Beyond the DNA sequence: mapping the dynamic epigenetic landscape for risk stratification and therapeutic intervention in acute myeloid leukemia 

Abstract

Acute myeloid leukemia (AML) is a genetically heterogeneous malignancy where traditional DNA sequence-based classification fails to fully explain clinical outcomes. This review explores the dynamic and multifaceted role of the epigenome as a critical driver of leukemogenesis, progression, and therapeutic resistance. We detail how disruptions in DNA methylation, histone modifications, chromatin architecture, and non-coding RNAs create a complex regulatory landscape that transcends genetic alterations. We highlight the emergence of single-cell multi-omics technologies, which are resolving intratumoral heterogeneity and uncovering the epigenetic signatures of leukemic stem cells. Furthermore, we discuss the significant translational potential of these discoveries, including the development of epigenetic biomarkers for refined risk stratification, minimal residual disease (MRD) monitoring, and predicting response to novel epigenetic therapies like menin and LSD1 inhibitors. Integrating these dynamic epigenetic layers into clinical decision-making promises to usher in a new era of precision medicine, ultimately improving prognostic accuracy and therapeutic outcomes for AML patients.

Introduction: the epigenetic imperative in AML

Acute myeloid leukemia (AML) represents a genetically heterogeneous hematologic malignancy characterized by the clonal expansion of immature myeloid blasts with blocked differentiation and aberrant self-renewal capacity [1, 2]. Despite advances in understanding its molecular pathogenesis, AML continues to pose significant therapeutic challenges, with overall survival rates remaining disappointingly low, particularly in older patients [3, 4]. Traditional classification systems have relied heavily on cytogenetic abnormalities and recurrent genetic mutations to stratify patients into prognostic groups and guide treatment decisions [3, 5]. The European LeukemiaNet (ELN) risk stratification system, which incorporates genetic alterations such as NPM1, FLT3-ITD, and CEBPα mutations, has been instrumental in predicting outcomes and informing therapeutic choices [3, 5]. However, the limitations of this genetic-centric approach are increasingly apparent, as significant outcome heterogeneity persists within established genetic subgroups, suggesting that factors beyond DNA sequence variations contribute critically to AML pathogenesis and clinical behavior [3, 6].

The epigenome encompasses heritable changes in gene expression that occur without alterations to the underlying DNA sequence, including DNA methylation, histone modifications, chromatin remodeling, and regulation by non-coding RNAs (ncRNAs) [1, 7, 8]. In normal hematopoiesis, epigenetic mechanisms orchestrate the delicate balance between self-renewal and differentiation of hematopoietic stem cells (HSCs), ensuring proper lineage commitment and maturation [8, 9]. Disruption of this epigenetic machinery is now recognized as a fundamental driver of leukemogenesis, with mounting evidence indicating that epigenetic dysregulation can independently foster malignant transformation or collaborate with genetic lesions to enforce the leukemic phenotype [1, 5, 10]. Somatic mutations in genes encoding epigenetic regulators, including DNMT3A, TET2, IDH1/2, ASXL1, and EZH2, are among the most frequently identified alterations in AML, occurring in over 50% of cases [4, 5, 8]. These mutations disrupt normal epigenetic patterning, leading to aberrant DNA methylation landscapes, distorted histone modification signatures, and ultimately, distorted transcriptional programs that promote self-renewal while impairing differentiation [1, 4, 5, 7].

The scope of epigenetic dysregulation in AML extends beyond mutational events to encompass broader epigenetic alterations that are not necessarily linked to genetic mutations. AML cells exhibit pervasive epigenetic abnormalities, including genome-wide DNA hypomethylation juxtaposed with locus-specific CpG island hypermethylation, aberrant histone modification patterns, and disrupted chromatin architecture [1, 6, 10]. Additionally, ncRNAs, particularly long non-coding RNAs (lncRNAs) and circular RNAs (circRNAs), have emerged as significant contributors to the epigenetic landscape of AML [8, 9, 11]. These RNA species regulate gene expression through diverse mechanisms, including chromatin modification, transcriptional interference, and post-transcriptional regulation, thereby adding another layer of complexity to the epigenetic regulation of AML [9, 11, 12]. For example, lncRNAs such as HOTAIRM1 and H19 have been implicated in normal and malignant myelopoiesis, while their dysregulation contributes to leukemogenesis and therapy resistance [9, 11].

Current genetic risk stratification in AML, like the ELN system, fails to fully explain clinical heterogeneity. Epigenetic dysregulation, aberrant DNA methylation, histone modifications, and ncRNAs, is a critical driver of leukemogenesis. This review explores how mapping the dynamic epigenetic landscape can refine prognostic assessment and unveil novel therapeutic interventions, moving beyond DNA sequence to improve outcomes in AML through precision medicine.

The foundational layer: DNA methylation biomarkers

AML is characterized by profound epigenetic dysregulation, with DNA methylation alterations serving as a cornerstone for understanding disease pathogenesis, classification, and risk stratification [1]. The identification of recurrent mutations in genes encoding epigenetic modifiers, such as DNMT3A, TET2, and IDH1/2, has provided crucial insights into the mechanistic underpinnings of these methylation abnormalities [13, 14]. These mutations are not merely passive associations but actively sculpt the methylome, contributing to the leukemogenic phenotype by silencing tumor suppressor genes, disrupting differentiation pathways, and conferring self-renewal capabilities to hematopoietic progenitors [1, 15]. The prognostic impact of these mutations is profound and often context-dependent. For instance, DNMT3A mutations, occurring in approximately 20–30% of cytogenetically normal (CN)-AML cases, are frequently associated with adverse outcomes, including higher relapse rates and reduced overall survival [14, 16]. The specific type and location of the mutation further refine this risk; the most common DNMT3A R882 missense mutations exhibit a dominant-negative effect and are particularly associated with inferior survival in older adults (≥ 60 years), while non-R882 mutations (e.g., nonsense, frameshift) appear more detrimental in younger patients [14]. Similarly, the prognostic significance of TET2 mutations, present in 12–20% of AML cases, is increasingly understood to be mutation-type specific. Truncating mutations (nonsense, frameshift), designated as “significant” TET2 mutations, portend a particularly dismal prognosis, with shorter overall survival and reduced complete remission rates, and importantly, their negative impact may not be ameliorated by allogeneic hematopoietic stem cell transplantation [17]. The oncometabolite-producing IDH1/2 mutations drive a CpG island hypermethylator phenotype (CIMP) by inhibiting TET2 function, effectively creating a functional TET2-mutated state [13]. The co-occurrence of mutations in these epigenetic drivers is not random and can lead to epigenetic antagonism; for example, in AMLs with both IDH and DNMT3A mutations, the hypermethylation signature typical of IDH mutation is often suppressed, resulting in a unique epigenetic and clinical profile associated with RAS pathway activation and potential sensitivity to MEK inhibition [13].

Moving beyond single mutation status, genome-wide analyses have revealed that the collective disturbance of the DNA methylome is a defining feature of AML [13, 18, 19]. These alterations are not uniformly distributed across the genome but are highly enriched at specific regulatory regions, particularly enhancers and CpG shores, which exhibit more pronounced differential methylation than traditional promoter-associated CpG islands [13]. AML blasts can display both global hypomethylation, contributing to genomic instability, and focal hypermethylation at specific loci, leading to the silencing of genes critical for hematopoietic differentiation and tumor suppression [15, 20]. Studies leveraging high-resolution techniques like enhanced reduced representation bisulfite sequencing (ERRBS) have demonstrated that differential methylation at these non-promoter elements is a primary driver of AML’s epigenetic heterogeneity and biological subtypes [13]. The patterns of differential methylation are so robust that they can be harnessed to identify specific differentially methylated regions (DMRs) associated with key genes such as WNT10A and GATA3, which may predict response to hypomethylating agents (HMAs) like azacytidine, although developing a single CpG-based classifier has proven challenging due to the complex, region-based nature of this epigenetic information [18].

This comprehensive understanding of AML methylome disruption has catalyzed the development of DNA methylation-based classifiers that transcend traditional genetic categorization. Unsupervised clustering of genome-wide methylation data has consistently identified three primary CpG island methylator phenotypes (CIMP): CIMP-high (CIMP-H), CIMP-medium (CIMP-M), and CIMP-low (CIMP-L), which exhibit distinct clinical outcomes, mutational landscapes, and immune microenvironments [19]. For example, the CIMP-L subtype, frequently harboring DNMT3A mutations, is associated with the most favorable prognosis, whereas the CIMP-H subtype correlates with the worst overall survival. These methylation subtypes are enriched with specific genetic alterations; RUNX1 mutations are common in CIMP-M, while DNMT3A mutations are a hallmark of CIMP-L [19]. Furthermore, these epigenetically defined subgroups exhibit differences in immune cell infiltration patterns and TIDE scores, suggesting varying susceptibilities to immunotherapy and providing a holistic view of the tumor microenvironment, which is increasingly recognized as a critical determinant of outcome. The construction of risk models, such as the CIMP-associated prognostic model (CPM) comprising 32 key genes, demonstrates the potent ability of methylation signatures to predict survival with significant accuracy at 0.5, 1, 3, and 5 years, independently of other clinical variables [19, 21]. This approach underscores the value of integrating methylation-based subtyping into existing risk stratification schemas to achieve a more refined prognostic assessment.

Despite the immense promise of DNA methylation biomarkers, several formidable challenges must be overcome for their successful translation into routine clinical practice. A primary obstacle is cellular heterogeneity within patient samples, where the methylation signal from leukemic blasts can be confounded by varying proportions of non-malignant immune and stromal cells [18, 22]. This necessitates the use of purified blast populations or computational deconvolution methods to ensure accurate methylation profiling. Furthermore, the epigenome is dynamic and can evolve under therapeutic selective pressure, leading to clonal selection and changes in methylation patterns at relapse that may differ from the diagnostic profile [17]. Tracking these dynamic changes requires serial monitoring, adding a layer of complexity to biomarker application. From a technical standpoint, moving from discovery-based platforms like genome-wide bisulfite sequencing to robust, cost-effective, and clinically actionable targeted assays is non-trivial [18, 22]. While microarrays (e.g., Illumina EPIC) offer a solution, their limited coverage can miss critical regulatory DMRs identified by sequencing. The application of machine learning (ML) and artificial intelligence (AI) is poised to address this complexity by integrating large, multidimensional methylome data sets to build predictive models for diagnosis and treatment response [22], including novel deep learning frameworks specifically designed for methylation prediction [23]. Supervised ML algorithms, such as random forest and support vector machines, have shown efficacy in classifying AML based on methylation patterns [22]. However, these models require large, well-annotated patient cohorts for training and validation to avoid overfitting and to ensure generalizability, a significant hurdle in a molecularly diverse disease like AML [18, 22]. Finally, prospective validation in independent, multi-institutional cohorts is an absolute prerequisite to establishing the clinical utility and reliability of any methylation-based biomarker or classifier before it can be integrated into standard diagnostic and treatment algorithms [18, 19]. The major DNA methylation alterations and their associated clinical implications are summarized in Table 1.

Table 1.

Key DNA methylation biomarkers and their clinical implications in AML

Epigenetic alteration/Biomarker	Clinical & prognostic significance	Ref.
DNMT3A mutations	Associated with adverse outcomes in CN-AML. R882 mutations confer inferior survival in older adults (≥ 60 yrs), while non-R882 mutations are more detrimental in younger patients	[14, 16]
“Significant” TET2 mutations	Truncating (nonsense, frameshift) mutations portend a dismal prognosis with shorter OS and reduced CR rates; negative impact may not be ameliorated by allogeneic HSCT	[17]
IDH1/2 mutations	Drive a CIMP phenotype by inhibiting TET2 function. Co-occurrence with DNMT3A mutations can lead to epigenetic antagonism and a unique profile with potential sensitivity to MEK inhibition	[13]
CIMP subtypes (CIMP-H, M, L)	CIMP-L (often with DNMT3A mutations) has the most favorable prognosis. CIMP-H correlates with the worst OS. These subtypes have distinct mutational landscapes and immune microenvironments	[19]
CPM	A 32-gene methylation signature that predicts survival at 0.5, 1, 3, and 5 years with significant accuracy, independent of other clinical variables	[19]

The expanding horizon: novel and emerging epigenetic layers

The understanding of AML pathogenesis has evolved considerably beyond genetic mutations to encompass complex epigenetic dysregulation. While DNA methylation and histone modifications have been extensively studied, recent investigations have revealed increasingly sophisticated layers of epigenetic control that contribute to leukemogenesis. These emerging dimensions include RNA modifications, three-dimensional chromatin architecture, and histone variants, which collectively form a dynamic regulatory network that governs gene expression patterns in myeloid malignancies. The integrative analysis of these epigenetic layers provides unprecedented opportunities for risk stratification and therapeutic intervention in AML, particularly for patients with high-risk disease or those who exhibit resistance to conventional chemotherapy. This section examines three rapidly advancing areas of epigenetic research in AML, highlighting their potential clinical applications and mechanistic insights into disease pathogenesis.

The epitranscriptome: RNA modifications as biomarkers

The epitranscriptome represents a recently discovered layer of epigenetic regulation involving post-transcriptional modifications of RNA molecules that significantly influence their function, stability, and translational efficiency. In AML, several RNA modifications have been identified as critical regulators of leukemogenesis, with N6-methyladenosine (m6A) being the most extensively studied. The m6A modification is dynamically regulated by writer complexes (including METTL3, METTL14, and WTAP), erasers (FTO and ALKBH5), and reader proteins (YTHDF family, IGF2BPs) that recognize the modification and mediate its functional outcomes. Recent evidence indicates that METTL3 is frequently overexpressed in AML and promotes leukemogenesis by enhancing the translation of key oncogenic drivers such as MYB, MYC, and BCL2 through m6A-dependent mechanisms [24]. The increased catalytic activity of METTL3 has been correlated with adverse clinical outcomes and resistance to chemotherapy, suggesting its potential utility as a predictive biomarker for treatment response [25].

Beyond m6A, other RNA modifications including 5-methylcytosine (m5C) and pseudouridine (Ψ) are gaining recognition as important contributors to AML pathogenesis. The m5C modification, catalyzed by writers such as NSUN2 and DNMT2, has been implicated in the stabilization of internal ribosomal entry site (IRES)-containing mRNAs that encode proteins involved in self-renewal and differentiation blockade. Interestingly, recent studies have demonstrated that aberrant m5C patterns can distinguish AML subtypes with different prognosis and treatment responses [26]. Similarly, pseudouridylation of ncRNAs, particularly snoRNAs and scaRNAs, has been associated with ribosomal dysfunction and impaired hematopoietic differentiation in AML. The discovery that specific pseudouridine patterns in circulating exosomes can serve as non-invasive biomarkers for AML detection represents a significant advancement in diagnostic approaches [24, 27, 28].

The therapeutic potential of targeting RNA-modifying enzymes is increasingly recognized in AML. Preclinical studies have demonstrated that inhibition of FTO, which is overexpressed in certain AML subtypes with t(11q23)/MLL rearrangements and FLT3-ITD mutations, sensitizes leukemic cells to conventional chemotherapy and tyrosine kinase inhibitors [25]. Similarly, small molecule inhibitors targeting METTL3 have shown promising antileukemic effects in patient-derived xenograft models, particularly in cases with secondary AML evolved from myelodysplastic syndrome. The development of selective inhibitors against specific readers of m6A modifications, such as YTHDF2, offers additional therapeutic avenues for disrupting the epitranscriptomic programs that sustain leukemic stem cells (LSCs) [24]. As research in this field advances, the comprehensive mapping of RNA modification landscapes in AML patients will likely facilitate the development of epitranscriptome-based classifiers for precision medicine approaches.

Chromatin architecture and 3D genomics

The three-dimensional organization of chromatin within the nucleus represents a fundamental regulatory layer for gene expression control, with profound implications for leukemogenesis. Central to this architecture are topologically associating domains (TADs), self-interacting genomic regions where DNA sequences within each domain physically interact more frequently than with sequences outside the domain [29–31]. In mammalian cells, TADs range in size from approximately 100 kb to 2 Mb and are demarcated by boundary elements enriched for the architectural proteins CTCF and cohesin, housekeeping genes, transfer RNA genes, and short interspersed nuclear elements (SINEs) [29, 32]. Recent high-resolution Hi-C studies in AML have revealed that malignant cells exhibit widespread alterations in TAD boundaries compared to normal hematopoietic stem and progenitor cells (HSPCs), with characteristic patterns associated with specific molecular subtypes [33]. These structural rearrangements frequently result in ectopic enhancer-promoter interactions that activate oncogenes or silence tumor suppressors, thereby contributing to the leukemogenic process.

The integrity of TAD boundaries is crucial for maintaining appropriate gene regulation, and their disruption in AML can lead to pathogenic rewiring of transcriptional programs. Comprehensive analyses of primary AML samples have identified recurrent boundary shifts, expansions, and contractions that affect hundreds of cancer-related genes [33, 34]. For instance, the disruption of TAD boundaries near the WT1 gene through structural variations or epigenetic alterations leads to its ectopic activation via engagement with a super-enhancer located in an adjacent domain, particularly in samples with TET2 and/or FLT3-ITD mutations [33]. Similarly, subtype-specific alterations in TAD organization have been documented near key leukemogenic drivers such as MYCN, MEIS1, and ERG, where newly formed loops connect these promoters with distal enhancers that are normally inactive in myeloid cells [33]. These architectural changes are not merely consequences of transformation but actively contribute to disease pathogenesis by creating oncogenic gene regulatory circuits.

Beyond TADs, higher-order chromatin structures including lamina-associated domains (LADs) and chromatin loops have emerged as significant players in AML pathogenesis. LADs represent genomic regions that interact with the nuclear lamina and are typically characterized by repressive chromatin marks and transcriptional silencing [30, 31, 34]. In AML, the redistribution of LADs has been associated with the reactivation of developmental genes that promote self-renewal and impair differentiation. Additionally, advanced chromatin conformation capture techniques have identified AML-specific promoter-enhancer and promoter-silencer loops that occur predominantly within TADs and involve approximately 42.4% and 11.2% of all loops, respectively [33]. These loops are frequently anchored by CTCF binding sites and enriched for specific histone modifications, with active loops marked by H3K27ac and repressive loops associated with H3K27me3 and EZH2 binding [33, 34].

The clinical implications of 3D genomic studies in AML are rapidly expanding. Hi-C and ATAC-seq profiling of primary samples has demonstrated that chromatin architecture patterns can stratify patients into distinct prognostic groups beyond conventional genetic markers [26, 33]. Specifically, the degree of compartment switching between active and inactive chromatin states has been correlated with treatment response and survival outcomes. Moreover, the identification of structural variation-induced enhancer-hijacking and silencer-hijacking events in AML genomes provides mechanistic explanations for the dysregulation of genes without coding mutations [33]. These findings open new therapeutic opportunities targeting the architectural machinery, including CTCF and cohesin complexes, or using HMAs to restore normal chromatin organization [33, 34]. As single-cell technologies for assessing chromatin architecture mature, they promise to reveal the heterogeneity of 3D genome organization within leukemic populations and its evolution during disease progression and treatment.

Histone variants and oncohistones

While mutations in epigenetic modifiers have been extensively documented in AML, recent investigations have revealed that mutations in the histones themselves, termed “oncohistones”, can drive leukemogenesis through profound alterations of the epigenetic landscape. Although historically associated with pediatric gliomas and sarcomas, oncohistone mutations are increasingly recognized in hematological malignancies, including AML [24, 35]. Systematic sequencing of histone genes in primary AML samples has identified recurrent mutations in several H3 variants at critical residues, including Q69H, A26P, R2Q, R8H, and the well-characterized K27M substitution [35]. These mutations occur with an overall frequency of approximately 1.6% in AML, with notably higher incidence in secondary AML (9%) arising from myelodysplastic syndrome or other precursor conditions. The presence of oncohistones in pre-leukemic HSCs and their persistence at high variant allele frequencies in major leukemic clones indicates that they represent early events in leukemogenesis that confer selective advantage to hematopoietic precursors [35].

The mechanistic consequences of oncohistone mutations in AML involve complex alterations to histone modification landscapes and chromatin structure. The H3K27M mutation, the best-characterized oncohistone, exerts a dominant-negative effect by inhibiting the catalytic activity of polycomb repressive complex 2 (PRC2), leading to global reduction of H3K27me3 levels and local gains of H3K27ac at specific genomic loci [24, 35]. This redistribution of histone modifications results in aberrant activation of gene expression programs that promote self-renewal and block differentiation. Interestingly, H3K27M mutations in AML are associated with distinctive gene expression signatures involving upregulation of genes involved in erythrocyte and myeloid differentiation pathways, suggesting lineage-specific effects 3. Beyond H3K27M, other histone mutations such as H3G34R/V disrupt the recognition of H3K36 methylation by reader proteins, interfering with transcriptional elongation and DNA damage response mechanisms [24]. These findings highlight the diverse mechanisms through which oncohistones subvert normal epigenetic regulation in hematopoietic cells.

The functional impact of oncohistones in AML has been validated through experimental models demonstrating their ability to enhance self-renewal capacity and impair differentiation of hematopoietic precursors. Introduction of H3K27M into human HSCs expands functional stem cell populations and alters their differentiation potential, ultimately promoting leukemic aggressiveness in transplantation assays [35]. These effects appear to be dependent on the specific histone mutation, with different variants conferring distinct phenotypic outcomes. Importantly, oncohistone-driven leukemogenesis can occur independently of cooperating mutations in established drivers like RUNX1, although they frequently co-occur with lesions in epigenetic modifiers such as ASXL1, TET2, and components of the spliceosome machinery [35]. This suggests that oncohistones can function as founding lesions that establish a permissive epigenetic context for additional mutational events during leukemic evolution.

From a translational perspective, the identification of oncohistones in AML has important implications for risk stratification and therapeutic development. Patients harboring H3K27M mutations exhibit distinctive clinical features, including enrichment in specific AML subtypes and associations with particular cooperating mutations [35]. The global reduction in H3K27me3 caused by H3K27M creates a unique epigenetic vulnerability that can be targeted with inhibitors of demethylases that remove H3K27me3, such as JMJD3/GSKJ4, or with agents that target compensatory chromatin modifications [24]. Additionally, the altered DNA damage response in cells with H3G34R/V mutations may confer sensitivity to PARP inhibitors or other agents that induce genomic instability [24]. As research in this area advances, comprehensive profiling of histone mutations in larger AML cohorts will be essential to fully elucidate their clinical significance and therapeutic implications. The development of targeted therapies for oncohistone-driven AML represents a promising frontier in precision epigenetics for hematological malignancies. The mechanisms and clinical potential of these novel epigenetic regulators are consolidated in Table 2.

Table 2.

Emerging epigenetic layers: mechanisms and translational potential in AML

Epigenetic layer	Key mechanisms & players in AML	Clinical & therapeutic potential	Ref.
m6A RNA modification	METTL3 (writer) overexpression enhances translation of oncogenes (MYB, MYC, BCL2). FTO (eraser) overexpression in MLL-rearranged/FLT3-ITD AML	METTL3/FTO inhibitors show promising antileukemic effects. METTL3 activity is a potential predictive biomarker for chemotherapy resistance	[24, 25]
3D chromatin architecture	Widespread TAD boundary alterations lead to ectopic enhancer-promoter interactions (e.g., near WT1, MYCN). Compartment switching correlates with prognosis	Chromatin patterns stratify patients beyond genetic markers. Targeting architectural machinery (CTCF/cohesin) or using HMAs to restore organization are potential strategies	[26, 30, 33]
Oncohistones (e.g., H3K27M)	Inhibit PRC2 activity, causing global loss of H3K27me3 and local gains of H3K27ac. Drive pre-leukemic HSC expansion and are early leukemogenic events	Create a unique vulnerability targetable by JMJD3/GSKJ4 inhibitors. H3G34R/V mutations may confer sensitivity to PARP inhibitors	[24, 35]

The functional regulators: NcRNAs in the driver’s seat

The epigenetic landscape of AML is profoundly shaped by the intricate networks of ncRNAs, which have emerged as pivotal regulators of leukemogenesis, disease progression, and therapeutic response. Moving beyond mere correlation, mechanistic studies have elucidated how specific microRNAs (miRNAs) and lncRNAs function as bona fide oncogenes or tumor suppressors by modulating key signaling pathways. For instance, miR-155 is frequently overexpressed in AML and drives myeloid proliferation by directly repressing SHIP1 and CEBPB, thereby hyperactivating PI3K/AKT signaling and impairing differentiation [36]. Conversely, the let-7 family acts as a tumor suppressor by targeting oncogenes like RAS and MYC, and its repression by LIN28B, which is often upregulated in AML, promotes self-renewal and blocks myeloid differentiation. Among lncRNAs, HOTAIRM1, transcribed from the HOXA cluster, plays a critical role in modulating myeloid differentiation by fine-tuning the expression of HOXA genes through interactions with chromatin modifiers. Similarly, the newly identified LINC01629 (also known as LNC-ECAL1) is upregulated in AML stem cells and promotes leukemogenesis by forming an RNA-protein complex with DNMT1 and EZH2, leading to epigenetic silencing of tumor suppressor genes like p15INK4B [37, 38]. These ncRNAs do not operate in isolation but form complex competing endogenous RNA (ceRNA) networks, where lncRNAs such as LINC01629 can sequester miRNAs, thereby derepressing their target mRNAs and adding another layer of post-transcriptional regulation [37, 39].

The remarkable stability of specific ncRNAs in circulating biofluids has unveiled unprecedented opportunities for non-invasive liquid biopsies in AML management. Circulating miRNAs and lncRNAs, protected from degradation by encapsulation in exosomes or binding to argonaute proteins, offer a dynamic snapshot of the disease state. Studies have demonstrated that plasma levels of miR-92a and miR-99a significantly correlate with bone marrow blast percentage, allowing for non-invasive disease monitoring and early detection of relapse [36]. Furthermore, exosomal lncRNAs, such as HOTAIR and PVT1, are enriched in the serum of AML patients and show differential expression between genetic subtypes, providing prognostic information independent of conventional risk factors [37, 40]. The implementation of droplet digital PCR (ddPCR) and next-generation sequencing (NGS) of plasma-derived ncRNAs has achieved sensitivities and specificities exceeding 90% for distinguishing AML from other hematologic malignancies and healthy controls, underscoring their clinical utility as minimally invasive biomarkers for diagnosis and residual disease monitoring [38, 41].

Beyond the well-characterized miRNAs and lncRNAs, cutting-edge research has identified tRNA-derived fragments (tRFs) and circRNAs as entirely new classes of epigenetic regulators with exceptional stability and disease-specific expression patterns. tRFs, such as tRF-3001a and tRF-3003a, are generated by specific cleavage of tRNAs by angiogenin or Dicer and can load into Argonaute complexes to silence target genes in a miRNA-like manner, influencing apoptosis and differentiation in AML blasts [42, 43]. Notably, a distinct tRF signature derived from the 5′ ends of specific tRNAs (tRF-5s) can predict transformation from myelodysplastic syndromes (MDS) to AML with high accuracy, highlighting their potential as predictive biomarkers [36]. Conversely, circRNAs, characterized by their covalently closed loop structure, exhibit extraordinary resistance to exonucleases and accumulate to high levels in AML cells. CircRNAs like circ-VIM (circ_009910) and circ_0004277 are dysregulated in AML and function as efficient miRNA sponges; for example, circ-VIM sequesters miR-20a-5p to derepress its target genes, promoting cell cycle progression and inhibiting apoptosis [38, 41]. Recent genome-wide screens have identified a CIMP in AML that is tightly associated with specific circRNA expression profiles, suggesting an interplay between DNA methylation and circRNA biogenesis [39, 44]. The functional significance of these ncRNAs is further underscored by their ability to encode peptides; for instance, circ-ANAPC7 was recently found to produce a novel peptide that regulates the ubiquitin-proteasome system in AML, adding a previously unrecognized dimension to their mechanistic roles [37, 39].

The translational potential of these ncRNAs is immense, yet several challenges must be addressed before clinical implementation. Standardized protocols for sample collection, RNA extraction, and data normalization are critical for the reproducibility of liquid biopsy assays, as hemolysis can significantly alter the plasma miRNA profile. Furthermore, the dynamic changes in ncRNA expression during therapy and the cellular heterogeneity of the bone marrow microenvironment necessitate serial sampling and single-cell approaches to deconvolute the specific contributions of leukemic versus non-malignant cells [36, 44]. Computational tools, such as the recently developed tRFTars, which uses a support vector machine model to predict tRF-mRNA interactions based on features like minimum free energy and seed pairing, are essential for deciphering the complex regulatory networks governed by these ncRNAs [43], as demonstrated by other advanced network-based inference methods for non-coding RNAs [45]. Prospective validation in large, multi-institutional cohorts is required to establish the clinical validity and utility of ncRNA-based biomarkers for risk stratification and treatment selection in AML. As these hurdles are overcome, the integration of ncRNA profiles with genetic and epigenetic data will undoubtedly pave the way for more precise and personalized therapeutic interventions in AML, ultimately improving patient outcomes [37, 38, 41]. The diverse functions and burgeoning clinical applications of ncRNAs in AML are highlighted in Table 3. These diverse functions of ncRNAs and their translational potential are visually summarized in Fig. 1, which illustrates representative examples of miRNAs, lncRNAs, circRNAs, and tRFs involved in AML pathogenesis and their clinical applications.

Table 3.

NcRNAs as functional regulators and biomarkers in AML

ncRNA category	Example & function	Clinical application & utility	Ref.
Oncogenic miRNA	miR-155: Drives proliferation by repressing SHIP1/CEBPB, hyperactivating PI3K/AKT	Plasma levels of miR-92a/99a correlate with BM blast %, enabling non-invasive monitoring	[36]
Tumor suppressor miRNA	let-7 family: Targets RAS and MYC; repressed by LIN28B to promote self-renewal	A distinct tRF-5 signature can predict MDS-to-AML transformation with high accuracy	[36]
Functional lncRNA	LINC01629: Forms a complex with DNMT1/EZH2 to silence tumor suppressors (e.g., p15INK4B)	Exosomal lncRNAs (HOTAIR, PVT1) provide prognostic info independent of conventional risk factors	[37, 38, 40]
circRNA	circ-VIM: Sponges miR-20a-5p to promote cell cycle progression and inhibit apoptosis	circRNA expression is associated with CIMP subtypes. circ-ANAPC7 produces a novel peptide regulating UPS	[38, 39, 41, 44]
tRF	tRF-3001a/3003a: Silences target genes in a miRNA-like manner, influencing apoptosis/differentiation	ddPCR of plasma ncRNAs achieves > 90% sensitivity for AML diagnosis and MRD monitoring	[36, 42, 43]

Fig. 1.

Representative ncRNAs involved in AML, outlining their functional roles in proliferation, apoptosis, and differentiation, as well as their potential utility as diagnostic, prognostic, and disease-monitoring biomarkers

Single-cell multi-omics: resolving the epigenetic heterogeneity

The advent of single-cell multi-omics technologies has revolutionized our understanding of AML epigenetics by enabling the simultaneous interrogation of genomic, transcriptomic, and epigenomic features within individual cells. These approaches have revealed an unprecedented degree of heterogeneity within leukemic populations, uncovering regulatory circuits driving therapy resistance and relapse. Single-cell assays now provide high-resolution maps of the epigenetic landscape, capturing dynamic changes during disease progression and treatment response that were previously obscured in bulk analyses [46, 47]. This technological revolution has particularly transformed our ability to study LSCs, the rare populations responsible for relapse, by identifying their unique epigenetic signatures and vulnerabilities [48, 49].

The technological revolution in single-cell epigenomics includes methods such as scATAC-seq (single-cell Assay for Transposase-Accessible Chromatin with sequencing), which maps genome-wide chromatin accessibility landscapes at single-cell resolution; scChIC-seq (single-cell chromatin immunocleavage sequencing), which detects specific histone modifications in individual cells; and CITE-seq (Cellular Indexing of Transcriptomes and Epitopes by sequencing), which simultaneously quantifies surface protein expression and transcriptomic profiles [46, 47, 50]. These techniques have revealed remarkable heterogeneity in chromatin states and transcriptional programs within seemingly homogeneous AML populations, providing insights into the functional diversity of leukemic cells [47]. For instance, integrated scRNA-seq and scATAC-seq profiling in t(8;21) AML identified TCF12 as a core component of the AML1-ETO-containing transcription factor complex (AETFC), driving a universally repressed chromatin state in blast cells [46]. Similarly, multi-omic analysis of therapy-resistant AML samples has uncovered heterogeneous lineage composition with cells primed for stem, progenitor, and even differentiated myeloid, erythroid, and lymphoid lineages, challenging conventional views of AML differentiation arrest [47].

Applying these technologies to LSCs has revealed their epigenetic plasticity and regulatory complexity. scATAC-seq profiling of 22 bone marrow aspirates from therapy-resistant AML patients demonstrated that LSCs exhibit distinct chromatin accessibility patterns characterized by enrichment of stemness-related transcription factor motifs and repression of differentiation pathways [47]. These LSC-specific epigenetic signatures persist following treatment and are associated with poor clinical outcomes, suggesting their role in therapeutic resistance [47, 51]. Importantly, single-cell multi-omics has enabled the identification of novel LSC-specific regulatory circuits, such as the SPI1-CEBPE feedforward loop in APL, which drives leukemic cell differentiation and could be therapeutically targeted 6. Furthermore, integrated analysis of chromatin accessibility and gene expression has revealed that LSCs in different AML subtypes utilize distinct enhancer networks to maintain stemness properties, providing new opportunities for subtype-specific interventions [49, 50].

Single-cell multi-omics has fundamentally transformed our ability to track clonal evolution and understand therapy resistance mechanisms in AML. By combining mitochondrial DNA sequencing with scATAC-seq (mtscATAC), researchers have demonstrated that distinct mitochondrially-defined clones can undergo convergent epigenetic evolution at relapse, acquiring similar chromatin accessibility signatures despite their genetic diversity. This epigenetic convergence occurs independently of genetic mutations in approximately 40% of relapsed AML cases, highlighting the importance of non-genetic mechanisms in treatment failure [48]. Moreover, longitudinal single-cell profiling of paired diagnosis-relapse samples has revealed that therapy-resistant subclones often exhibit pre-existing epigenetic features that become selected under therapeutic pressure, rather than emerging de novo after treatment [48, 51]. These observations suggest that epigenetic profiling could identify high-risk subclones undetectable by conventional DNA sequencing alone, enabling earlier intervention strategies [47, 48].

The integration of single-cell multi-omics data with ML approaches has further enhanced our ability to predict clinical outcomes and identify novel therapeutic vulnerabilities. For instance, ML-based integration of scRNA-seq and scATAC-seq data from t(8;21) AML patients generated a robust 9-gene prognostic signature that effectively predicted survival outcomes across multiple independent cohorts [46]. Similarly, computational analysis of single-cell chromatin accessibility data has identified enhancer elements that are specifically activated in LSCs and associated with drug resistance, providing potential targets for epigenetic therapies [47, 50]. These approaches have also revealed how ncRNAs, particularly lncRNA and circRNA, contribute to epigenetic regulation in AML by modulating chromatin remodeling complexes and transcription factor activity [49, 52].

Despite these advances, several challenges remain in the implementation of single-cell multi-omics technologies for clinical applications. Technical considerations include the need for improved methods for simultaneous profiling of multiple epigenetic modalities, enhanced computational tools for integrating multi-omic datasets, and standardized protocols for processing limited clinical specimens [47, 50]. Furthermore, the translation of these discoveries into clinical practice requires validation in larger prospective cohorts and the development of targeted therapies that specifically exploit identified epigenetic vulnerabilities [46, 51]. Nevertheless, single-cell multi-omics approaches already provide unprecedented insights into the epigenetic heterogeneity of AML, offering new avenues for risk stratification and therapeutic intervention [46–48]. As these technologies continue to evolve, they will undoubtedly yield further discoveries into the dynamic epigenetic landscape of AML, ultimately improving outcomes for this challenging disease.

Translating epigenetic biomarkers to the clinic

The integration of epigenetic biomarkers into clinical decision-making represents a paradigm shift in the management of AML, moving beyond traditional genetic-based risk stratification to incorporate dynamic layers of molecular information. While genetic alterations remain fundamental to AML classification, epigenetic markers offer complementary insights into disease biology, prognosis, and therapeutic vulnerabilities. The clinical translation of these biomarkers leverages technological advances in genome-wide methylation sequencing, chromatin accessibility mapping, and bioinformatic analytics to generate actionable insights for patient management. Several epigenetic biomarkers have demonstrated superior sensitivity to conventional methods for minimal residual disease (MRD) detection, prediction of treatment response, and identification of novel therapeutic targets [53, 54]. This section examines the most promising clinical applications of epigenetic biomarkers in AML, focusing on integrated risk stratification models, epigenetic MRD monitoring, and predictive biomarkers for emerging epigenetic therapies.

Refining risk stratification: integrating epigenetic classifiers with genetic data

The development of integrated epigenetic-genetic classifier systems represents a significant advancement in AML risk stratification, addressing limitations of conventional genetic-only classification systems. Current ELN guidelines rely primarily on cytogenetic and molecular genetic abnormalities for risk categorization; however, considerable outcome heterogeneity exists within each genetic risk group [55]. DNA methylation profiling has identified distinct epigenetic subtypes that transcend conventional genetic classifications, providing additional prognostic precision. For instance, a landmark study demonstrated that a seven-gene methylation score (CD34, RHOC, SCRN1, F2RL1, FAM92A1, MIR155HG, and VWA8) significantly predicted overall survival independent of genetic mutations, with patients exhibiting low scores showing significantly higher complete remission rates (94% vs. 87%) and improved survival outcomes [55]. This epigenetic classifier maintained prognostic significance in multivariable models incorporating established genetic risk factors, supporting the development of a unified “Epi-Genetic” risk model that could enhance the current ELN framework.

The integration of epigenetic markers with genetic profiling enables more precise identification of high-risk disease features that would otherwise be misclassified. For example, within the CN-AML subgroup, which represents the largest cytogenetic category, DNA methylation patterns can discriminate patients with favorable versus adverse outcomes despite similar genetic profiles [55]. Specifically, hypermethylation of promoter regions in genes involved in hematopoietic differentiation and tumor suppression is associated with inferior outcomes, even in genetically intermediate-risk patients. These epigenetic alterations frequently occur in conjunction with mutations in epigenetic modifiers such as DNMT3A, TET2, and IDH1/2, creating distinct epigenetic signatures that reflect the functional convergence of genetic and epigenetic abnormalities [53, 55]. The clinical implementation of such integrated classification requires standardized methylation profiling platforms and bioinformatic pipelines that can be deployed in diagnostic laboratories, with ongoing efforts focused on developing targeted methylation panels that capture the most informative CpG sites for clinical prognostication.

Recent technological advances have facilitated the development of practical epigenetic profiling approaches suitable for clinical implementation. Methods such as MethylCap-seq and reduced representation bisulfite sequencing provide comprehensive genome-wide methylation data, while targeted approaches using pyrosequencing or digital droplet PCR offer focused analysis of specific regulatory regions with turnaround times compatible with clinical decision-making [53]. These techniques have been validated across multiple independent cohorts, demonstrating consistent prognostic value across age groups and treatment intensities. The future of AML risk stratification likely involves the integration of multidimensional data including genetic mutations, methylation signatures, chromatin accessibility profiles, and transcriptomic data into unified prediction models that better capture the biological complexity of the disease and provide more personalized prognostic assessment [53–55].

Epigenetic MRD (eMRD): detecting residual disease through aberrant methylation panels

MRD monitoring has emerged as a critical tool for post-treatment risk assessment in AML, with eMRD detection offering several advantages over conventional flow cytometry or NGS-based approaches. eMRD leverages the stability of cancer-specific methylation patterns, which are less susceptible to clonal evolution and antigenic shifts that can compromise immunophenotypic MRD detection [53]. Methylation-based MRD assays typically focus on panels of hypermethylated promoter regions that are highly specific to leukemic cells, with demonstrated sensitivity reaching 0.001% (10 − 5), surpassing the sensitivity threshold of many current MRD methods [53]. This exceptional sensitivity is particularly valuable for detecting early relapse and guiding preemptive therapeutic interventions, especially in patients who achieve morphologic remission but maintain persistent measurable residual disease.

The development of effective eMRD assays requires careful selection of methylation markers that exhibit consistent leukemia-specific patterns with minimal background methylation in normal hematopoietic cells. For instance, hypermethylation of promoters controlling tumor suppressor genes and developmental regulators is frequently employed due to their high prevalence in AML and stability during disease course [53]. These markers can be combined into multiplex panels that capture the heterogeneity of methylation patterns across different AML subtypes, increasing the applicability across diverse patient populations. Technical approaches for eMRD detection have evolved from genome-wide discovery methods to targeted assays suitable for routine clinical use, including methylation-specific PCR, pyrosequencing, and digital droplet PCR platforms that offer quantitative assessment of methylation levels with rapid turnaround times [53]. These targeted approaches facilitate serial monitoring throughout the treatment course, providing dynamic information about treatment response and emerging resistance.

Clinical validation studies have demonstrated strong correlation between eMRD status and clinical outcomes, with persistent methylation positivity after induction therapy predicting significantly higher relapse risk and inferior survival [53]. Importantly, eMRD detection may identify patients at risk of relapse who would be classified as MRD-negative by conventional methods, highlighting its complementary value to existing approaches. The implementation of eMRD in clinical practice requires standardization of analytical methods, definition of clinically relevant thresholds, and establishment of quality control measures to ensure reproducibility across laboratories [53]9. Future directions include the development of integrated MRD assessment combining genetic, immunophenotypic, and epigenetic markers to maximize detection sensitivity and clinical utility, as well as the exploration of liquid biopsy approaches that detect methylation markers in cell-free DNA as a non-invasive alternative to bone marrow aspiration [53].

Predicting and monitoring response to epigenetic therapy

The emergence of epigenetic therapies including HMAs and novel targeted drugs has created an urgent need for predictive biomarkers to guide patient selection and treatment monitoring. For conventional HMAs such as azacitidine and decitabine, response prediction remains challenging, with current clinical factors (e.g., blast count, cytogenetics) providing limited predictive value [56, 57]. DNA methylation profiling has identified potential predictive signatures, with studies suggesting that baseline global methylation levels and specific methylation patterns may correlate with HMA response [53, 55]. For instance, patients with higher degrees of promoter hypermethylation in specific gene sets (e.g., tumor suppressor genes, developmental regulators) may exhibit better responses to demethylating therapy, although validation in large prospective cohorts is ongoing. Additionally, dynamic changes in methylation patterns during treatment may serve as pharmacodynamic markers of target engagement, with early demethylation of specific loci potentially predicting subsequent clinical response. Beyond traditional HMAs, research into combination regimens includes agents like dihydroartemisinin, which has been shown to sensitize AML cells to cytarabine by modulating the Nrf2/HO-1 signaling pathway [58].

For novel epigenetic therapies targeting specific molecular pathways, biomarker-driven patient selection has proven essential for clinical efficacy. Menin inhibitors demonstrate striking activity in AML with KMT2A rearrangements or NPM1 mutations, which constitute approximately 5–10% and 30% of AML cases, respectively [59–61]. These genetic alterations create a dependency on the menin-KMT2A interaction for maintenance of leukemogenic gene expression programs, particularly involving HOXA cluster genes and MEIS1 [59, 60, 62]. Response to menin inhibitors correlates with reduction in HOXA/MEIS1 expression, making these transcriptional changes potential pharmacodynamic biomarkers for target inhibition [59, 61, 63]. Recent clinical trials of revumenib have demonstrated complete response rates of 26% and overall response rates of 48% in heavily pretreated R/R AML patients with NPM1 mutations, with 63% of responders achieving MRD negativity [61]. Similarly, biomarkers for DOT1L inhibitors in KMT2A-rearranged AML include reduction in H3K79 methylation and decreased expression of leukemogenic genes such as HOXA9 and MEIS1 [57, 60, 63, 64].

LSD1 inhibitors represent another class of epigenetic therapeutics with biomarker-driven application opportunities. High LSD1 expression is associated with poor prognosis in AML and maintains an immature, differentiation-blocked state in leukemic blasts. Predictive biomarkers for LSD1 inhibition include expression levels of LSD1 itself and specific genetic alterations such as MLL rearrangements [65, 66]. Response to LSD1 inhibitors is characterized by differentiation of leukemic cells, which can be monitored through immunophenotypic changes (e.g., increased CD11b and CD14 expression) and morphological evidence of differentiation. Combination approaches with all-trans retinoic acid (ATRA) have demonstrated synergistic effects in preclinical models, with ongoing clinical trials exploring this combination [65, 66]. For BET inhibitors, although predictive biomarkers are less well defined, high BRD4 expression and specific genetic contexts such as NPM1 mutations or MLL rearrangements may enrich for responsive populations [57]. The successful clinical translation of these targeted epigenetic therapies will require companion diagnostic assays that reliably identify dependent malignancies and monitor therapeutic response through appropriate pharmacodynamic biomarkers [57, 59, 61]. Looking forward, the combination of epigenetic therapies with other modalities, such as monoclonal antibody-based immunotherapies, represents a promising frontier for enhancing treatment efficacy and overcoming resistance [67]. A summary of the most promising clinical applications of epigenetic biomarkers is provided in Table 4. These applications are further illustrated in Fig. 2, which provides an overview of representative epigenetic biomarkers and their utility in risk stratification, MRD detection, and prediction of therapeutic response in AML.

Table 4.

Clinical translation of epigenetic biomarkers: applications and examples

Clinical application	Epigenetic tool/Biomarker	Utility & performance	Ref.
Integrated risk stratification	7-gene methylation score (CD34, RHOC, SCRN1, F2RL1, FAM92A1, MIR155HG, VWA8)	Predicts OS independent of genetic mutations. Low-score patients had higher CR rates (94% vs. 87%)	[55]
eMRD	Panels of hypermethylated promoters (e.g., tumor suppressor genes)	Sensitivity reaches 0.001% (10⁻⁵), surpassing many conventional methods. Detects MRD in flow/NGS-negative patients	[53]
Predicting response to epigenetic therapy	Reduction in HOXA/MEIS1 expression	Pharmacodynamic biomarker for response to menin inhibitors in NPM1mut/KMT2Ar AML	[59, 61, 63]
	Immunophenotypic differentiation (↑CD11b/CD14)	Marker of response to LSD1 inhibitors	[65, 66]
	Global methylation levels/specific patterns	Potential predictive signatures for response to HMAs (azacitidine/decitabine)	[53, 55]

Fig. 2.

Clinical translation of epigenetic biomarkers in AML, highlighting representative tools and their applications in risk stratification, MRD detection, and prediction of response to epigenetic therapies

Challenges and future perspectives

The integration of epigenetic profiling into clinical decision-making for AML faces several significant challenges that must be addressed before realizing its full potential in precision medicine. While the dynamic nature of the epigenome offers unprecedented opportunities for risk stratification and therapeutic intervention, this complexity also presents substantial obstacles in deciphering clinically actionable patterns from the vast multidimensional data generated by current technologies [1, 68]. This section examines the key challenges and future directions in mapping the epigenetic landscape of AML, highlighting critical pathways toward clinical translation.

Technical and analytical challenges

Current epigenetic profiling technologies face significant limitations in sensitivity, reproducibility, and scalability that hinder their widespread clinical implementation. Single-cell multi-omics approaches, while powerful, remain technically demanding and cost-prohibitive for routine diagnostic use [46, 69]. The analysis and integration of multidimensional epigenetic data require sophisticated computational infrastructure and advanced bioinformatics expertise that are not universally available in clinical settings [6, 68]. Furthermore, the critical impact of data preprocessing steps on the analysis of next-generation sequencing data adds another layer of complexity that must be standardized for robust clinical translation [70]. There is a pressing need for standardized protocols and analytical frameworks to ensure reproducibility across different laboratories and platforms [46]. Additionally, the digital nature of massively parallel sequencing data introduces specific challenges in AML samples with low tumor cellularity or high stromal contamination, potentially leading to false-negative results or inaccurate quantification of epigenetic modifications [71]. Future efforts must focus on developing streamlined, cost-effective epigenetic profiling platforms that can be integrated into routine diagnostic workflows without compromising analytical depth or accuracy.

Biological complexity and heterogeneity

The remarkable epigenetic heterogeneity within AML tumors presents a fundamental challenge for developing unified classification systems and targeted therapies. Intratumoral epigenetic diversity enables functional specialization among leukemic subclones, facilitating adaptation to therapeutic pressures and contributing to relapse [1, 46]. This heterogeneity is further complicated by the dynamic interplay between genetic and epigenetic alterations, where mutations in epigenetic regulators such as DNMT3A, TET2, and IDH1/2 initiate widespread epigenetic reprogramming that evolves throughout disease progression [5, 72]. The bone marrow microenvironment adds another layer of complexity, providing niche-specific signals that influence the epigenetic state of LSCs and promote therapy resistance [68, 73]. Future research must prioritize longitudinal studies tracking epigenetic evolution from diagnosis through relapse to identify stable epigenetic vulnerabilities that can be therapeutically targeted despite ongoing clonal evolution. Additionally, greater emphasis on understanding the spatial organization of epigenetic states within the bone marrow niche will be crucial for developing effective combination therapies.

Therapeutic development Obstacles

Current epigenetic therapies, including HMAs and histone deacetylase inhibitors, demonstrate limited efficacy as monotherapies and often produce heterogeneous responses across molecularly defined AML subtypes [72, 73]. The non-specific nature of these agents leads to widespread epigenetic modulation rather than targeted correction of disease-specific alterations, resulting in off-target effects and variable therapeutic windows [72]. Additionally, the development of resistance remains a significant concern, with multiple mechanisms including upregulation of alternative epigenetic regulators, metabolic adaptation, and selection of pre-resistant subclones undermining long-term efficacy [1, 73]. Future drug development should focus on next-generation epigenetic therapies with improved specificity, such as inhibitors targeting specific chromatin readers, writers, or erasers that are dysregulated in particular AML subtypes [72, 74, 75]. Combination strategies that simultaneously target epigenetic mechanisms and synergistic pathways offer particular promise, but require careful optimization to maximize efficacy while minimizing toxicity [6, 73]. The development of rational therapeutic combinations will depend on improved understanding of epigenetic networks and their functional interactions with genetic and signaling pathways in AML pathogenesis.

Clinical translation challenges

The translation of epigenetic biomarkers into clinical practice faces significant validation and standardization hurdles. While numerous studies have proposed epigenetic classification systems and prognostic signatures, few have undergone rigorous validation in prospective clinical trials or across diverse patient populations [6, 76, 77]. The integration of epigenetic risk assessment into established frameworks such as ELN guidelines requires demonstration of independent prognostic value and clinical utility beyond current genetic markers [6, 73]. Additionally, the development of clinically applicable assays for monitoring epigenetic evolution and treatment response presents technical and logistical challenges, particularly for DNA methylation-based biomarkers that may exhibit tissue-specific patterns and dynamic changes over time [69, 71]. Implementation barriers including cost reimbursement, regulatory approval, and interoperability with existing diagnostic platforms further complicate clinical adoption [68, 69]. Future efforts should prioritize large-scale collaborative studies validating epigenetic biomarkers in prospective cohorts and developing standardized assays that can be deployed across clinical laboratories. The incorporation of epigenetic profiling into MRD monitoring represents a particularly promising application that may enable earlier detection of relapse and guide preemptive therapeutic interventions [6, 73].

Future perspectives and directions

The future of epigenetic research in AML lies in the development of integrated multidimensional approaches that capture the complexity of the epigenetic landscape and its functional consequences. Technological innovations including third-generation sequencing, spatial multi-omics, and CRISPR-based epigenetic screening will provide unprecedented insights into the regulatory architecture of AML and its therapeutic vulnerabilities [46, 69], aided by sophisticated deep learning frameworks for predicting transcriptional regulators from epigenomic data [78]. The application of AI and ML approaches to large-scale epigenetic datasets will be essential for identifying predictive patterns and generating clinically actionable classifiers [6], building on foundational work in network-based analysis of complex biological systems across tissues [79]. From a therapeutic perspective, the development of more specific epigenetic drugs and rational combination regimens holds promise for overcoming current limitations and achieving durable responses [72, 73, 80]. Clinically, the integration of epigenetic profiling into dynamic risk assessment models that evolve throughout the disease course may enable truly personalized treatment approaches adapted to each patient’s changing disease biology [6, 68]. Realizing this vision will require collaborative efforts across disciplines including basic science, computational biology, drug development, and clinical trial design to translate our growing understanding of epigenetic mechanisms into improved outcomes for AML patients.

Conclusion

The journey “Beyond the DNA Sequence” reveals a complex and dynamic epigenetic landscape that is fundamental to the pathogenesis and clinical behavior of AML. This review has underscored that genetic mutations alone provide an incomplete picture, while epigenetic alterations, including DNA methylation patterns, histone modifications, 3D chromatin architecture, and regulatory ncRNAs, offer a deeper, more nuanced understanding of disease heterogeneity, prognosis, and therapeutic vulnerabilities. The advent of single-cell multi-omics is particularly transformative, enabling the dissection of this complexity at an unprecedented resolution and identifying the resilient epigenetic circuits within LSCs that drive relapse. Translating these insights into the clinic through epigenetic risk classifiers, sensitive eMRD monitoring, and biomarkers for targeted therapies represents the next frontier in AML management. Despite challenges in standardization and clinical integration, the concerted effort to map the epigenetic landscape is poised to refine prognostic models and unlock novel, personalized therapeutic interventions, moving us closer to improving survival for patients with this challenging disease.

Author contributions

Hamed Soleimani Samarkhazan wrote the main manuscript text and prepared figures and Jiqian Xie Revise the manuscript. All authors read final version of manuscript.

Funding

This research did not receive any financial support from public, commercial, or nonprofit organizations.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.