetectable in single-omics analysis [262,263]. For example, comprehensive inflammatory proteomic profiling of AML patient samples identified oncostatin M receptor (OSMR) as a novel prognostic biomarker [264]. Similarly, pharmacogenomics and epigenomics analysis revealed the efficacy of Menin inhibitors in KMT2A-rearranged AML [59].
Single-cell technologies have further elucidated extensive intratumoral heterogeneity, the evolutionary dynamics underlying therapeutic resistance and functional diversity within AML. These methods have uncovered rare subpopulations, including leukemia stem cells (LSCs), therapy-persistent cells, and transient cell states that are often obscured in bulk analyses [262,265]. Single-cell multi-omics profiling has delineated three clonal evolution patterns (monoclonal, linear and branched polyclonal) in complex-karyotype AML (CK-AML) [266]. Single-cell metabolic analysis, using innovative methods such as SCENITH, a flow-cytometry-based method to functionally profile energy metabolism with single-cell resolution [267], have revealed interactions between leukemia redox metabolism and EV signature, thereby providing a more comprehensive metabolic landscape in AML [268].
Single-cell spatial transcriptomics have added an additional layer of resolution by enabling the precise mapping of leukemic cells within their native BM microenvironment. Studies demonstrate that primitive-like AML cells localize near the endosteal niche in BM, whereas committed-like and granulocyte–monocyte progenitor (GMP)-like cells reside more distally. Niche-specific signaling pathways, including CXCL12-CXCR4 and PI3K/AKT/mTOR, regulate leukemic cell differentiation, migration and extramedullary infiltration [269]. Moreover, integrating single-cell spatial transcriptomics with ligand–receptor interaction analysis has revealed that immunotherapy reshapes the spatial organization of leukemic cells and the local enrichment of specific immune cell populations [270].
In addition, genome-wide CRISPR-Cas9 screening has provided causal insights into AML biology by systematically identifying genes and pathways essential for leukemic cell survival and differentiation blockade. For instance, Fermt3, a master regulator of integrin signaling identified in MLL-rearranged AML, exhibits significant prognostic relevance [271], while the RNA-binding protein ZFP36L2 has been recognized as a critical regulator of AML maintenance [272]. CRISPR/Cas9 screening has also facilitated the identification of gene markers predictive of chemo-resistance or sensitivity to ADE (cytarabine, daunorubicin, etoposide) chemotherapy in pediatric AML, offering potential biomarkers and therapeutic targets to overcome drug resistance [273].
Collectively, the integration of muti-omics has paved the way for precision immunotherapy and advanced computational modeling. Emerging concepts such as digital twins and generative AI enable patient-specific therapeutic optimization [274,275]. Deep learning frameworks, including the Transcription Regulator Activity Prediction Tool (TRAPT), provide innovative strategies for identifying transcriptional regulators (TRs) [276]. AML digital twins represent a computational paradigm that models individual disease trajectories and therapeutic responses by integrating multi-omics data with clinical and longitudinal treatment information. Unlike traditional static models, digital twins are continuously updated with new data, supporting adaptive treatment optimization and true precision medicine [277,278].
However, several limitations currently hinder their clinical translation. For the multi-omics datasets, precise delineation of cell boundaries, difference in platforms, batch effects and sample bias limit reproducibility and interpretability. From a practical perspective, many advanced multi-omics technologies remain costly and technically demanding for sample quality and processing, which constrains their feasibility for routine clinical. In addition, a substantial proportion of current discoveries are derived from retrospective analyses or relatively small patient cohorts, necessitating validation in large, multicenter, and prospective clinical studies before their utilization as diagnostic, prognostic, or predictive biomarkers can be firmly established. Emerging computational paradigms, including digital twins and generative artificial intelligence, hold great promise for patient-specific therapeutic optimization. Moreover, these approaches are highly dependent on data quality, longitudinal sampling, and model generalizability. Issues related to algorithm interpretability, regulatory approval, and clinical trust further limit their immediate adoption. At present, most AML digital twin frameworks remain at the proof-of-concept stage and require extensive clinical validation.

10. Conclusions
Acute myeloid leukemia (AML) is a blood cancer driven by genetic and epigenetic changes that disrupt normal hematopoiesis, allowing leukemic stem cells to thrive. These cells are organized hierarchically and rely on signals from the bone marrow microenvironment to maintain self-renewal, resist therapy, and drive disease progression. Contemporary AML research increasingly leverages integrative systems biology approaches, including multi-omics profiling, single-cell and spatial analyses, and computational modeling, to dissect the mechanisms by which molecular programs, cellular states, and clonal dynamics are established and maintained over time, moving beyond purely descriptive characterization. These insights facilitate the rational design of therapies that address cellular heterogeneity and microenvironmental dependencies, offering broader implications for understanding fundamental principles of cancer biology.