Abstract
Background:
The clinical phenomenon of metabolic memory, in which diabetic complications may progress despite later glycemic improvement, lacks a complete explanation. Persistent oxidative stress, advanced glycation, mitochondrial dysfunction, and epigenetic remodeling are established contributors, but how these processes intersect with regulated cell death programs is still unresolved.
Main Body:
This review proposes and critically appraises the hypothesis that epigenetic dysregulation increases the susceptibility of diabetic tissues to ferroptosis. We employ the term “epigenetic-ferroptosis link” as an operational term denoting a potential mechanistic connection between hyperglycemia-associated chromatin/RNA-regulatory alterations and shifts in ferroptosis vulnerability. A key methodological tenet of our analysis is the clear demarcation between direct evidence from diabetic models and supportive mechanistic insights drawn from other pathological contexts, ensuring a rigorous assessment of the hypothesis against the available data.
Conclusion:
Ferroptosis is one execution pathway among several that can contribute to diabetic tissue injury. This framing highlights the critical importance of organ- and cell-type specificity, the experimental rigor required to conclusively demonstrate ferroptosis involvement, the inherent limitations of extracellular-vesicle biomarkers and epigenetic-editing approaches, and the validation steps that remain necessary before any clinical translation can be contemplated. Antioxid. Redox Signal. 00, 000–000.
Introduction
Diabetes mellitus imposes a substantial global burden through devastating microvascular complications, such as retinopathy, nephropathy, and neuropathy, which lead to blindness, renal failure, and nerve dysfunction. A central and enduring clinical paradox is the phenomenon of “metabolic memory”: complications continue to progress even after glycemic control is achieved, indicating that a transient metabolic insult leaves a durable, pathological legacy inscribed within tissues (Chen et al., 2024b; Kern et al., 2022; Singh et al., 2020).
Researchers increasingly attribute this persistence to epigenetic reprogramming. Heritable changes in DNA methylation, histone modifications, and noncoding RNA networks promote and sustain pro-inflammatory, pro-fibrotic, and pro-oxidative transcriptional programs, encoding the memory of past hyperglycemia (Cannito et al., 2025; Chen et al., 2024; Yamunadevi et al., 2021). For instance, such epigenetic dysregulation in endothelial cells and neurons perpetuates dysfunction, mechanistically linking early metabolic stress to late complications like retinopathy and neuropathy (Kern et al., 2022; Sanusi et al., 2025).
Concurrently, ferroptosis, an iron-dependent form of regulated cell death (RCD) marked by lipid peroxide accumulation, contributes to diabetic tissue injury in a cell-type-dependent manner. Its relevance depends on cell type, iron availability, membrane lipid composition, GPX4/system xc- defense capacity, and the extent to which ferroptosis-specific rescue has been demonstrated in a given model. Current models often discuss epigenetic memory and ferroptosis in parallel, but they rarely test whether a lasting regulatory change actually changes the ferroptosis threshold in a diabetic cell. Without that step, the field cannot tell whether the two processes simply coexist or one helps set up the other.
Innovation
This review offers three specific contributions. First, it synthesizes the scattered literature on metabolic memory, epigenetic regulation, and ferroptosis into a set of experimentally addressable questions. Second, it clearly distinguishes diabetic evidence from indirect support, particularly concerning feedback loops and propagation via extracellular vesicles. Third, it establishes practical standards for future studies, including criteria for assessing ferroptosis, approaches for modeling metabolic memory, and guidelines for interpreting therapeutic or biomarker claims without overextrapolating from the current evidence base.
Here, we review the diabetic and nondiabetic literature with that distinction in mind. Prior hyperglycemia may leave durable chromatin or RNA-regulatory marks that weaken antioxidant defense, alter iron handling, or change membrane lipid composition. In some cells, those changes could lower the threshold for iron-dependent lipid peroxidation; in others, apoptosis, senescence, inflammation, fibrosis, or mitochondrial dysfunction may be the dominant outcome. The aim is therefore practical: to define what would count as convincing evidence, to identify where diabetic models already provide support, and to mark the places where the argument still depends on inference.
The unresolved question is not whether oxidative stress, epigenetic changes, or ferroptosis occur in diabetes; each has substantial support. The specific gap is whether persistent epigenetic or RNA-regulatory alterations causally tune ferroptosis sensitivity in diabetic target cells, whether this occurs before overt injury, and whether reversal of these marks changes disease trajectory. We treat the epigenetic-ferroptosis axis as a testable framework.
Metabolic memory refers to the persistent biological effect of previous hyperglycemic exposure after subsequent glucose improvement. Epigenetic memory refers more narrowly to durable chromatin, DNA methylation, histone modification, chromatin accessibility, noncoding RNA, or epitranscriptomic changes that can maintain altered gene regulation. Ferroptosis priming refers to a state in which cells have a lower threshold for iron-dependent phospholipid peroxidation because defenses such as system xc-, GPX4, glutathione (GSH) synthesis, NRF2 signaling, FSP1-CoQ10, or mitochondrial DHODH are insufficient relative to iron load and oxidizable lipid substrates.
Evidence-Graded Framework: Linking Metabolic Memory, Epigenetic Regulation, and Ferroptosis Susceptibility
The clinical enigma of metabolic memory and a critical theoretical gap
The enduring legacy of early hyperglycemia, termed metabolic memory, poses a central paradox in diabetes care. Landmark trials (DCCT/EDIC, UKPDS) provided strong evidence that initial glycemic exposure imprints a lasting risk for vascular complications, which persists despite subsequent glucose normalization (Lachin and Nathan, 2021; Miller and Orchard, 2020). This clinical reality underscores a fundamental dilemma: intensive control instituted later often fails to prevent damage, highlighting the irreversibility encoded within tissues and the critical need for early intervention.
Prevailing theories attribute this memory to mitochondrial reactive oxygen species (ROS) bursts and advanced glycation end-product (AGE) accumulation. These drivers are thought to induce sustained oxidative stress and inflammation, leading to persistent epigenetic reprogramming, such as DNA methylation, histone modifications, and chromatin remodeling, that locks cells into a pathological state (Chen et al., 2024; Chen and Natarajan, 2022). For instance, hyperglycemia-induced DNA hypermethylation in endothelial cells perpetuates dysfunction long after glucose levels are corrected (Vasishta et al., 2026), while histone marks and noncoding RNAs maintain pro-inflammatory gene expression (Kern et al., 2022; Lü et al., 2022).
The paradigm that links ROS and AGEs to epigenetic changes harbors a critical gap: it explains the initiation of a dysfunctional cellular state but not the downstream cell-fate decisions, including survival, senescence, or death, that culminate in tissue loss. While these factors alter the epigenetic landscape, the downstream mechanisms determining cell fate remain opaque. The role of RCD, specifically ferroptosis (an iron-dependent, lipid peroxidation-driven process), remains less systematically integrated into models of metabolic memory. Current models often treat ferroptosis as a separate, downstream event, disconnected from the upstream epigenetic reprogramming of metabolic memory. By doing so, they overlook that epigenetic modifications may both record and preconfigure the cellular sensitivity threshold to ferroptosis, linking past insult to a primed cell-death pathway.
Recent studies support parts of this link, particularly epigenetic regulation of ferroptosis-associated genes such as GPX4 and SLC7A11. Many studies remain correlative or use nondiabetic models. We treat these findings as a rationale for structured evaluation, not as proof that ferroptosis is the universal terminal mechanism of metabolic memory.
ROS/AGE-centric theories explain major initiating and sustaining components of diabetic tissue injury but do not fully resolve how prior metabolic exposure changes later cell-fate thresholds. Evaluating ferroptosis susceptibility within an epigenetic-memory framework identifies when iron-dependent lipid peroxidation contributes to tissue loss and when other regulated death or dysfunction programs predominate.
The proposed epigenetic-ferroptosis axis as a testable, evidence-graded model
We propose a restricted formulation of the model: in some diabetic tissues, metabolic memory may include a sustained regulatory state that lowers the threshold for ferroptosis. This state is not assumed to be present in all complications or cell types. Its existence must be demonstrated by persistent regulatory marks, altered ferroptosis-defense or lipid/iron-handling networks, and functional rescue by targeted interventions.
Correlation is not enough for this question. A stronger study would need to show three things: the regulatory change appears before lipid peroxidation or tissue injury, the change is sufficient to increase ferroptosis sensitivity, and reversing the change or blocking ferroptosis reduces the injury while preserving the memory-like design. Recent primary studies provide examples of these validation components, including glucose-induced metabolic-memory mechanisms in endothelial cells, epigenetic regulation of SLC7A11/GPX4/TfR1-related ferroptosis nodes, and ferroptosis-directed rescue in diabetic or diabetes-related models (Vasishta et al., 2026; Wang et al., 2024a; Wei et al., 2025; Zhang et al., 2025a).
Temporal precedence: Epigenetic or epitranscriptomic changes at ferroptosis-relevant loci should be detectable during early or memory-like hyperglycemic exposure before overt lipid peroxidation, cell loss, fibrosis, or neurovascular degeneration. Sufficiency: Experimental induction of a defined mark, such as methylation-dependent GPX4 repression or PRC2/EZH2-linked SLC7A11 repression, should lower the ferroptosis threshold in otherwise nondiabetic cells or tissues. Necessity: Reversal of the upstream regulatory change, or selective blockade of ferroptosis using accepted criteria, should reduce diabetic tissue injury in models that otherwise preserve the metabolic-memory phenotype.
With these criteria, the model becomes testable (Table 1). The better-supported direction runs from metabolic or epigenetic stress toward ferroptosis sensitivity. The reverse direction, in which ferroptosis products reshape the epigenome in a stable and locus-specific way, remains much less secure in diabetic tissues.
Working Criteria Used to Grade Evidence for the Proposed Epigenetic-Ferroptosis Framework
This table is used as an explicit screening rule: a study is not described as strong evidence for ferroptosis unless it demonstrates lipid peroxidation together with iron dependence and ferroptosis-directed rescue; similarly, a study is not treated as metabolic-memory evidence unless its design separates prior hyperglycemic exposure from later normoglycemic or longitudinal effects. Evidence is graded as direct when derived from diabetic models or human diabetic tissues, supportive when imported from cancer, ischemia, neurodegeneration, or other inflammatory models to illustrate mechanism, and hypothetical when it remains to be tested.
DNMT, DNA methyltransferase; TET, ten-eleven translocation.
To keep the evidence transparent, the remainder of the review uses the following labels. Direct diabetic evidence refers to data obtained in diabetic cells, animal models of diabetes, or human diabetic tissues, ideally with a memory-compatible design. Supportive or indirect evidence refers to mechanistically plausible findings from cancer, ischemia-reperfusion, neurodegeneration, or other inflammatory models that illustrate how an epigenetic-ferroptosis axis can operate but that do not by themselves prove it operates in diabetic complications. Hypothesis denotes predictions that remain to be tested. Similarly, we do not assume the axis applies uniformly to diabetic nephropathy, retinopathy, neuropathy, cardiomyopathy, osteoporosis, and vascular injury. These organs differ in iron handling, lipid composition, antioxidant capacity, cell turnover, and exposure to metabolic stress; the framework should therefore be applied in a modular, organ-specific manner (Table 2).
Organ-Specific Evidence Map for the Proposed Epigenetic-Ferroptosis Framework
DKD, diabetic kidney disease.
Forward component (epigenetic/RNA regulation to ferroptosis susceptibility): Hyperglycemia-associated regulatory alterations may rewire ferroptosis-relevant networks by repressing protective systems such as SLC7A11, GPX4, NRF2-linked antioxidant genes, FSP1-CoQ10, or DHODH, and/or by increasing iron uptake and oxidizable lipid substrates through regulators such as TfR1, ACSL4, LPCAT3, ALOX enzymes, Elongation of very-long-chain fatty acids (ELOVL)/Fatty acid desaturase (FADS) enzymes, or m6A-linked transcript control.
Putative feedback component (ferroptosis-associated products to epigenetic remodeling): Lipid aldehydes, labile iron, lactate, inflammatory mediators, and extracellular-vesicle (EV) cargo could influence chromatin regulators or RNA-regulatory networks. In diabetic complications, this feedback remains incompletely proven and should be investigated as a high-priority hypothesis rather than presented as established causality.
This narrower view explains why a tissue may remain vulnerable after glycemia improves while leaving room for other injury programs to dominate in particular organs or stages (Fig. 1).
Evidence-graded framework linking metabolic memory, epigenetic regulation, and ferroptosis susceptibility in diabetic complications.
BOX 1. Evidence-graded interpretation of the proposed epigenetic-ferroptosis framework
Better-supported direction: Hyperglycemic or glucolipotoxic stress can alter chromatin, noncoding RNA, and epitranscriptomic regulation of antioxidant defense, iron handling, and lipid-remodeling genes. In selected diabetic models, these changes are associated with increased ferroptosis susceptibility.
Hypothesis requiring validation: Ferroptosis-associated metabolites or EV cargo may influence epigenetic states, but direct locus-specific evidence for a stable feedback loop in diabetic tissues remains limited.
Therapeutic testing should determine whether combined targeting of regulatory priming and ferroptosis execution provides benefit beyond standard metabolic, hemodynamic, anti-inflammatory, and antioxidant approaches.
Core Mechanisms of the Axis: Definition and Basic Principles
Metabolic memory: An epigenetic primer for ferroptosis
Metabolic memory—the persistence of cellular dysfunction after transient hyperglycemia—is likely sustained by multiple interacting processes, including AGE accumulation, mitochondrial dysfunction, chronic low-grade inflammation, vascular remodeling, and epigenetic reprogramming, rather than initiated exclusively by any single mechanism. Within this broader landscape, we examine the hypothesis that epigenetic and RNA-regulatory changes may, in selected diabetic tissues, lower the threshold for ferroptosis. We therefore treat the epigenetic-ferroptosis axis as a proposed, testable component of metabolic memory, not as an established universal mechanism. Hyperglycemia-induced mitochondrial dysfunction alters the flux of key metabolite cofactors. A pivotal shift is the decreased α-ketoglutarate (α-KG)-to-succinate ratio, which inhibits α-KG-dependent dioxygenases (e.g., ten-eleven translocation [TET], JmjC-domain histone demethylases), constraining epigenetic erasure (Che et al., 2026).
A critical distinction runs through this entire review. Metabolic memory specifically requires that a biological effect persists after the initiating hyperglycemic exposure is removed or corrected. Studies that use acute high glucose, sustained high glucose, or glucolipotoxicity without a normalization or washout phase inform stress responsiveness or chronic diabetic injury, but they do not by themselves demonstrate memory. Conversely, transient hyperglycemia followed by normoglycemia, oscillating glucose designs, or long-term cohorts that separate prior glycemic exposure from current glycemia are needed to claim memory. In Table 1 and throughout the article we classify evidence accordingly: continuous high-glucose findings are interpreted as supportive or associative, whereas true memory designs are required for stronger inference. This distinction is essential because the clinical phenomenon we seek to explain is persistent injury despite glycemic correction.
This metabolic constraint extends beyond TCA cycle intermediates via three key metabolite-chromatin bridges that define the “priming” phase:
The One-Carbon Metabolism and S-adenosylmethionine/S-adenosylhomocysteine Ratio: Hyperglycemia disrupts the methionine cycle, elevating the SAM-to- SAH ratio. This enhanced SAM/SAH status promotes DNA methyltransferase (DNMT) activity and may contribute to hypermethylation of cytoprotective gene promoters (e.g., GPX4, NFE2L2) (Ciminera et al., 2021; Herrmann et al., 2005; Ntambi et al., 2023). Acetyl-CoA Availability and Histone Acetylation: Cytosolic acetyl-CoA, generated by Adenosine triphosphate-citrate lyase (ACLY), is essential for histone acetyltransferases (HATs). Diabetic metabolic inflexibility alters this pool, dysregulating site-specific histone acetylation. This can lead to repressed chromatin states at antioxidant genes (e.g., SOD2), while in some contexts, increased ACLY may drive activating marks (e.g., H3K27ac) on pro-ferroptotic genes (Wellen et al., 2009). The NAD+/Sirtuin Redox Sensor: The diabetic “pseudohypoxic” state, characterized by a high NADH/NAD+ ratio, inhibits NAD+-dependent sirtuins (SIRT1/3). SIRT1 normally deacetylates and stabilizes NRF2. Its impairment leads to NRF2 degradation, thereby suppressing the transcription of ferroptosis defense genes like SLC7A11 and GPX4 (Liang et al., 2025; Singh et al., 2018; Xie et al., 2022), creating a redox-epigenetic silencing loop.
This concerted epigenetic reprogramming reshapes ferroptosis-relevant regulatory networks. Defense genes (e.g., GPX4, SLC7A11) accumulate repressive marks (DNA hypermethylation, H3K9me3, H3K27me3) (Wang et al., 2024; Zhang et al., 2025), while pro-ferroptotic genes (e.g., ACSL4, TFR1) may gain activating modifications (H3K4me3, H3K27ac), establishing a primed, pro-ferroptotic transcriptomic state.
The response is also cell-type specific. Podocytes, endothelial cells, glia, neurons, and Schwann cells begin from different chromatin states and use different stress-response programs, so the same glucose exposure need not produce the same ferroptosis phenotype in every tissue.
Metabolic memory is a composite effect of TCA cycle blockage, methyl-donor imbalance, and redox-dependent deacetylase failure. The resulting epigenetic inertia, reinforced by feedback between metabolites and chromatin regulators, links metabolic dysregulation to altered cell-fate susceptibility, explaining disease chronicity. The metabolic memory-epigenetic-ferroptosis axis offers a multipronged therapeutic frontier, with nodes targeting metabolite replenishment, epigenetic editing, and ferroptosis inhibition to interrupt this feed-forward process (Table 3).
Evidence Hierarchy for the Epigenetic-Ferroptosis Framework in Diabetic Complications
Detailed analysis: Epigenetic targeting of the ferroptosis network
The durable pro-ferroptotic state characteristic of metabolic memory is proposed to be executed, in part, through epigenetic modifications that target core regulatory nodes of the ferroptosis pathway. We emphasize that much of the evidence discussed below comes from diabetic cell or animal models, whereas mechanistic details (e.g., specific histone-modifying enzymes, m6A writers/readers, or lipid-remodeling pathways) are sometimes inferred from cancer, ischemia-reperfusion, or other nondiabetic inflammatory models. In those instances, the diabetic relevance is plausible but not proven, and we label them as supportive or inferential rather than direct evidence.
The regulatory architecture of ferroptosis susceptibility extends beyond DNA methylation and canonical histone marks. m6A-dependent epitranscriptomic control can stabilize or destabilize transcripts encoding iron import, lipid remodeling, and antioxidant defense proteins; for example, METTL3/YTHDF-linked regulation of TfR1 has been implicated in diabetic kidney disease (DKD). NRF2 links redox state to chromatin-regulated antioxidant capacity because its activity is shaped by acetylation, methylation, KEAP1-dependent degradation, and promoter/enhancer context at target genes, including SLC7A11, GCLC, NQO1, and GPX-related networks. Lipid remodeling also involves ACSL4 together with LPCAT3, ALOX enzymes, ELOVL family members, FADS enzymes, and CoQ10/DHODH/FSP1-dependent suppressor systems, which may vary by tissue. To make the evidence grade explicit, findings obtained in diabetic cells, animal models, or human diabetic tissues are treated as direct diabetic evidence; mechanistic details inferred from cancer, ischemia-reperfusion, or other nondiabetic inflammatory models are labeled as supportive evidence and do not by themselves prove diabetic relevance.
Ferroptosis diagnosis: a stricter standard. Demonstrating ferroptosis in diabetic tissues requires more than altered GPX4, SLC7A11, ACSL4, malondialdehyde (MDA), 4-HNE, ROS, or iron levels. These are supportive but, taken individually, insufficient. We encourage future studies to include (i) direct lipid-peroxidation assays (e.g., BODIPY-C11, LC-MS/MS oxidized phospholipids, MDA/4-HNE only as part of a panel), (ii) iron dependence (chelation or genetic iron-trafficking manipulation), (iii) rescue by validated ferroptosis inhibitors (e.g., ferrostatin-1, liproxstatin-1) or genetic restoration of GPX4/SLC7A11, and (iv) exclusion of dominant apoptosis, necroptosis, pyroptosis, and senescence pathways where feasible. Studies that do not meet these criteria should be interpreted as suggestive, and we have downgraded their evidence accordingly in Tables 1 and 4.
Regulated Cell-Death Pathways in Diabetic Complications and Criteria for Distinguishing Ferroptosis from Apoptosis, Pyroptosis, Necroptosis, and Senescence
Epigenetic causality: from correlation to necessity and sufficiency. DNA methylation, histone modifications, noncoding RNAs, and m6A changes may correlate tightly with diabetic injury, but causality requires targeted manipulation of the regulatory mark itself. Ideal evidence includes targeted reversal (e.g., dCas9-TET/demethylase or dCas9-DNMT/methylase, locus-specific histone-modifier recruitment, or antagomir/mimic rescue) with concordant changes in ferroptosis-regulator expression, lipid peroxidation, and cell death. We now state explicitly that many currently cited studies report associations or global epigenetic-drug effects and therefore do not yet establish necessity or sufficiency. We have revised Table 5 and the main text to separate correlation, global pharmacologic modulation, and locus-specific causal evidence.
Epigenetic and RNA-Regulatory Modifiers of Ferroptosis Susceptibility in Diabetic Complications
PRC2, polycomb repressive complex 2.
Silencing of Cytoprotective Machinery:
SLC7A11 (Cystine/Glutamate Antiporter): The expression of SLC7A11, essential for cystine uptake and GSH synthesis, appears to be suppressed in selected diabetic models via multiple epigenetic layers. This includes DNMT3B-mediated DNA hypermethylation of its promoter, which inhibits transcription. Concurrently, the Polycomb repressive complex 2 (PRC2) catalyzes the deposition of the repressive histone mark H3K27me3 at the SLC7A11 locus, further consolidating its silencing in a durable manner. Noncoding RNAs, such as lncRNA H19, contribute to this repression by sequestering or guiding epigenetic complexes. GPX4: As the central enzyme that reduces lipid hydroperoxides, GPX4 is a critical epigenetic target. Its promoter undergoes hypermethylation, likely mediated by DNMTs, leading to transcriptional downregulation and impaired antioxidant defense at the final step of the ferroptosis pathway.
Activation of Pro-Ferroptotic Drivers:
Acyl-CoA Synthetase Long-Chain Family Member 4 (ACSL4): This enzyme catalyzes the esterification of polyunsaturated fatty acids (PUFAs) into membrane phospholipids, providing substrates for lipid peroxidation. Hyperglycemia promotes an activating epigenetic landscape at the ACSL4 gene, including increases in H3K4me1 and H3K27ac, which are associated with enhanced enhancer and promoter activity, respectively. Studies implicate the histone methyltransferase SETD7, which catalyzes H3K4me1, in this process. Transferrin Receptor 1 (TfR1): To fuel the iron-dependent peroxidation process, cellular iron uptake is epigenetically enhanced. The TfR1 gene acquires activating histone modifications such as H3K4me3, facilitating its increased expression and contributing to intracellular iron overload.
This coordinated epigenetic rewiring—repressing defense and promoting attack—creates a transcriptomic state primed for ferroptosis. The stability of these DNA and histone modifications may allow this pro-death regulatory state to persist, encoding the “memory” of past hyperglycemic stress (Fig. 2).
Evidence-graded, proposed epigenetic programming of ferroptosis susceptibility in diabetic complications. Putative schematic of the core molecular architecture. Central ferroptosis regulators (SLC7A11, GPX4, ACSL4, TfR1) are proposed to be regulated by hyperglycemia-associated epigenetic modifications (red: repression; green: activation), establishing a putative pro-ferroptotic state that requires validation in organ-specific diabetic models.
Ferroptosis susceptibility as one regulated cell-death output of epigenetic programming
The epigenetic reprogramming detailed above establishes a cell in a long-term “primed” state for iron-dependent death. It is characterized by constrained GSH synthesis, suppressed expression of the key antioxidant enzyme GPX4, and elevated membrane incorporation of peroxidation-prone PUFAs via upregulated ACSL4 (Wei et al., 2025). This biochemical remodeling creates a cellular environment with diminished antioxidant buffering capacity and an enriched substrate pool for lipid peroxidation, effectively lowering the threshold for ferroptotic execution.
A key consequence of this priming is increased sensitization of cells to subsequent, otherwise sublethal, oxidative insults prevalent in the diabetic milieu, such as mild hypoxia or inflammatory cytokines (Mudaliar et al., 2021). When such a “second hit” occurs, the already weakened GPX4/GSH defense system may become insufficient. This can promote iron-catalyzed Fenton chemistry, leading to an irreversible burst of lipid peroxidation that disrupts membrane integrity and executes cell death. Epigenetic programming does not cause death but decisively lowers the barrier for it, explaining how a transient hyperglycemic insult can predispose tissues to acute damage long after glucose normalization.
Epigenetic priming also has to be interpreted within a wider stress network. Transcript stability, protein turnover, mitochondrial function, iron trafficking, inflammatory cytokines, and the local lipid pool can all change whether a primed cell survives, senesces, or crosses into ferroptosis.
Beyond ACSL4, lipid remodeling and parallel ferroptosis-defense systems should be evaluated explicitly. LPCAT3 controls incorporation of polyunsaturated fatty acyl chains into phospholipids, whereas ALOX enzymes can oxidize susceptible phospholipids once antioxidant defenses fail. ELOVL and FADS enzymes shape the pool of elongation and desaturation products that determine membrane peroxidizability. In parallel, FSP1-CoQ10 and mitochondrial DHODH-CoQ defense axes can suppress lipid peroxidation independently of GPX4 in some contexts. A diabetic study that measures only ACSL4 or GPX4 should be interpreted as partial ferroptosis evidence unless lipid-substrate remodeling and these compensatory defense systems are also considered.
A state transition model and the epigenetic ferroptosis signature
Integrating the above mechanisms, we propose a dynamic, three-phase state transition model that conceptualizes the progression of diabetic complications through the epigenetic-ferroptosis axis. This model captures the evolution from a reversible metabolic insult to an autonomous, self-propagating pathological state.
Phase 1: Priming (Reversible). Transient hyperglycemia induces early, labile chromatin changes, such as increased H3K4me1 at enhancers of pro-ferroptotic genes (e.g., ACSL4). These alterations are theoretically reversible upon timely glycemic normalization. Prediabetes or early hyperglycemic exposure is the clinical correlate. Phase 2: Sensitization (Locked-In). Sustained metabolic stress catalyzes the conversion of priming marks into stable, repressive modifications. This includes DNA hypermethylation and H3K27me3 deposition at cytoprotective loci (e.g., GPX4, SLC7A11), while activating marks consolidate at pro-death genes. This establishes a durable “pro-ferroptotic epigenome,” which critically lowers the ferroptosis threshold. The clinical correlate is the establishment of “metabolic memory” and early complications (the “legacy effect”). Phase 3: Execution and Propagation. Cells in the sensitized state are primed for ferroptosis. A secondary insult (e.g., localized inflammation or hypoxia) acts as a “trigger,” initiating the lethal lipid peroxidation cascade. The execution of ferroptosis itself generates two key outputs: (i) reactive metabolites (e.g., lipid aldehydes) that can further remodel the epigenome, and (ii) EVs carrying pro-ferroptotic signals. These outputs propagate the pro-ferroptotic priming signal to neighboring cells, thereby extending the “field” of cellular susceptibility and potentially contributing to the spatial propagation of susceptibility that amplifies tissue damage (Fig. 3).
Proposed three-phase validation framework: regulatory priming, susceptibility, and execution/propagation. Each phase should be tested with time-resolved epigenetic, lipid-peroxidation, and functional rescue experiments.
To operationalize the detection of the critical Sensitization phase, we define the epigenetic ferroptosis signature (EFS) as its molecular hallmark. The EFS comprises: (i) DNA methylation: Hypermethylation of the GPX4 and SLC7A11 promoters; (ii) Histone modifications: Enrichment of repressive H3K27me3 at defense gene loci (e.g., SLC7A11) coupled with activating H3K4me3/H3K27ac at drivers like ACSL4; (iii) Noncoding RNA: Elevated lncRNA H19 and suppressed miR-29b. The detection of an EFS signifies that a tissue is in the “Sensitization” phase—epigenetically primed for ferroptosis—and thus supports testing interventions aimed at modifying persistent regulatory states rather than mere antioxidant rescue.
BOX 2. The epigenetic ferroptosis signature
Definition: A composite molecular profile indicative of the “Sensitization” phase, characterized by promoter hypermethylation of key cytoprotective genes (e.g., GPX4, SLC7A11); enrichment of repressive histone marks (e.g., H3K27me3) at their loci; and dysregulation of specific noncoding RNAs (e.g., elevated lncRNA H19, suppressed miR-29b).
Function: The EFS does not directly kill cells. Instead, it stably reprograms the cellular phenotype by silencing the ferroptosis-defense system, thereby critically lowering the threshold for lethal lipid peroxidation and “locking in” a state of heightened susceptibility.
Clinical implication: The EFS is a research-stage candidate signature, not a validated biomarker. Before clinical use, it must show organ specificity, reproducibility across cohorts and platforms, temporal precedence over tissue injury, and added predictive value beyond HbA1c, albuminuria, retinal imaging, neuropathy testing, and conventional inflammatory or oxidative-stress markers. Clinical implication: The EFS is a research-stage candidate signature and currently a hypothesis, not a validated biomarker. Before clinical use, it must be evaluated against existing clinical predictors, including HbA1c history, albuminuria, eGFR, retinal imaging, neuropathy scores, blood pressure, lipid status, and diabetes duration. Prospective studies must demonstrate incremental predictive value—improved discrimination, calibration, or reclassification—over and above these established predictors. Without such benchmarking, biomarker claims risk being overstated.
Putative feedback from ferroptosis-associated products to epigenetic remodeling
Ferroptosis-associated lipid peroxidation products may contribute to secondary regulatory changes, but direct evidence for a self-amplifying epigenetic feedback loop in diabetic tissues remains limited and is currently one of the weaker parts of the framework. Candidate mediators include lipid aldehydes (e.g., 4-HNE, MDA), labile iron-dependent redox signaling, lactate-linked histone lactylation, inflammatory cytokines, and extracellular-vesicle cargo. While each is biologically plausible, stable locus-specific feedback—where a ferroptosis product reproducibly alters a defined chromatin or RNA-regulatory mark in diabetic target cells—has not yet been demonstrated. We therefore present this feedback route as a high-priority hypothesis rather than an established driver of metabolic memory.
Aldehyde modification of DNMTs and histone deacetylases (HDACs) can impair their function, derepressing genes involved in oxidative stress and iron metabolism. In DKD, while EZH2-mediated repression of SLC7A11 promotes ferroptosis, subsequent lipid peroxide accumulation may feedback to inhibit such repressive complexes, dynamically reshaping gene expression (Wang et al., 2024). Similarly, in diabetic osteoporosis, DNMT-mediated GPX4 promoter hypermethylation suppresses this key defense (Wei et al., 2025); lipid aldehydes from ongoing ferroptosis could then inhibit DNMT activity, creating epigenetic heterogeneity.
The feedback component should therefore be read as a hypothesis rather than as an established driver of irreversibility. It may help explain persistent injury in selected contexts, but direct diabetic evidence linking ferroptosis products to stable, locus-specific chromatin remodeling remains insufficient.
This putative feedback route is best treated as a mechanistic test bed. Scavenging reactive aldehydes, modulating lactylation, or blocking EV cargo transfer should be evaluated first for necessity and temporal ordering before being advanced as axis-disrupting therapy.
Boundary conditions: Contextualizing the axis within the cell death landscape
This review treats the epigenetic-ferroptosis link as one part of diabetic memory. Mitochondrial dysfunction, AGE/RAGE signaling, inflammation, fibrosis, vascular remodeling, senescence, apoptosis, pyroptosis, and necroptosis all remain part of the same clinical problem.
First, ferroptosis is not universally the initial or dominant cell-death driver. In early diabetic retinopathy (DR), for example, pericyte loss often reflects apoptosis and vascular stress, whereas ferroptosis-related injury may become more relevant in glial, neuronal, or photoreceptor contexts depending on iron load, lipid composition, and antioxidant reserve. This model therefore predicts context-dependent overlap rather than replacement of established mechanisms.
Second, not all hyperglycemia-induced epigenetic remodeling culminates in ferroptosis. A distinct form of “inflammatory memory,” such as H3K9me3-mediated sustained NF-κB activation in monocytes, drives chronic inflammation without triggering lethal lipid peroxidation. The epigenetic-ferroptosis axis represents a specific, deleterious subset of metabolic memory dedicated to cell loss, separable from programs sustaining inflammation or hypertrophy.
Finally, ferroptosis can occur independently of epigenetic memory during overwhelming acute stress, including ischemia-reperfusion, toxin exposure, or severe GSH depletion. Such settings should not be used as direct proof of metabolic memory unless persistence after normalization of the initiating stress is demonstrated.
In practice, ferroptosis is likely to be dominant only in specific contexts: when labile iron is abundant, oxidizable membrane PUFAs are enriched, GPX4/system xc-/NRF2/FSP1/DHODH defenses are persistently repressed, and rescue by ferroptosis inhibitors attenuates injury. In other contexts, apoptosis, senescence, pyroptosis, necroptosis, fibrosis, endothelial dysfunction, or immune activation may be the primary driver. We therefore do not argue that ferroptosis replaces these pathways but rather that it may act as a parallel or secondary execution mechanism that becomes clinically important in selected cell types and stages.
Organ-Specific Decoding: Heterogeneous Driving of Axes in Microvascular Complications
Diabetic nephropathy: Divergent epigenetic “memory” and synergistic pathogenesis in podocytes and tubular cells
Before considering individual complications, it is essential to state that a single identical mechanism is unlikely to operate uniformly in kidney, retina, nerve, heart, and vasculature. The proposed axis should be interpreted as a modular framework: prior hyperglycemia may tune chromatin or RNA regulation, but the ferroptosis-relevant output depends on cell-specific iron handling, lipid composition, mitochondrial load, antioxidant reserve, local inflammation, and repair capacity. The kidney may be dominated by tubular iron handling and podocyte vulnerability, the retina by neurovascular and glial lipid-peroxidation stress, and peripheral nerves by axon-Schwann-cell metabolic coupling.
DN progression is driven by cell-type-specific epigenetic reprogramming that encodes a durable “memory” of hyperglycemic insult. Podocytes, as terminally differentiated cells, exhibit a stable epigenetic landscape characterized by persistent alterations in chromatin accessibility and DNA methylation. This entrenched memory predisposes them to detachment and loss, a key event in glomerular injury, and persists despite subsequent glycemic control, contributing to the irreversibility of damage. In contrast, renal tubular epithelial cells (TECs) display epigenetic plasticity, undergoing dynamic reprogramming toward a pro-fibrotic phenotype that promotes epithelial-to-mesenchymal transition (EMT) and extracellular matrix deposition, thereby driving tubulointerstitial fibrosis.
Mechanistically, the diabetic renal microenvironment—marked by metabolic stress, hypoxia, and oxidative stress—fosters a pro-ferroptoticepigenome. Methylglyoxal a reactive dicarbonyl elevated in diabetes, can induce oxidative stress and ferroptosis in renal tubular epithelial cells, providing a candidate mechanistic link between metabolic stress and tubular injury (Zhang et al., 2025c). This sensitizes both cell types to iron-dependent death. Injured podocytes release paracrine mediators (e.g., Angptl4) that are taken up by TECs, delivering epigenetic regulators such as microRNAs (e.g., miR-23a-3p) that reprogram tubular chromatin. This intercellular crosstalk establishes a feed-forward loop: podocyte injury propagates a “memory” signal to TECs, amplifying fibrogenic responses and consolidating a synergistic axis of glomerular and tubular damage.
Integrated bioinformatics of DKD patient data identifies hub genes (e.g., JUN, PTGDS, SLC22A17) linking epigenetic dysregulation to ferroptosis and inflammation. Immune-cell infiltration further modulates this microenvironment. Therapeutic strategies targeting the epigenetic-ferroptosis axis, via iron chelation, antioxidant pathways, or interception of microRNA-mediated communication, may have experimental value for mitigating podocyte loss and tubular fibrosis. Decoding the distinct yet interconnected epigenetic memories of podocytes and TECs provides a refined framework for precision intervention in DN.
Diabetic retinopathy: Glial, vascular, and neuronal susceptibility within the neurovascular unit
DR is a neurovascular disorder in which hyperglycemia disrupts coordinated interactions among neurons, glia, pericytes, and vascular endothelial cells, leading to neurodegeneration and vasoregression (Antonetti, 2021; Ji et al., 2021; Kolibabka et al., 2023). Retinal models show ferroptosis-related lipid peroxidation, but its timing and dominance relative to apoptosis, inflammation, and vascular dysfunction remain context-dependent.
Hyperglycemia may alter Müller glial stress responses and ferroptosis-related defenses, but direct evidence that Müller-cell ferroptosis spreads as a self-sustaining epigenetic relay in DR remains incomplete. EVs carrying oxidized lipids, noncoding RNAs, or inflammatory signals are biologically plausible mediators; however, demonstrating that they propagate ferroptosis susceptibility requires more than altered cargo profiles. Necessary experiments include EV depletion or neutralization, cargo rescue/replacement, cell-type tracing of EV donor–recipient pairs, and functional transfer of ferroptosis vulnerability in recipient cells using the strict diagnostic criteria outlined above. Until such data are available, EV-mediated propagation should be treated as a working hypothesis.
This intercellular dialogue is amplified by activated microglia and astrocytes, which release pro-inflammatory cytokines (TNF-α, IL-1β) and vascular endothelial growth factor (VEGF), further destabilizing the neurovascular unit (NVU) and potentiating ferroptotic pathways (Bai et al., 2025; Srivastava et al., 2026; Zhang et al., 2026). Single-cell transcriptomics reveals heterogeneous Müller cell subpopulations (e.g., Aqp4hi vs. Aqp4lo) with specialized roles, whose differential responses to diabetic stress may be fine-tuned by EV-carried epigenetic regulators (Zhang et al., 2023). Single-cell RNA-sequencing analyses furtherlink ferroptosis to diabetes-associated cognitive decline,identifying cell-type-specificsusceptibility programs (Zhang et al., 2025b).
Thus, a testable model emerges: glial stress, EV-mediated communication, endothelial dysfunction, and neuronal lipid-peroxidation injury may interact within the diabetic NVU. We consider this as a hypothesis for spatial propagation rather than an established spreading process (Fig. 4). The EV-mediated propagation component, in particular, should be treated as a hypothesis until EV depletion, cargo rescue, donor–recipient tracing, and functional transfer experiments are performed in diabetic models.
Organ-specific versions of the proposed framework in kidney and retina. (
Translational redox and neuroinflammatory context. Beyond the diabetic retina and nerve, recent translational work underscores how NADPH oxidase 2 (NOX2)-driven oxidative stress and neuroinflammation converge to produce neuronal injury and cognitive impairment and how neuroprotection can be achieved by targeting redox pathways. Singh and colleagues showed that selective NOX2 inhibition after status epilepticus attenuates epileptogenesis and cognitive impairment in a sex-dependent manner (Singh et al., 2025a) and that NOX2 targeting reduces seizure susceptibility, oxidative stress, and neuroinflammation in a chemically induced seizure model (Singh et al., 2025b). In parallel, a thioredoxin-mimetic peptide was reported to attenuate epilepsy progression and neurocognitive deficits (Singh et al., 2026). These studies are not diabetic models, but they illustrate a broader principle relevant to diabetic complications: oxidative stress and neuroinflammation are shared proximal drivers of neuronal dysfunction, and inhibiting redox signaling can produce measurable cognitive protection. We therefore discuss them as mechanistic support that strengthens the rationale for testing NOX2- and thioredoxin-centered redox interventions in diabetic neuropathy and cognitive impairment (Wang et al., 2024c), while clearly noting that their direct relevance to diabetic tissues remains to be validated.
Diabetic neuropathy: Cell-type-specific epigenetic priming drives axonal and Schwann cell vulnerability
Diabetic neuropathy involves the progressive degeneration of peripheral nerves, characterized by distinct pathologies in sensory axons and myelinating Schwann cells. Their differential vulnerability reflects their biological roles: long axons demand robust energy and lipid homeostasis, and Schwann cells require precise lipid synthesis for myelin integrity. Hyperglycemia exploits these differences via cell-specific epigenetic reprogramming, converging on ferroptotic execution.
In sensory neurons, high glucose epigenetically represses key mitochondrial fusion genes, such as MFN2. This leads to mitochondrial fragmentation within axons, impairing bioenergetics and amplifying ROS production—a critical driver of iron-dependent lipid peroxidation and ferroptosis. The resulting oxidative damage disrupts axonal transport and integrity, directly contributing to sensory deficits.
Schwann cells undergo epigenetic upregulation of lipid metabolism genes, notably ACSL4. This enzyme promotes the incorporation of peroxidation-susceptible PUFAs into myelin phospholipids, altering myelin composition and rendering Schwann cells particularly susceptible to oxidative damage and ferroptotic death. Demyelination and loss of axonal support ensue.
These cell-specific epigenetic programs create a feed-forward, synergistic process. Axonal mitochondrial dysfunction and ROS emission exacerbate the pro-oxidative microenvironment, further stressing Schwann cells. Simultaneously, Schwann cell injury and demyelination deprive axons of metabolic and insulating support, accelerating degeneration. This bidirectional crosstalk, mediated by the epigenetic-ferroptosis axis, underpins the co-occurrence of axonal loss and demyelination in diabetic neuropathy.
Recent models, including those using human pluripotent stem cell-derived Schwann cells, support their selective vulnerability to glucotoxic stress and highlight nodal regions, specialized axon-glia interfaces, as hotspots of metabolic vulnerability. Therapeutic strategies that modulate this axis, such as agents targeting epigenetic regulators of mitochondrial dynamics or lipid metabolism, may inform strategies for preserving both neuronal and glial function, providing a mechanistic foundation for halting neuropathic progression (Table 2) (Majd et al., 2023; Malheiro et al., 2021; Pham et al., 2023).
Translational Medicine Perspective: New Strategies for Diagnosis and Treatment
Biomarkers: from static injury markers to dynamic readers of epigenetic memory
Clinicians see the problem as a lag between present glucose control and past tissue exposure. A useful biomarker would read that history. We describe three levels at which such a signal might be sought: molecular marks at ferroptosis-related loci, EV-associated epigenetic or RNA cargo, and tissue-level patterns of spread.
Molecular Memory: Stable and heritable epigenetic modifications at specific loci (e.g., GPX4 promoter hypermethylation, repressive histone marks at SLC7A11), representing the fundamental layer of information storage (Chen et al., 2025; Yamunadevi et al., 2021). Cellular Memory: The resulting global transcriptional state, characterized by the locked-in dysregulation of both the antioxidant defense and iron homeostasis networks, which collectively establish a cell-intrinsic pro-ferroptotic phenotype (Sanusi et al., 2025). Tissue Memory: The spatial propagation of susceptibility via paracrine signaling. This involves the transfer of molecular “memory” through EVs carrying epigenetic regulators, lipids, and noncoding RNAs, creating a “field defect” of pro-ferroptotic priming within an organ (Mao et al., 2025; Yin et al., 2026).
Conventional biomarkers of diabetic injury, such as plasma MDA, 4-HNE, or inflammatory cytokines, primarily reflect acute or ongoing stress and damage. They fail to distinguish transient oxidative bursts from the entrenched epigenetic reprogramming that underlies true metabolic memory. For instance, while albuminuria indicates glomerular damage, it does not report the historical epigenetic insult that predisposes the tissue to such damage (Goycheva et al., 2023; Raja et al., 2021; Wang et al., 2024b).
That gap has prompted interest in markers closer to the proposed biology, including DNA methylation, histone marks, noncoding RNAs, and EV cargo linked to ferroptosis-relevant genes. At present, these are research tools, not clinical tests.
Human biofluid evidence is insufficient to claim that an EFS predicts complication progression. Detectability, tissue origin, analytical sensitivity, pre-analytical handling, and confounding by systemic inflammation remain major obstacles. The EFS concept is therefore proposed as a research tool requiring prospective validation, not as a ready clinical biomarker.
Accordingly, EFS measurement should currently be restricted to exploratory studies that pair EV- or cell-type-specific epigenomic profiling with orthogonal evidence of lipid peroxidation, iron handling, and ferroptosis-specific rescue in matched tissue or cellular models.
Advances in single-vesicle analysis now allow ultrasensitive profiling of histone and DNA modifications within individual EVs. Profiling candidate epigenetic or RNA-regulatory signals in organ-enriched EVs may provide an early, organ-specific indicator of an activated epigenetic-ferroptosis axis, potentially preceding clinical symptoms (Jha et al., 2024; Wu et al., 2025). Integrating these epigenetic reads with EV-carried proteins, metabolites, and noncoding RNAs can yield a candidate multi-omic signature for early detection, risk stratification, and therapy monitoring (Fig. 5) (Gauthier et al., 2022; Hu et al., 2020).
A proposed noninvasive diagnostic pipeline: Profiling extracellular vesicle epigenetic signatures to detect the candidate diabetic epigenetic-ferroptosis axis. Hypothetical workflow for isolating tissue-specific EVs from patient biofluids and analyzing their multi-omic cargo to derive an integrated research-stage risk score, potentially enabling early detection and monitoring of the candidate epigenetic-ferroptosis memory signature. Clinical validation against established predictors is required.
Turning this idea into a reliable biomarker faces technical and biological hurdles. Plasma EVs come from many tissues, diabetes creates systemic inflammatory noise, and fragmented nucleic acid makes epigenomic assays technically fragile.
Organ-Specific EV Isolation: Plasma contains a heterogeneous mixture of EVs from multiple tissues. Distinguishing target organ-derived EVs (e.g., from kidney tubules) from the abundant background, due to the current lack of perfectly specific surface markers, remains a primary technical challenge. Confounding by Systemic Inflammation: The chronic low-grade inflammation inherent to diabetes can alter the global EV release profile and cargo, potentially confounding the detection of signatures specifically arising from localized ferroptotic processes in the target organ. Technical Limitations of Epigenomic Analysis: The DNA within small EVs is often highly fragmented and of low yield. Performing bisulfite sequencing or methylation-specific Polymerase chain reaction on such material demands ultrasensitive methods, which can introduce technical noise and may not accurately reflect low-frequency but pathologically critical methylation changes.
Future studies must adopt rigorous controls and work toward standardized protocols. Combining multi-omic data from EVs—epigenomic, transcriptomic, and lipidomic—may provide cross-validated, composite signatures (e.g., an EV-based EFS, EV-EFS) to enhance specificity and predictive power.
Dynamic epigenetic-memory biomarkers remain promising but premature for clinical use. Key challenges include standardizing EV isolation, validating organ-specific surface markers, proving that EV epigenetic cargo reflects tissue ferroptosis susceptibility, and testing whether these signatures predict outcomes beyond albuminuria, retinal imaging, neuropathy scores, and established inflammatory or oxidative biomarkers.
Therapeutic strategies: Experimental axis-disruption approaches and translational constraints
Therapeutic implications must be interpreted cautiously. Ferroptosis inhibitors, epigenetic drugs, EV-based diagnostics, and CRISPR/dCas9 epigenetic editing are useful experimental probes, but none currently establishes a near-term disease-modifying therapy for diabetic complications. Conventional antioxidants may fail because they incompletely target membrane phospholipid peroxidation, do not restore GPX4/system xc-/NRF2-linked defenses, and do not erase persistent chromatin or RNA-regulatory changes. This distinction, rather than a simple ROS-versus-ferroptosis dichotomy, is the mechanistic rationale for studying ferroptosis-specific and epigenetic approaches. None of the strategies discussed below should be read as a current clinical recommendation. They remain experimental, mechanistically appealing approaches that require extensive preclinical validation before any consideration of human use.
Experimental Strategy 1: Precision Epigenetic Editing. Programmable tools such as CRISPR-dCas9 fused to TET, DNMT, HAT, or histone demethylase domains could test whether specific marks at GPX4, SLC7A11, ACSL4, or NRF2-linked loci are causal. At present, this is primarily a mechanistic validation strategy rather than a near-term therapy because of delivery constraints, off-target epigenetic effects, immune responses, durability concerns, and regulatory barriers.
The translational path for this approach faces substantial barriers:
Delivery Constraints: The large size of CRISPR-dCas9-epigenetic effector fusions challenges the packaging capacity of common viral vectors (e.g., AAV). Nonviral methods currently lack efficiency and specificity. Off-Target Epigenetic Perturbations: The enzymatic domains fused to dCas9 could catalyze modifications at off-target genomic sites with sequence similarity, leading to long-lasting and potentially pathogenic dysregulation of gene networks in an unpredictable manner. This necessitates more selective systems and comprehensive off-target profiling. Durability in Postmitotic Tissues: A critical unknown is the longevity of the epigenetic correction in nondividing target cells (e.g., podocytes, neurons), which dictates whether this is a one-time intervention or requires repeated administration. Regulatory Hurdles: In vivo epigenetic editing falls under stringent gene therapy regulations, demanding extensive safety assessments to exclude risks of germline alteration or genotoxicity.
Emerging Strategy 2: Sequential Combination Therapy. A more immediately translatable paradigm involves temporally sequencing epigenetic modulators with ferroptosis inhibitors. The rationale is mechanistically sequential: first, low-dose epigenetic drugs (e.g., DNMT or HDAC inhibitors) are administered to remodel chromatin and reactivate transcription of defense genes like GPX4 (Wang et al., 2024; Wei et al., 2025). This process requires time for protein resynthesis. Subsequently, ferroptosis inhibitors are introduced to protect the cells as their endogenous antioxidant capacity is being rebuilt. This capitalizes on a therapeutically primed state, enhancing efficacy and allowing dose reduction.
The clinical implementation of this sequential strategy hinges on several practical considerations:
Therapeutic Window and Timing Problem: This strategy is conceptually aligned with intervening during the “Sensitization” phase of our state transition model, aiming to reverse the pro-ferroptotic epigenome before massive tissue loss and fibrosis (the “Propagation” phase) render intervention futile. This directly addresses the timing problem: if ferroptosis susceptibility is epigenetically primed, the therapeutic window is likely early and potentially reversible; once tissue loss is advanced, blocking ferroptosis may offer only limited benefit. This distinction is critical for translating the model into prevention versus treatment and underscores the importance of early detection. Drug Safety and Specificity: The epigenetic-modulating drugs themselves can cause genome-wide epigenetic disturbances and cytotoxicity. This calls for targeted delivery systems or pulsed, low-dose regimens that achieve transient epigenetic modulation without sustained global alterations. Optimization of Regimen: Determining the optimal sequence, timing, dosing, and duration for the epigenetic modulator and the ferroptosis inhibitor requires careful pharmacokinetic/pharmacodynamic modeling and validation in advanced disease models.
To guide translational efforts, we evaluate key nodes within the axis from a druggability perspective:
Direct Executor Inhibition (High Feasibility): Targeting GPX4 (with activators) or ACSL4/TfR1 (with inhibitors) offers a direct, acute rescue. The challenge is ensuring durable benefit without addressing the upstream epigenetic drive. Epigenetic Modulator Inhibition (Variable Specificity/Risk): Pan-DNMT inhibitors (e.g., decitabine) can broadly reverse hypermethylation but risk genomic instability. More specific agents, like EZH2 inhibitors (tazemetostat), target repressive H3K27me3 marks with potentially better tolerability (Wang et al., 2024). Precision Editing (High Impact, High Technical Hurdle): CRISPR-based editing offers high locus-level specificity but faces major barriers related to delivery, off-target control, and long-term safety validation outlined above.
Overall, the near-term priority is not to claim disease-modifying “memory erasure,” but to determine whether targeted modulation of regulatory priming plus ferroptosis inhibition improves outcomes in rigorously designed diabetic models. Translation will require organ-targeted delivery, dose and timing optimization, long-term safety studies, and biomarkers that show predictive value in human cohorts (Fig. 6).
Experimental and translational strategies for testing the proposed axis. The figure contrasts established clinical management with investigational ferroptosis-targeted, epigenetic, and combination approaches, emphasizing delivery and safety barriers.
Critical Limitations and Open Questions
Before this line of work can guide treatment, several practical questions remain.
Methodological constraints obscuring causal resolution
Current evidence faces two major technical hurdles. The first is the limitation in spatial and cellular resolution. Current evidence faces three technical hurdles. The first is the need to consider type 1 and type 2 diabetes as distinct biological contexts. Type 1 diabetes is typically characterized by autoimmune β-cell destruction, absolute insulin deficiency, younger age at onset, and often longer cumulative glycemic exposure, whereas type 2 diabetes is accompanied by insulin resistance, obesity-related chronic inflammation, dyslipidemia, and heterogeneous mitochondrial and lipid metabolism. These differences influence iron handling, lipid composition, antioxidant capacity, immune cell infiltration, and epigenetic remodeling. The epigenetic-ferroptosis axis may operate differently between the two types, and future studies should analyze them separately or stratify analyses by diabetes type rather than assuming a single unified model. The second is the limitation in spatial and cellular resolution. Bulk omics approaches (RNA-seq, Bisulfite-seq) cannot determine if the cells harboring the pro-ferroptotic epigenome are the same cells that ultimately undergo lipid peroxidation. Single-cell multi-omics (e.g., scRNA-seq with scATAC-seq or scNOMe-seq) are essential to deconvolute this causal relationship at cellular resolution. A third limitation relates to temporal dynamics. Longitudinal studies that map the establishment of specific epigenetic marks (e.g., GPX4 promoter methylation) against the onset of ferroptotic execution are scarce, blurring the line between priming event and terminal phase.
The feedback loop: a graded evidence landscape
The “Ferroptosis-to-Epigenetics” feedback component remains the least mature part of the model. Current evidence supports several plausible mediators but does not yet establish a self-reinforcing circuit in diabetic tissues.
Relatively stronger diabetic evidence: Lactate accumulation and histone lactylation have been linked to fibrotic and inflammatory programs in diabetic kidney-related contexts, and some studies connect lactylation-associated pathways with ferroptosis regulators. Whether this represents direct ferroptosis-to-chromatin feedback remains to be tested. Primary diabetic nephropathy studies support lactylation-linked ferroptosis regulation, whereas EV-mediated ferroptosis propagation has been shown more directly in hypoxic kidney injury models and should therefore be treated as indirect mechanistic support for diabetic complications (Hong et al., 2026; Wang et al., 2024d). Indirect evidence from related pathologies: Reactive aldehydes such as 4-HNE can modify proteins including metabolic or chromatin-associated enzymes in cancer, ischemia, and cardiac models. Their locus-specific effects on DNMTs, TETs, HDACs, or histone marks in diabetic tissues remain insufficiently defined. Hypothetical and high-priority: Direct feedback from ferroptotic lipid peroxidation to stable DNA methylation or histone modification at ferroptosis-regulatory loci in diabetic target cells remains unproven and should be tested with time-resolved multi-omics, biochemical adduct mapping, and rescue experiments.
The next validation step should use EV-depletion and cargo-rescue experiments, time-resolved single-cell or spatial multi-omics, biochemical mapping of lipid-aldehyde adducts, and ferroptosis-specific rescue designs to determine whether ferroptotic products precede, cause, and sustain stable epigenetic remodeling in diabetic tissues.
Resolving the survival-death paradox
Cells with similar epigenetic burdens can die by ferroptosis or enter a senescent, pro-inflammatory state (SASP). We propose that cell-type-specific metabolic set points, especially the basal labile iron pool, govern this fate decision. Tissues with inherently high iron turnover (e.g., renal tubular cells) may cross the ferroptosis threshold more readily upon epigenetic priming. Cell types with tighter iron regulation (e.g., mesangial cells) might instead default to senescence in response to the same stress. Cell-intrinsic iron metabolism is a critical decoder of the pro-ferroptotic epigenome and a key area for future investigation.
Future Perspectives and Resources
This review turns a scattered literature into testable predictions without settling the mechanism.
A reference table of epigenetic modulators in the ferroptosis axis
The establishment and maintenance of the pro-ferroptotic epigenome involve a coordinated set of epigenetic writers, erasers, and readers. Table 5 consolidates current evidence linking specific epigenetic enzymes and modifications to the regulation of core ferroptosis genes, serving as a resource for future mechanistic and therapeutic studies.
Predictions, testable hypotheses, and falsifiability
A useful model should make predictions that can fail. For this topic, the most important tests are straightforward: the regulatory change should come first, it should predict later injury better than current markers, and intervention should work best before the tissue crosses into irreversible loss.
Core predictions for validation, revised to match the evidence grade:
Regulatory reversal prediction: If a defined epigenetic or RNA-regulatory alteration truly lowers ferroptosis threshold, targeted reversal of that alteration should restore GPX4/SLC7A11/NRF2 or lipid/iron-homeostasis function and reduce lipid peroxidation in the same vulnerable cell population. This prediction is supported by primary studies in which manipulation of EZH2/SLC7A11, DNMT/GPX4, METTL3/YTHDF3-TfR1, or lactylation-linked pathways altered ferroptosis-relevant readouts in diabetic kidney disease, diabetic osteoporosis, or high-glucose models (Hong et al., 2026; Wang et al., 2024; Wei et al., 2025; Zhang et al., 2025). Biomarker prediction: A candidate EFS should precede or predict conventional injury markers only if it is measurable in organ-enriched EVs or tissue samples, reproducible across cohorts, and additive to existing clinical predictors. For biomarker predictiveness, human urinary exosomal miRNA studies in diabetic kidney disease provide a primary-literature precedent for discovery-validation designs, although they also illustrate the need for prospective outcome testing and specificity beyond renal dysfunction alone (Zang et al., 2019). Timing prediction: Regulatory modulation should be most effective during a reversible susceptibility phase, whereas late fibrosis, neurodegeneration, or vascular dropout may be less responsive even if ferroptosis markers remain detectable. The timing prediction is consistent with primary clinical evidence for legacy effects after early intensive glycemic control in type 1 and type 2 diabetes, which supports testing whether regulatory intervention is most effective before irreversible tissue loss is established (DCCT/EDIC Research Group, 2000; Holman et al., 2008). Falsifiability Criteria:
The model would fail if any of the following were demonstrated:
Failed Rescue by Ferroptosis Inhibition: If potent ferroptosis inhibitors (e.g., Ferrostatin-1, Lip-1) fail to mitigate tissue damage in animal models where complications progress after glycemic normalization (i.e., the pure “memory phase”), it would suggest that alternative, nonferroptotic death or dysfunction pathways are dominant in that context. Lack of Epigenetic Priming: If comprehensive epigenomic profiling (e.g., ATAC-seq, ChIP-seq for relevant marks) in target tissues from patients with progressing complications reveals no stable, reproducible epigenetic alterations at the loci of key ferroptosis genes (e.g., GPX4, SLC7A11, ACSL4) compared with controls or nonprogressors, the core premise of an “epigenetically programmed” state would be invalid. Independence from Iron Dysregulation: If the progression of complications in established metabolic memory models proceeds normally in the setting of profound systemic iron chelation or in genetic models of cellular iron deficiency, the centrality of iron-dependent lipid peroxidation in that model would be weakened.
These failure conditions prevent the review from mistaking an appealing narrative for a definitive conclusion. If ferroptosis blockade does not rescue injury, if epigenetic priming is absent, or if iron dysregulation is irrelevant, the model should be revised (Fig. 7).
A rigorous validation framework: Testable predictions and falsifiability criteria for the epigenetic-ferroptosis axis model. The model generates specific, testable hypotheses (left) and defines clear conditions for its potential falsification (right), establishing a roadmap for future experimental and clinical validation.
Interplay with other cell-death pathways: Context and specificity
Ferroptosis is not the sole cell-death output of diabetic tissue injury. Apoptosis, pyroptosis, necroptosis, autophagy-dependent death, senescence-associated dysfunction, and inflammatory Damage-associated molecular pattern (DAMP) signaling can coexist or occur sequentially. Diabetic epigenetic memory may bias certain vulnerable cells toward ferroptosis under specific iron-lipid-redox conditions, while other cells may undergo alternative fates.
Similar regulatory logic may occur in chronic kidney disease, ischemia-reperfusion injury, neurodegeneration, cancer therapy responses, and other chronic inflammatory states. What may distinguish diabetes is the initiating metabolic history: recurrent hyperglycemia, AGE/RAGE signaling, mitochondrial overproduction of ROS, altered one-carbon metabolism, NAD+/sirtuin imbalance, and tissue-specific exposure to glucolipotoxic stress. This model therefore emphasizes diabetic context and organ specificity rather than claiming a diabetes-exclusive mechanism.
Comparative context
Apoptosis is often triggered by specific death receptor engagement or intracellular damage sensors, leading to caspase activation and an immunologically “silent” phagocytic clearance. Necroptosis and pyroptosis are more inflammatory, involving RIPK1/RIPK3/MLKL and inflammasome/caspase-1/GSDMD axes, respectively. Ferroptosis is uniquely defined by iron-dependent phospholipid peroxidation and is genetically and morphologically distinct.
The epigenetic bias hypothesis
We propose that the hyperglycemia-induced pro-ferroptotic epigenome does not merely increase general cell death susceptibility but actively biases the cell’s demise toward ferroptosis when a subsequent insult occurs. This bias may result from the epigenetic silencing of alternative RCD pathways or the simultaneous priming of ferroptosis components. For instance, repression of GPX4 directly removes a key brake on lipid peroxidation, while the epigenetic activation of ACSL4 enriches membranes with peroxidation-susceptible lipids, creating a biochemical milieu that favors ferroptosis over other death modalities.
Implications and co-existence
This model allows the co-occurrence of other RCD forms. In advanced disease, ferroptosis-derived damage (e.g., release of DAMPs, inflammation) could secondarily trigger apoptosis or necroptosis in neighboring cells. The specific tissue context (e.g., availability of iron, lipid composition, expression of caspase-8) may determine the dominant death pathway in different complications. Future single-cell multi-omics studies in diabetic tissues are needed to map the co-occurrence and sequence of different RCD pathways and to test the “epigenetic bias” hypothesis directly.
These interactions broaden the conceptual scope of the axis, highlight the specificity of the proposed mechanism, and connect it to the wider field of cell death biology (Table 4).
Materials and Methods
This narrative review synthesized peer-reviewed literature indexed in PubMed/Medline, Web of Science, Scopus, and Google Scholar up to May 2026. Search terms combined diabetes or diabetic complications with metabolic memory, epigenetic memory, DNA methylation, histone modification, noncoding RNA, m6A, ferroptosis, lipid peroxidation, GPX4, SLC7A11, ACSL4, TfR1, NRF2, FSP1, DHODH, EVs, and RCD. Inclusion criteria prioritized diabetic human, animal, or cellular studies that reported epigenetic or RNA-regulatory changes together with ferroptosis-relevant endpoints. Nondiabetic disease models were included only when they clarified mechanism and were labeled as indirect evidence. Studies were excluded from strong-evidence statements when they lacked ferroptosis rescue experiments, used only continuous high-glucose exposure without a memory-like phase, or did not separate ferroptosis from apoptosis, pyroptosis, necroptosis, senescence, or nonspecific oxidative injury.
Because glucose modeling varies widely across studies, we recorded whether experiments used transient high glucose followed by normoglycemia, continuous high glucose, oscillating glucose, or advanced glycation/lipid stress, and whether exposure duration reflected acute stress, short-term priming, or longer memory-like persistence. We interpret findings from continuous high-glucose models more cautiously than true metabolic-memory paradigms.
Conclusion
The epigenetic-ferroptosis link is a useful but still incomplete way to think about diabetic complications. It is strongest where diabetic models connect metabolic stress to durable regulatory changes and ferroptosis-relevant injury and weakest where it relies on feedback loops, biofluid biomarkers, or evidence imported from other diseases. The next step is to test the pathway more cleanly: temporal precedence, sufficiency, necessity, organ specificity, and ferroptosis-selective rescue should decide which parts of the model survive.
Authors’ Contributions
Q.W. wrote the initial draft of the article. Q.W. and H.F. contributed to article revision and editing. Q.W. and H.F. prepared the final version. All authors reviewed and approved the submitted article.
Footnotes
Statement on the Use of AI-Assisted Technology
AI-assisted tools were used only for language editing, readability checking, and article revision support. No AI-assisted tool was used to generate original scientific ideas, select literature, analyze data, interpret findings, or determine the conclusions of the article. Grammarly Premium was used for grammar, punctuation, and readability checking. The authors independently verified all references, scientific claims, interpretations, and final wording and take full responsibility for the content of the article.
Author Disclosure Statement
The authors declare no conflict of interest.
Funding Information
This research was funded by grants from the Natural Science Foundation of Beijing Municipality (7232047, 7242192) and the Beijing Municipal Administration of Hospitals Incubation Program (PX2023019).