Integrative Biomedical Research

1. Introduction

Acute myeloid leukaemia (AML) is, at its core, a disease of arrested identity. It is an aggressive hematologic malignancy characterised by the clonal expansion of undifferentiated myeloid precursor cells that accumulate in the bone marrow and peripheral blood, progressively crowding out normal haematopoiesis (Schrezenmeier & Huntly, 2025; Zhang & Zhang, 2025). For a long time — perhaps too long — the field explained this arrest almost exclusively through the language of somatic mutation and chromosomal translocation, treating AML as a genetic bookkeeping problem: tally the lesions, assign a risk category, choose a chemotherapy backbone (Anandappa et al., 2026; Schrezenmeier & Huntly, 2025). That framework was not wrong, exactly. It was incomplete.

What the genetic lens tends to underplay is that DNA in the nucleus is not simply a linear code waiting to be read; it is a physical, three-dimensional object, folded into loops, domains, and compartments that determine which distal enhancers can actually reach which gene promoters (Fischer et al., 2024; King et al., 2025). This is worth pausing on, because it changes the unit of analysis. A mutation does not have to alter a coding sequence to be pathogenic — it can simply alter where a piece of DNA sits in space relative to its regulatory partners. In AML, this spatial organisation is profoundly and repeatedly hijacked, enforcing aberrant transcriptional programmes that block differentiation and sustain leukaemic self-renewal (Schrezenmeier & Huntly, 2025; Wu & Zhang, 2026).

Among the machinery responsible for higher-order chromatin structure, the cohesin complex — built around the core subunits STAG2, RAD21, SMC1A, and SMC3 — functions as something close to a gatekeeper of spatial chromatin looping and genomic insulation (Fischer et al., 2024; Schrezenmeier & Huntly, 2025). Mutations affecting cohesin genes are found in roughly 10% to 15% of AML cases, and the X-linked subunit STAG2 is, by a wide margin, the one most frequently affected (Fischer et al., 2024; Schrezenmeier & Huntly, 2025). What is striking, and not entirely intuitive, is how unevenly these mutations distribute across cytogenetic subgroups: they are markedly enriched in t(8;21) core-binding factor (CBF) AML, where they frequently co-occur with mutations in epigenetic regulators such as ASXL1, ASXL2, and EZH2, yet they are almost never seen in inv(16) CBF-AML (Duployez et al., 2016; Fischer et al., 2024). Why one CBF subtype should tolerate — or even favour — cohesin loss while its sibling subtype does not remains an open and genuinely interesting question.

Mechanistically, loss of STAG2 substantially reduces cohesin occupancy across the genome and selectively erodes short-range enhancer-promoter interactions, producing either local chromatin deactivation or, somewhat paradoxically, aberrant activation depending on genomic context (Fischer et al., 2024). One might expect the paralogous subunit STAG1 to step in and compensate, but it does not: STAG1-containing complexes cannot substitute for STAG2 at these short-range loops, leaving the genome biochemically exposed at precisely the sites that matter most for lineage commitment (Fischer et al., 2024). In hematopoietic stem and progenitor cells (HSPCs), STAG2 deficiency impairs terminal myeloid and erythroid differentiation while promoting an aberrant, almost stubborn self-renewal phenotype (Fischer et al., 2024; Schrezenmeier & Huntly, 2025). At the mechanistic level, this manifests as a paradoxical increase in overall chromatin accessibility, which in turn boosts binding of transcription factors such as RUNX1 at key target sites and cooperates with leukaemic oncoproteins to drive myeloid transformation (Duployez et al., 2016).

Cohesin mutations are only one route into this same architectural problem. AML-associated oncofusion proteins take a more direct approach, physically rewiring the 3D genome and reshaping DNA accessibility landscapes on their own terms (Saeed et al., 2012). Fusion oncoproteins such as PML-RARA, characteristic of t(15;17) acute promyelocytic leukaemia (APL), and AML1-ETO, arising from t(8;21) AML, preferentially dock onto already-open, accessible chromatin regions (Saeed et al., 2012). Once bound, they do not act passively; they actively recruit repressive machinery — histone deacetylases and Polycomb complexes among them — to install a locally hypoacetylated, transcriptionally muted chromatin state at enhancers and promoters that would otherwise be active (King et al., 2025; Saeed et al., 2012).

PML-RARA in particular seems to operate at a genome-wide scale rather than locus by locus (Wang et al., 2016). It triggers a broad collapse of active chromatin loops, depleting the enhanceosome and severing the RNA Polymerase II (RNAPII)-mediated physical bridges that normally connect promoters to their distal regulatory elements. The consequence of this collapse is the eviction of myeloid-lineage transcription factors — PU.1, IRF1, CEBPB among them — along with the coactivator p300, from promoters and super-enhancers that would otherwise drive the myeloid differentiation programme forward. The net effect is a suppressed differentiation programme and, ultimately, leukemogenesis.

Perhaps the clearest illustration of how disrupted loop architecture translates into disease biology comes from the regulation of the proto-oncogene MYC (Pulikkan et al., 2018; Schrezenmeier & Huntly, 2025). In inv(16) AML, the fusion protein CBFb-SMMHC sustains leukaemic cell viability by preventing RUNX1 from carrying out its normal job of repressing MYC transcription (Pulikkan et al., 2018). Pharmacologically severing the CBFb-SMMHC–RUNX1 interaction — for instance with the small-molecule inhibitor AI-10-49 — allows RUNX1 to re-engage three distal downstream enhancers (ME1, ME2, and E3) that are physically looped to the MYC promoter. RUNX1 then recruits the Polycomb Repressive Complex 1 (PRC1) subunit RING1B in place of the activating SWI/SNF component BRG1, an enhancer-complex switch that produces local chromatin compaction, deposition of the repressive H3K27me3 mark, and rapid MYC downregulation — ultimately triggering apoptosis in the leukaemic cells.

A structurally different, but conceptually related, mechanism plays out through long non-coding RNAs (lncRNAs) and intronic regulatory elements (Camera et al., 2023). In NPM1-mutated AML, transcripts arising from intron 8 of the FTO gene (termed FTO-lncAML) promote physical contact between an intronic enhancer cluster and the promoter of the Iroquois homeodomain transcription factor gene IRX3. This intronic module effectively locks cells out of differentiation; depleting FTO-lncAML dissolves the enhancer-promoter loop, downregulates IRX3 together with HOXA stemness programmes, and releases the differentiation block.

These insights into disrupted spatial topology are, encouragingly, already reshaping AML therapeutics rather than sitting purely in the academic literature. Recent work has characterised so-called "3D chromatin hubs" as spatial regulatory centres that coordinate expression of genes essential for leukaemic identity and survival (Gambi et al., 2025). Master transcription factors — MYB most notably — have emerged as critical organisers of these hubs, and pharmacological degradation or inhibition of MYB dissolves the hub structure, collapsing the loop networks sustaining oncogenes such as MYC and NOTCH1. This offers a genuinely novel epigenetic strategy for overcoming therapeutic resistance and, potentially, eradicating leukaemia stem cells (LSCs) (Wu & Zhang, 2026).

Despite this progress, an integrated account of how genetic lesions translate into physical 3D genome disruption — and from there into transcriptional dysregulation — remains, frankly, incomplete. Existing work has tended to characterise individual fusions or individual chromatin-modifying enzymes in isolation rather than as parts of a connected system. This review sets out to address that gap through four specific objectives.

First, to systematically delineate the molecular mechanisms by which mutations in the cohesin complex — particularly STAG2 and RAD21 — and epigenetic modifiers reshape 3D genome looping, enhancer-promoter contacts, and transcriptional control in AML (Duployez et al., 2016; Fischer et al., 2024; Schrezenmeier & Huntly, 2025).

Second, to evaluate the spatial chromatin alterations induced by major AML-associated oncoproteins — PML-RARA, AML1-ETO, and CBFb-SMMHC among them — focusing on how each disrupts enhanceosome assembly, drives localised histone hypoacetylation, and silences the genes that define differentiation (Pulikkan et al., 2018; Saeed et al., 2012; Wang et al., 2016).

Third, to investigate long-range enhancer-promoter loop dynamics at key oncogenic loci, including MYC, IRX3, and the HOX clusters, and to examine how non-coding RNA elements such as FTO-lncAML act in both cis and trans to stabilise aberrant, differentiation-resistant topological states (Camera et al., 2023; Pulikkan et al., 2018; Schrezenmeier & Huntly, 2025).

Fourth, to trace the translation of 3D genomics into clinical practice, with particular attention to therapeutic targeting of chromatin hub regulators such as MYB, pharmacological disruption of oncogenic loops, and the use of single-cell epigenomic profiling — scATAC-seq and CUT&Tag among the relevant platforms — for precision risk stratification and minimal residual disease (MRD) tracking (Anandappa et al., 2026; Gambi et al., 2025; Schrezenmeier & Huntly, 2025; Wu & Zhang, 2026).

2. The Malignant Epigenetic–Transcriptional Network in Acute Myeloid Leukaemia

For decades, the working model of AML pathogenesis was reassuringly linear — the so-called "two-hit" hypothesis, in which leukemogenesis required the cooperative convergence of Class I mutations driving proliferation and survival (tyrosine kinase signalling, for example) with Class II mutations disrupting the transcription factors responsible for hematopoietic differentiation (Duployez et al., 2016; Schrezenmeier & Huntly, 2025). That framework served its purpose; it gave clinicians a workable way to subclassify disease and choose treatment. But next-generation sequencing and multi-omic profiling have, over roughly the past decade, quietly dismantled its underlying assumption of simplicity (Anandappa et al., 2026; Schrezenmeier & Huntly, 2025). What emerges instead — and this is really the organising idea of the present review — is a densely interconnected network in which myeloid precursors do not merely stall due to one or two isolated lesions, but undergo a coordinated, almost systemic reprogramming of their epigenetic, transcriptional, and spatial architecture (Wu & Zhang, 2026).

To make this network tractable, we have organised the literature into four interacting functional layers, or nodes: a DNA methylation and metabolic node, a higher-order 3D chromatin topology node, an epitranscriptomic regulatory node, and a targeted clinical translation node (Figure 1). Genetic fusions and somatic mutations act across these layers simultaneously, physically reshaping the nuclear environment to enforce differentiation arrest and sustain leukaemic stem cell (LSC) self-renewal.

2.1 Node I — Epigenetic Remodelling and DNA Methylation Dynamics

At what might be called the foundational layer of the leukaemic network sits DNA methylation — a mechanism that, under normal physiological conditions, governs lineage commitment during haematopoiesis (Schrezenmeier & Huntly, 2025). In AML, this landscape is not simply perturbed but systematically inverted in places: promoter CpG islands become hypermethylated, silencing tumour suppressors, while specific enhancer elements become hypomethylated, aberrantly switching on oncogenes (Schrezenmeier & Huntly, 2025). The somatic mutations underlying these changes tend to converge, somewhat predictably once you know to look for it, on the enzymes that write, read, or erase methylation marks (Quattrocchi et al., 2023; Schrezenmeier & Huntly, 2025).

DNMT3A, which encodes a de novo DNA methyltransferase, is mutated in approximately 20% to 25% of adult AML cases — a remarkably high frequency for a single gene (Schrezenmeier & Huntly, 2025). The dominant lesion is a single-amino-acid substitution at codon 882, replacing arginine with histidine (DNMT3A R882), which disrupts the protein's oligomerisation interface (Schrezenmeier & Huntly, 2025). This hypomorphic variant produces focal hypomethylation and, in doing so, tips the balance toward stem cell self-renewal at the expense of terminal differentiation (Schrezenmeier & Huntly, 2025). Interestingly, and this is a detail easily overlooked, systematic profiling of non-R882 variants shows that roughly 74% of these lead to outright loss of methyltransferase activity, with about half undergoing rapid, proteasome-mediated degradation via the DCAF8 E3 ubiquitin ligase adaptor (Schrezenmeier & Huntly, 2025).

This methylation machinery does not operate in isolation; it is tightly coupled to cellular metabolism, most conspicuously through mutant isocitrate dehydrogenase 1 and 2 (IDH1/IDH2) (Schrezenmeier & Huntly, 2025). Wild-type IDH1/2 normally convert isocitrate to α-ketoglutarate (α-KG), a co-factor required by the ten-eleven translocation 2 (TET2) DNA demethylase (Schrezenmeier & Huntly, 2025). Mutant IDH1/2 enzymes acquire a neomorphic activity that instead reduces α-KG to the oncometabolite 2-hydroxyglutarate (2-HG) (Schrezenmeier & Huntly, 2025). Accumulated 2-HG competitively inhibits TET2, producing a global DNA hypermethylation phenotype that effectively locks the differentiation programme shut (Schrezenmeier & Huntly, 2025).

This same epigenetic blockade can also arise through a purely metabolic route, without any direct mutation in IDH1 or IDH2 (Schrezenmeier & Huntly, 2025). Overexpression of the branched-chain amino acid transaminase BCAT1 depletes intracellular α-KG pools, reproducing — almost identically — the hypermethylation profile and stem-cell-like signature otherwise seen in IDH1/2-mutant AML (Schrezenmeier & Huntly, 2025). Aberrant methyltransferase activity also cooperates directly with lineage-specific oncogenic transcription factors to sculpt distinct disease subtypes (Choudhury et al., 2026). In pediatric acute megakaryoblastic leukaemia (AMKL), the CBFA2T3-GLIS2 (C/G) fusion protein partners physically with the de novo methyltransferase DNMT3B to drive a highly specific promoter hypermethylation programme (Choudhury et al., 2026). Somewhat counterintuitively,

Figure 1. Conceptual map of AML pathobiology organised into four interacting functional nodes — epigenetic remodelling and metabolism, 3D genomic chromatin architecture, epitranscriptomic regulation, and targeted clinical translation. Representative molecular drivers are listed beneath each node to illustrate how distinct genetic and biochemical lesions converge on shared regulatory outcomes. This schematic synthesises mechanistic relationships described throughout Sections 2 and 4 and is not itself a primary data figure.

Figure 2. Schematic comparison of enhancer-promoter looping at a representative cohesin-dependent locus under wild-type versus STAG2-deficient conditions. Left panel: an intact STAG2-cohesin ring stabilises a short-range (<500 kb) loop, permitting active transcription of loci such as ITGA9/DACT1. Right panel: STAG2 loss collapses the loop, and the paralogous subunit STAG1 fails to functionally compensate, resulting in loss of enhancer-promoter contact and gene silencing. Conceptually adapted from mechanistic data reported by Fischer et al. (2024).

C/G-positive leukaemia relies on a dual gene-expression strategy in which both hypomethylated active promoters (CMTM5, GP1BA) and hypermethylated promoters (HPSE2, DLX3) coexist within active chromatin states to drive disease progression (Choudhury et al., 2026); knocking out DNMT3B restores normal methylation and re-sensitises otherwise resistant clones to apoptotic therapy (Choudhury et al., 2026).

2.2 Node II — Higher-Order Chromatin Topology and 3D Genomic Architecture

Within the eukaryotic nucleus, genetic material is not simply coiled at random; it is arranged into organised loops, topologically associating domains (TADs), and multi-connected "3D chromatin hubs" (Gambi et al., 2025; Schrezenmeier & Huntly, 2025). This architecture regulates transcription by bringing distal enhancers into physical proximity with target promoters (Fischer et al., 2024; Gambi et al., 2025). In AML, both structural mutations and oncofusion proteins converge on this architecture, reshaping it to block differentiation pathways and sustain leukaemic transcriptional programmes (Fischer et al., 2024; Saeed et al., 2012).

The multi-subunit cohesin complex — RAD21, SMC1A, SMC3, together with either STAG1 or STAG2 — is the principal driver of chromatin loop extrusion and genome insulation (Fischer et al., 2024; Schrezenmeier & Huntly, 2025). As noted above, cohesin mutations affect roughly 10–15% of AML cases, with STAG2 the most frequent target, and they show strikingly non-random subtype distribution: enriched in t(8;21) CBF-AML, essentially absent in inv(16) CBF-AML (Duployez et al., 2016; Fischer et al., 2024). Loss of STAG2 substantially reduces overall chromatin-bound cohesin and selectively erases short-range regulatory loops under roughly 500 kb, without measurably affecting the larger, CTCF-anchored structural domains, which remain preferentially occupied by residual STAG1-cohesin complexes (Fischer et al., 2024). The functional consequence is real and specific: STAG2-deficient hematopoietic stem and progenitor cells show impaired terminal myeloid and erythroid differentiation and remain locked in an immature, stem-like state, in part because genes such as the cell-migration regulator ITGA9 and the tumour suppressor DACT1 lose the short-range enhancer contacts they depend upon (Fischer et al., 2024) (see Figure 2).

Oncofusion proteins reach the same architectural endpoint by a different route (Saeed et al., 2012). PML-RARA binds accessible chromatin regions containing retinoic acid response elements and, once recruited, triggers a collapse of active chromatin loops together with loss of RNAPII-mediated promoter-enhancer connectivity (Wang et al., 2016). This physical dissociation strips key myeloid transcription factors — PU.1, IRF1, CEBPB — from differentiation-defining genes, effectively freezing the cell in a promyelocytic state (Wang et al., 2016).

A related but mechanistically distinct example plays out in inv(16) AML, where CBFb-SMMHC sustains blast viability by blocking RUNX1-mediated repression of MYC (Pulikkan et al., 2018). Pharmacological disruption of this interaction with AI-10-49 allows RUNX1 to bind three distal enhancers (ME1, ME2, E3) physically looped to the MYC promoter, triggering an enhancer-complex switch — RING1B for BRG1 — that compacts chromatin, deposits H3K27me3, and rapidly silences MYC, precipitating apoptosis (Pulikkan et al., 2018).

Non-coding RNA adds a further layer of complexity to this picture. Chromatin loops can also be rewired by unannotated lncRNAs originating from intronic enhancers (Camera et al., 2023). In roughly 10% of AML cases — most notably those bearing NPM1 mutations — transcripts arising from intron 8 of FTO (FTO-lncAML) stabilise physical contact between an intronic enhancer cluster and the IRX3 promoter (Camera et al., 2023). This enhancer-hijacking module locks cells in an undifferentiated state; depleting FTO-lncAML dissolves the loop, downregulates both IRX3 and HOXA programmes, and lifts the differentiation block (Camera et al., 2023).

2.3 Node III — Epitranscriptomic Regulatory Dynamics and Translational Integrity

Beyond the structural genome, the leukaemic network is heavily regulated at the post-transcriptional level, through chemical modification of RNA bases — a layer of control now generally referred to as the epitranscriptome (Quattrocchi et al., 2023; Schrezenmeier & Huntly, 2025). The most abundant of these internal mRNA modifications is N6-methyladenosine (m6A), which governs mRNA stability, splicing, and the kinetics of translation (Quattrocchi et al., 2023; Schrezenmeier & Huntly, 2025).

m6A marks are installed co-transcriptionally by a writer complex built around METTL3 and METTL14 (Quattrocchi et al., 2023). Both are consistently upregulated in AML and function as bona fide oncogenic drivers (Quattrocchi et al., 2023). Mechanistically, METTL3 localises to the transcription start sites of active genes and deposits m6A within coding regions, which prevents ribosome stalling and boosts translation of stemness-associated transcripts such as MYC, PTEN, and BCL2 (Quattrocchi et al., 2023). Downstream of this, the cytoplasmic reader YTHDF2 selectively targets differentiation-inducing transcripts for degradation, while YTHDC1 forms nuclear condensates that stabilise leukaemic transcripts and further reinforce the differentiation block (Quattrocchi et al., 2023) (Figure 3).

These marks are, of course, reversible — m6A is erased by demethylases such as FTO and ALKBH5 (Quattrocchi et al., 2023). FTO is highly expressed in KMT2A-rearranged, PML-RARA-positive, and NPM1-mutant leukaemias, where it demethylates targets including RARA to suppress differentiation (Quattrocchi et al., 2023). This reversibility is not merely of academic interest: the small-molecule FTO inhibitor FB23-2 meaningfully reduces cell viability, blocks self-renewal, and induces differentiation in primary AML blasts, underscoring the pathway's therapeutic tractability (Quattrocchi et al., 2023).

Epitranscriptomic rewiring is not restricted to mRNA; it extends to ribosomal RNA (rRNA) and transfer RNA (tRNA) as well (Quattrocchi et al., 2023). In t(8;21) AML, AML1-ETO recruits the transcription factor AES to upregulate C/D box small nucleolar RNAs, driving overexpression of fibrillarin (FBL), a methyltransferase catalysing rRNA 2'-O-methylation (Quattrocchi et al., 2023). This ribosomal remodelling increases translation efficiency for stemness transcripts, sustaining self-renewal in leukaemia-initiating cells (Quattrocchi et al., 2023).

2.4 Node IV — Targeted Clinical Translation and Epigenetic Triplets

Mechanistic clarity across these three nodes has, gratifyingly, translated into genuinely effective therapies (Pienkowski et al., 2025; Watts et al., 2026). Among these, disrupting the chromatin-binding interaction between menin and KMT2A (MLL1) stands out as one of the most clinically successful strategies to date (Schrezenmeier & Huntly, 2025; Watts et al., 2026).

Menin acts as an obligate cofactor for both wild-type KMT2A and oncogenic KMT2A fusions, binding the N-terminus of KMT2A on chromatin to activate leukemogenic programmes such as HOXA9 and MEIS1 (Schrezenmeier & Huntly, 2025; Watts et al., 2026). Menin inhibitors (MENi) physically occlude this interaction, displacing the fusion protein from chromatin and downregulating its oncogenic targets (Watts et al., 2026). First-in-class agents — the pyrimidine/triazine compound revumenib (FDA-approved 2024) and the thienopyrimidine derivative ziftomenib (FDA-approved 2025) — achieve overall response rates around 50% or higher in relapsed/refractory disease (Watts et al., 2026).

These response rates are genuinely encouraging, yet single-agent efficacy is frequently undermined by rapid resistance (Watts et al., 2026). Within roughly 60 days of revumenib exposure, somatic point mutations emerge in MEN1 at codons G331, M327, and T349 (Watts et al., 2026). These mutations produce a steric clash that blocks drug binding while leaving the essential KMT2A interaction intact, rendering the inhibitor ineffective (Watts et al., 2026) (Figure 4). To pre-empt this, combination trials are now evaluating frontline "triplet" regimens pairing MENi with hypomethylating agents (such as azacitidine) and the BCL-2 inhibitor venetoclax (Watts et al., 2026); these combinations suppress subclonal outgrowth and produce high response rates in treatment-naïve patients (Watts et al., 2026).

Resistance can also be approached from the signalling side rather than the epigenetic side (Pienkowski et al., 2025). Highly proliferative AML clones frequently carry FLT3-ITD or c-KIT mutations that drive active tyrosine kinase signalling (Pienkowski et al., 2025). The glycolysis inhibitor 2-deoxy-D-glucose (2-DG) disrupts N-linked glycosylation, preventing these receptor tyrosine kinases from reaching the cell membrane, which in turn suppresses downstream STAT5 and ERK signalling and re-sensitises tyrosine-kinase-inhibitor-resistant clones to cytarabine and venetoclax (Pienkowski et al., 2025).

Finally, the epitranscriptomic machinery is itself entering

Figure 3. Schematic of co-transcriptional m6A installation at DRACH consensus motifs by the METTL3–METTL14–WTAP writer complex, and its bifurcated downstream consequence: stabilisation and enhanced translation of oncogenic transcripts (MYC, BCL2, PTEN) versus YTHDF2-mediated degradation of differentiation-associated transcripts, jointly sustaining leukaemia stem cell self-renewal. Synthesised from mechanistic findings reported by Quattrocchi et al. (2023).

Figure 4. Conceptual pathway from menin-inhibitor monotherapy (revumenib/ziftomenib) through target displacement of the KMT2A fusion, to the two principal escape routes reported in the clinical literature — direct MEN1 target-site mutation and FLT3-ITD/NRAS signalling bypass — culminating in the rationale for frontline 'epigenetic triplet' therapy combining a menin inhibitor with venetoclax and a hypomethylating agent. Synthesised from Watts et al. (2026).

clinical evaluation (Quattrocchi et al., 2023). Orally bioavailable METTL3 inhibitors — STM2457 and its clinical derivative STC-1 — selectively induce apoptosis in leukaemic cells while sparing healthy haematopoietic progenitors, offering a plausible route to eliminating the LSC reservoir and delaying relapse (Quattrocchi et al., 2023).

3. Methods

Because this manuscript synthesises mechanistic and translational evidence on chromatin architecture in AML rather than reporting a single primary experiment, we adopted a structured narrative-review methodology, reported here in enough operational detail that another investigator could reconstruct our search and selection process — an expectation broadly consistent with PubMed/MEDLINE indexing standards for reproducible reviews.

3.1 Search Strategy and Information Sources

We searched PubMed/MEDLINE, Scopus, and Web of Science from database inception through June 2026, supplemented by manual screening of reference lists from all included articles (a form of backward citation chasing) and of citing articles identified through Google Scholar (forward citation chasing). Search terms combined controlled vocabulary (MeSH: "Leukemia, Myeloid, Acute"; "Chromatin Assembly and Disassembly"; "Cohesins"; "Epigenesis, Genetic") with free-text keywords, connected using Boolean operators — for example: ("acute myeloid leukemia" OR "AML") AND ("chromatin architecture" OR "3D genome" OR "chromatin loop*" OR "TAD" OR "cohesin" OR "STAG2" OR "enhancer-promoter" OR "epitranscriptom*" OR "m6A"). No language restriction was applied at the search stage, though non-English full texts were subsequently excluded at screening for practical reasons.

3.2 Eligibility Criteria

Studies were eligible for inclusion if they (a) reported primary experimental or computational data, or synthesised such data, on chromatin topology, cohesin biology, oncofusion-protein chromatin binding, epitranscriptomic regulation, or targeted therapeutics in human AML or closely related myeloid neoplasms; (b) were published in a peer-reviewed journal between 2012 and 2026; and (c) were available in full text in English. Studies were excluded if they addressed chromatin biology exclusively in non-myeloid malignancies without direct relevance to AML mechanisms, were conference abstracts without corresponding full manuscripts, or were non-systematic opinion pieces lacking a clearly described evidentiary basis.

3.3 Study Selection and Data Extraction

Titles and abstracts identified through the search strategy were screened independently against the eligibility criteria described above; full texts of potentially eligible records were then retrieved and assessed in full. For each included study, we extracted, into a structured data-extraction template, the AML subtype or model system studied, the specific molecular target or pathway examined (e.g., STAG2/cohesin, PML-RARA, METTL3/METTL14), the experimental or analytical methodology employed (e.g., Hi-C, ChIP-seq, CUT&Tag, single-cell RNA-seq, in vitro pharmacological perturbation), the principal quantitative or mechanistic findings, and — where reported — associated clinical or translational outcomes. Discrepancies in extraction were resolved by consensus discussion and, where necessary, re-review of the primary source.

3.4 Data Synthesis and Conceptual Mapping

Rather than pooling data quantitatively — which is neither appropriate nor feasible for a body of evidence spanning molecular biology, structural genomics, and early-phase clinical trials — we performed a structured narrative synthesis. Extracted findings were mapped onto four a priori conceptual nodes reflecting distinct but interacting layers of leukaemic biology: (I) epigenetic remodelling and metabolism, (II) 3D genomic chromatin architecture, (III) epitranscriptomic regulation, and (IV) targeted clinical translation. This mapping (Figure 1) was developed iteratively, refined against the extracted evidence, and cross-checked for internal consistency by tracing each molecular mechanism through to its reported functional and, where available, clinical consequence.

3.5 Table and Figure Construction

Four synthesis tables were constructed to consolidate the extracted evidence: Table 1 (molecular, cytogenetic, and clonal classification of AML subtypes), Table 2 (epigenetic and metabolic regulators), Table 3 (spatial chromatin topology alterations), and Table 4 (epigenetic, epitranscriptomic, and signalling therapeutics). Each table entry was populated directly from the extracted primary-source data described above, with every claim traceable to its cited reference. Four schematic figures (Figures 1–4) were then generated to visually represent, respectively, the overall four-node conceptual network, cohesin-dependent enhancer-promoter looping, the m6A epitranscriptomic pathway, and the trajectory from menin-inhibitor monotherapy to acquired clinical resistance; each figure was constructed to be a faithful, simplified visual translation of the underlying cited mechanistic data rather than a new empirical claim.

3.6 Reproducibility Statement

The search strategy, eligibility criteria, and data-extraction fields described above are reported in sufficient detail to allow independent replication. No individual patient-level data were generated or analysed; all synthesised findings derive from previously published, peer-reviewed sources, each cited at the point of use throughout this manuscript.

4. Chromatin Topology and Epitranscriptomic Regulation in Acute Myeloid Leukemia

4.1 Overview of the Synthesised Evidence Base

The literature retained after screening spans structural genomics (Hi-C and CUT&Tag-based chromatin conformation studies), functional genetics (CRISPR-based reconstitution and knockout experiments), and early-phase clinical trial reporting, collectively covering six major AML cytogenetic subtypes and four interacting biological layers, as summarised in Table 1. Across this evidence base, a consistent pattern emerged: whatever the initiating genetic lesion, downstream convergence onto 3D chromatin topology and epitranscriptomic control was the rule rather than the exception.

4.2 Molecular and Cytogenetic Landscape of AML Subtypes

Cytogenetic classification of AML continues to carry strong prognostic weight, but the mechanistic basis for that prognosis is increasingly attributable to chromatin-level biology rather than the fusion event alone (Table 1). Core-binding factor AML bearing t(8;21) shows a high co-mutation burden in chromatin modifiers (42%) and cohesin subunits (18%), whereas its inv(16) counterpart is nearly devoid of cohesin mutations, instead accumulating RAS-pathway lesions in over half of cases (Duployez et al., 2016). Both subtypes nonetheless carry favourable overall prognosis, though relapse rates of 33–40% in t(8;21) disease correlate specifically with co-occurring signalling and cohesin/chromatin lesions (Duployez et al., 2016; Fischer et al., 2024). Acute promyelocytic leukaemia [t(15;17)] remains the clearest example of a curable AML subtype once its chromatin-collapsing fusion protein is pharmacologically degraded (Saeed et al., 2012; Wang et al., 2016; Wu & Zhang, 2026), in sharp contrast to KMT2A-rearranged disease and pediatric AMKL, both of which carry adverse risk and depend on distinct, chromatin-anchored oncogenic mechanisms (Choudhury et al., 2026; Schrezenmeier & Huntly, 2025).

4.3 Epigenetic and Metabolic Regulatory Alterations

Table 2 summarises the five most extensively characterised epigenetic/metabolic regulators — DNMT3A, TET2, IDH1/IDH2, ASXL1, and EZH2 — and shows a coherent pattern: whether through direct enzymatic loss of function (TET2, EZH2), gain of a neomorphic metabolic activity (IDH1/2), or dominant-negative oligomerisation defects (DNMT3A R882), each lesion converges on a shared endpoint of aberrant chromatin compaction or de-compaction at lineage-defining loci, consistently favouring self-renewal over terminal differentiation (Schrezenmeier & Huntly, 2025).

4.4 Spatial Chromatin Topology Alterations

The five topological mechanisms catalogued in Table 3 — STAG2 cohesin defects, RAD21/SMC1A/SMC3 mutations, PML-RARA loop collapse, 3D chromatin hub formation, and FTO-lncAML enhancer hijacking — each independently disrupt enhancer-promoter connectivity, but through mechanistically distinct routes: selective short-range loop loss (STAG2), global loop-network collapse (PML-RARA), enhancer-complex switching (CBFb-SMMHC/MYC), or lncRNA-stabilised aberrant looping (FTO-lncAML). Notably, several of these alterations are reported as pharmacologically reversible — acute STAG2/RAD21 reconstitution rapidly reduces global chromatin loop size (Yang et al., 2026), and ATRA/ATO-induced PML-RARA degradation restores enhancer-promoter connectivity (Wang et al., 2016) — supporting the broader conceptual claim that chromatin topology in AML represents a druggable, rather than fixed, disease state.

4.5 Epitranscriptomic Contributions

METTL3/METTL14-mediated m6A deposition and YTHDF2-mediated selective mRNA decay together establish a translational bias favouring oncogenic

Table 1. This table delineates the genomic and cytogenetic landscape of AML across six major subtypes, integrating recurrent chromosomal translocations, somatic co-mutation profiles, molecular pathophysiology, ELN 2022 risk category, and associated prognosis. Data are synthesised from the primary sources cited in the right-hand column; each row represents a distinct, clinically actionable disease entity.

Table 2. This table catalogues the five most extensively characterised epigenetic writers, erasers, and metabolic enzymes mutated in AML, detailing their wild-type biochemical function, mutant-induced chromatin state, and downstream impact on myeloid lineage commitment and leukaemia stem cell self-renewal.

transcripts (MYC, BCL2, PTEN) over differentiation-associated transcripts (Quattrocchi et al., 2023) (Figure 3). This layer of control operates in parallel with, rather than downstream of, the structural chromatin alterations described above, and represents an independently targetable node, as evidenced by the selective pro-apoptotic activity of METTL3 inhibitors (STM2457/STC-1) in leukaemic but not healthy progenitor cells (Quattrocchi et al., 2023).

4.6 Therapeutic Landscape and Resistance Mechanisms

Table 4 catalogues five therapeutic classes — menin inhibitors, IDH1/2 inhibitors, FLT3/AXL tyrosine kinase inhibitors, METTL3 inhibitors, and MDR1-bypassing liposomal anthracyclines — alongside their principal mechanisms of acquired resistance. A recurring theme across all five classes is that resistance arises through one of two broad routes: direct target-site mutation (e.g., MEN1 G331/M327/T349; IDH1-D835) or signalling-pathway bypass (e.g., FLT3-ITD/NRAS subclonal outgrowth) (Table 4; Watts et al., 2026) (Figure 4). This pattern directly informed the rationale for combination "epigenetic triplet" regimens (menin inhibitor + venetoclax + a hypomethylating agent), which are reported to suppress subclonal outgrowth more effectively than any single agent (Watts et al., 2026).

5. Discussion: Reframing AML as a Disease of Spatial Genome Dysregulation

5.1 From Linear Genetics to an Interconnected Network

Perhaps the central message of this synthesis — if a single message can be extracted from a body of evidence this varied — is that AML resists being reduced to a simple genetic checklist. The classical two-hit model was never really wrong; it just described a much smaller slice of the true biology than anyone initially appreciated. What the four-node framework developed here (Figure 1) makes visible is that genetic lesions operate as entry points into a shared regulatory network, one in which DNA methylation, 3D chromatin folding, RNA modification, and downstream signalling reinforce one another rather than acting as independent variables (Wu & Zhang, 2026).

5.2 Convergent Vulnerability at the Level of Chromatin Topology

A finding that we think deserves more attention than it currently receives is the degree of mechanistic convergence at the level of 3D chromatin architecture. STAG2 loss, PML-RARA binding, and CBFb-SMMHC activity are, on paper, entirely unrelated molecular events — different genes, different pathways of origin, different cytogenetic subtypes — yet each ultimately narrows down to the same functional outcome: loss or reorganisation of enhancer-promoter contact (Fischer et al., 2024; Pulikkan et al., 2018; Wang et al., 2016) (Table 3; Figure 2). This convergence is, in a sense, good news clinically. It suggests that therapeutic strategies aimed at restoring or exploiting chromatin architecture — rather than targeting each individual oncoprotein separately — could plausibly have broader applicability across cytogenetically distinct AML subtypes than current subtype-specific approaches allow.

5.3 The Epitranscriptome as an Underappreciated Layer of Control

It is tempting, when discussing chromatin biology, to focus almost exclusively on DNA-level events and treat RNA modification as a secondary or downstream phenomenon. The evidence reviewed here (Quattrocchi et al., 2023) argues against that framing. METTL3/METTL14-dependent m6A deposition operates as an independent, parallel control point that reinforces — but does not simply mirror — chromatin-level differentiation blocks (Figure 3). The selective sensitivity of leukaemic cells, but not healthy progenitors, to METTL3 inhibition (Quattrocchi et al., 2023) hints at a therapeutic window that is mechanistically distinct from, and potentially combinable with, chromatin-directed or menin-directed strategies.

5.4 Resistance as a Predictable Consequence of Single-Node Targeting

The resistance patterns catalogued in Table 4 and Figure 4 are, on reflection, not particularly surprising once the network framing is taken seriously. If leukaemic transcription is sustained by multiple, partially redundant nodes, then blocking only one node — however effectively — leaves the others available for compensatory escape, whether through direct target mutation (MEN1 G331/M327/T349) or through signalling bypass (FLT3-ITD/NRAS outgrowth) (Watts et al., 2026). This reasoning provides a coherent rationale for the field's evident move toward combination "epigenetic triplet" regimens, and arguably argues for extending that combinatorial logic to include epitranscriptomic and topological agents alongside the more established menin- and BCL2-directed drugs.

5.5 Limitations

Table 3. This table summarises five distinct mechanisms by which cohesin-complex mutations, oncofusion proteins, 3D chromatin hubs, and long non-coding RNAs physically rewire the three-dimensional genome to drive or sustain leukaemic transcriptional programmes, together with their reported reversibility and therapeutic targeting strategy.

Table 4. This table outlines five current and emerging targeted therapeutic classes in AML, detailing their primary molecular target, mechanism of action, active clinical trial combinations, and the molecular basis of acquired secondary resistance reported to date.

Several limitations of the present synthesis warrant acknowledgement. First, as a narrative rather than quantitative systematic review, it does not provide pooled effect estimates and remains subject to a degree of selection bias in the studies emphasised. Second, much of the mechanistic evidence derives from cell-line and murine model systems, and translation to primary patient material — while present in several of the reviewed studies — is not universal. Third, the clinical-trial data summarised in Table 4 largely reflect early-phase or ongoing studies, and response durability and long-term survival outcomes for several combination regimens remain to be fully established. Finally, because the field is moving quickly, some of the resistance mechanisms and combination strategies described here may already be superseded by findings published after our search cut-off of mid-2026.

5.6 Future Directions

Looking ahead, the most promising direction may not be the discovery of additional individual chromatin regulators, but the systematic mapping of how existing nodes interact — for instance, whether STAG2-deficient clones show differential sensitivity to METTL3 inhibition, or whether single-cell multi-omic profiling (scATAC-seq, CUT&Tag) can prospectively identify which patients are most likely to benefit from chromatin-directed versus epitranscriptomic-directed therapy (Anandappa et al., 2026; Gambi et al., 2025). Such work would move the field from a descriptive network model toward a genuinely predictive, patient-specific one.

Taken as a whole, this review argues that acute myeloid leukaemia is best understood not as a static collection of driver mutations but as a dynamic, layered disruption of nuclear architecture. Cohesin loss, oncofusion-driven loop collapse, aberrant DNA methylation, and m6A-dependent translational bias each contribute to a shared endpoint: a stable, self-reinforcing block on myeloid differentiation. Because these marks and structures are, in principle, reversible, they constitute genuinely druggable vulnerabilities rather than fixed genetic sentences. Menin inhibitors, IDH1/2 inhibitors, and emerging METTL3-directed compounds already demonstrate that targeting this network — rather than any single mutation in isolation — can produce meaningful clinical responses, even as resistance mechanisms confirm the network's underlying redundancy. Moving forward, integrating single-cell epigenomic and epitranscriptomic profiling into risk stratification, and designing rational combination regimens that target multiple nodes simultaneously, are likely to be the most productive paths toward durable remission. In short, mapping the spatial and chemical architecture of the leukaemic genome is no longer a peripheral academic exercise; it is fast becoming central to how AML will be diagnosed, risk-stratified, and treated.

6. Conclusion

This review reframes acute myeloid leukaemia as a disease of disrupted spatial genome logic rather than a simple accumulation of driver mutations. STAG2/cohesin loss, PML-RARA– and CBFb-SMMHC–driven loop remodelling, and METTL3/METTL14-dependent epitranscriptomic control converge on a common outcome: a stable, self-reinforcing block on myeloid differentiation that sustains leukaemia stem cell self-renewal. Because these chromatin and RNA-modification states are pharmacologically reversible, they represent genuinely actionable therapeutic targets, as demonstrated by the clinical success — and characteristic resistance patterns — of menin inhibitors, IDH1/2 inhibitors, and emerging METTL3-directed compounds. The recurring emergence of resistance through target-site mutation or signalling bypass reinforces the network's underlying redundancy and strengthens the rationale for rational, multi-node combination therapy. Future work integrating single-cell epigenomic and epitranscriptomic profiling into clinical risk stratification is likely to determine how effectively this reframing can be translated into durable, precision-guided remission for patients with AML.

Author Contributions

G.B.Y.S. contributed to the conception and design of the review, literature search, analysis and synthesis of the relevant evidence, development of the conceptual framework, and drafting of the manuscript. R.V. contributed to the literature search, interpretation of evidence concerning chromatin architecture, cohesin dysfunction, and oncofusion proteins, and critical revision of the manuscript. K.A. contributed to the analysis and interpretation of epigenetic and metabolic mechanisms in acute myeloid leukaemia and critically revised the manuscript. R.P. contributed to the evaluation of 3D genomic architecture, transcriptional regulation, and therapeutic vulnerabilities and critically revised the manuscript. M.J.M. contributed to the interpretation of translational and therapeutic evidence and critical revision of the manuscript for important intellectual content. All authors reviewed and approved the final version of the manuscript and agreed to be accountable for all aspects of the work.

Acknowledgements

The authors would like to acknowledge the Faculty of Medicine, Universitas Indonesia; the Department of Iranian Traditional Medicine, School of Medicine, Bushehr University of Medical Sciences; the Department of Traditional Medicine, School of Persian Medicine, Shahed University; and the Department of Hematology, Faculty of Allied Medicine, Bushehr University of Medical Sciences, for their academic and institutional support. The authors also acknowledge the researchers whose published studies contributed to the scientific foundation of this review.

References