Milk-derived exosome-based strategy targeting ferroptosis–glycolysis network promotes bone regeneration in diabetic aging comorbidity Authors: Zengguang Wang, Haiyang Lin, Yuntao Li, Yiming Li, Chen Xu, Yaokai Gan Published: 2026 ABSTRACT AbstractDiabetes and aging synergistically impair bone regeneration, primarily driven by functional deterioration of bone marrow mesenchymal stem cells (BMSCs). Here, we uncover an m6A-centered regulatory network integrating ferroptosis and glycolytic metabolism that dictates BMSCs dysfunction under diabetes–aging comorbidity. Mechanistically, diabetic stress selectively activated the ALKBH5–ferroptosis axis, driving excessive ferroptosis in young BMSCs, whereas senescent BMSCs exhibited an additional repression of METTL3-mediated glycolytic flux, rendering ferroptosis inhibition alone insufficient to restore osteogenic capacity. These intersecting molecular defects cooperatively exacerbated osteoporosis progression in diabetic aging. Guided by this mechanistic framework, we developed a combinatorial regenerative strategy integrating milk-derived exosomes (MiExos) and anaerobic exercise to simultaneously target ferroptotic stress and metabolic insufficiency. Mechanistically, MiExos selectively suppressed ferroptosis by stabilizing NRF2 signaling through inhibition of ubiquitination, without perturbing glycolytic metabolism, while anaerobic exercise robustly enhanced intraosseous glycolysis. Notably, the dual intervention markedly improved bone regeneration in diabetic-aged mice. Collectively, our findings establish milk-derived exosomes as a natural nanotherapeutic capable of precise ferroptosis modulation and demonstrate that coordinated suppression of ferroptosis and activation of glycolysis is essential for overcoming bone regenerative failure in diabetes–aging comorbidity.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12951-026-04588-8.Keywords: M6A, Glycolysis, Ferroptosis, MiExos, Anaerobic exercise Introduction Diabetes has emerged as one of the most prevalent and serious chronic diseases worldwide, affecting hundreds of millions of individuals and imposing a substantial burden on public health systems [1]. Accumulating epidemiological evidence indicates that diabetes and aging frequently coexist, forming a highly overlapping comorbid condition. The interplay between diabetes and aging exacerbates systemic metabolic dysregulation and markedly increases the susceptibility to a wide range of complications [2, 3]. Among these, osteoporosis represents a common yet long underrecognized consequence, substantially elevating the risk of fragility fractures [4–6]. Although diabetes and aging independently disrupt bone regeneration through multiple mechanisms, including epigenetic dysregulation, ferroptosis, and metabolic reprogramming [7–9], the molecular mechanisms underlying osteoporosis in the context of diabetes–aging comorbidity, as well as the development of effective targeted therapeutic and preventive strategies, remain poorly understood. BMSCs play a pivotal role in maintaining bone homeostasis by serving as the primary source of osteoblasts and orchestrating the dynamic balance between osteogenic and adipogenic differentiation [10, 11]. Emerging evidence has identified N6-methyladenosine (m6A) RNA modification as a critical epigenetic regulator of stem cell fate and stress adaptation [12, 13]. Among m6A regulatory machinery, the methyltransferase METTL3 and the demethylase ALKBH5 act as central “writer” and “eraser,” respectively, coordinating m6A deposition and removal [14]. Dysregulation of this m6A machinery has been implicated in impaired osteogenesis and accelerated osteoporosis progression [15–18]. However, whether m6A-dependent regulatory networks integrate ferroptosis and metabolic remodeling to drive BMSCs dysfunction under diabetes–aging comorbidity remains largely unknown. Current therapeutic strategies for osteoporosis primarily focus on anti-resorptive and anabolic agents, including bisphosphonates, denosumab, and parathyroid hormone analogs, often combined with bone-friendly lifestyle interventions such as weight-bearing exercise [19, 20]. Although these interventions are effective in reducing fracture risk and improving bone mineral density, their long-term clinical use is often limited by safety concerns, high costs, and variable patient adherence [21]. Moreover, most existing therapies predominantly target bone remodeling outcomes rather than addressing the underlying cellular dysfunction and metabolic imbalance that drive osteoporosis progression, particularly in complex pathological settings such as diabetes–aging comorbidity [22]. As a result, there remains a pressing need for alternative therapeutic approaches that are safe, sustainable, and capable of modulating fundamental pathogenic mechanisms to restore bone regenerative capacity. Natural milk-derived exosomes (MiExos) are nanoscale extracellular vesicles widely present in mammalian milk. As diet-derived bioactive carriers, milk-derived exosomes are continuously ingested by humans via the oral route and exhibit excellent biocompatibility and biological safety [23, 24]. In recent years, accumulating evidence has demonstrated that milk-derived exosomes possess considerable therapeutic potential in preclinical models of intestinal inflammation [25], metabolic disorders [26], tumors [27], and immune-related diseases [28], with oral administration being the primary routes of delivery [29]. Notably, milk-derived exosomes display remarkable stability in vivo and favorable tissue distribution profiles [30]. Moreover, milk-derived exosomes are enriched with multiple functional components, including microRNAs, proteins, and lipids, which exert pleiotropic biological effects such as anti-inflammatory, antioxidant, and immunomodulatory activities [31, 32], thereby positioning milk-derived exosomes as promising candidates for addressing complex multifactorial disease processes. In this study, we uncover an age-dependent divergence in osteogenic regulation under diabetic conditions, wherein ferroptosis inhibition effectively rescues young BMSCs but fails to restore osteogenic capacity in senescent BMSCs. Mechanistically, this resistance arises from a profound imbalance in m6A modification, characterized by hyperactivation of the ALKBH5–ferroptosis axis and concomitant suppression of the METTL3–glycolysis axis, leading to a decoupling of metabolic and redox homeostasis. Building on this mechanistic framework, we establish a combinatorial intervention integrating milk-derived exosomes to restrain ferroptosis and anaerobic exercise to restore glycolytic flux, thereby synergistically promoting bone regeneration in diabetic-aged mice. Our findings reveal an m6A-centered regulatory network governing BMSCs dysfunction and propose a simple, noninvasive, and translationally promising strategy for treating osteoporosis under diabetes–aging comorbidity (Scheme 1). Graphical abstract of m6A dysregulation–induced glycolysis–ferroptosis imbalance in BMSCs and the combinatorial therapeutic strategy using milk-derived exosomes and anaerobic exercise Materials and methods Cell cultures The RAW264.7 macrophage and HUVEC cell line was purchased from Shanghai Fuheng Biology (FH0328, FH-H040). Cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin (Gibco, USA). Primary BMSCs were isolated from the tibiae of 8-week-old male C57BL/6 J mice. Briefly, bone marrow was flushed out and cells were cultured in minimum essential medium-α (MEM-α) (Gibco, USA) supplemented with 15% FBS and 1% penicillin–streptomycin. All cell types were maintained at 37 °C in a humidified incubator containing 5% CO₂. Animal experiments All animal experiments were approved by the Institutional Animal Ethics Review Board of the Shanghai Ninth People’s Hospital, Shanghai Jiao Tong University School of Medicine (Approve No.SH9H-2025-A1614-1). C57BL/6J mice were purchased from Shanghai Jihui Laboratory Animal Care Co., Ltd. and maintained under specific pathogen-free (SPF) conditions at a controlled temperature of 22–24 °C with a 12 h light/12 h dark cycle, with ad libitum access to food and water. The experimental duration was 12 weeks. The body weight, blood glucose and fasting insulin of each mouse were measured every week until the end of the animal experiment. At the end of the study, mice were euthanized by CO₂ inhalation followed by cervical dislocation, blood and tissue samples were collected for subsequent analyses. A type 2 diabetes mouse model was established according to previously reported protocols [7]. After one week of acclimatization, mice were randomly assigned to the following groups: normal control group (Con, n = 6, 2 months old), aging group (Aging, n = 6, 12 months old), diabetes group (Diabetes, n = 6, 2 months old), and diabetes and aging comorbidity group (Diabetes + Aging, n = 6, 12 months old). For mice receiving gene intervention, lentiviruses carrying the target gene or control lentivirus were injected intratibially prior to diabetes induction. For the 2-deoxy-D-glucose (2-DG) treatment group, mice were administered 2-DG (0.5 mg/g, ST1024, Beyotime, Shanghai, China) every three days from the initiation of diabetes modeling for 12 weeks. For MiExos treatment, MiExos were supplemented in drinking water at a concentration of 400 µg/mL throughout the 12-week experimental period. For the combined treatment group, mice received MiExos supplementation together with anaerobic exercise training performed on a treadmill with a defined incline at 70% of maximal running speed for 1 h per session, every three days for 12 weeks, as previously described [33]. Cell cytotoxicity assay Cell viability was evaluated using a Cell Counting Kit-8 (C0037, Beyotime) and a Live/Dead Cell Staining Kit (C2015M, Beyotime) according to the manufacturer’s instructions. Briefly, BMSCs or HUVECs were seeded and subjected to the indicated treatments. The culture medium was then replaced with fresh medium containing 10% CCK-8 solution, followed by incubation at 37 °C for 1 h. The absorbance at 450 nm was measured using a microplate reader (Infinite M200 Pro, Tecan, Männedorf, Switzerland). For live/dead staining, treated cells were incubated with Calcein-AM and PI for 30 min at room temperature in the dark. After washing with PBS, fluorescence images were captured using a fluorescence microscope (DMi8, Leica Microsystems, Wetzlar, Germany). Hemolysis assay The hemocompatibility of MiExos was assessed by an in vitro hemolysis assay. Briefly, anticoagulated whole blood was collected from mice and centrifuged at 1,000 rpm for 10 min to isolate red blood cells (RBCs). The obtained RBCs were washed three times with PBS and diluted to a final concentration of 5% (v/v) in PBS. Subsequently, 1 mL of the RBC suspension was added to each well of a 48-well plate, followed by the addition of MiExos at the indicated concentrations. The samples were incubated at 37 °C for 1 h. PBS and Triton X-100 solution were used as negative and positive controls, respectively. After incubation, the suspensions were centrifuged at 1,000 rpm for 10 min, and the supernatants were collected. The hemolytic activity was visually observed and photographed. Meanwhile, the absorbance of the supernatant was measured at 540 nm using a microplate reader (Infinite M200 Pro, Tecan). Quantitative real-time PCR analysis BMSCs were seeded in six-well plates at a density of 1 × 10⁵ cells per well and cultured for 24 h. After treatment, total RNA was extracted using TRIzol reagent (R0016, Beyotime) according to the manufacturer’s protocol. Reverse transcription was performed with PrimeScript RT Master Mix (RR036A, Takara, Shiga, Japan). Quantitative real-time PCR was conducted using SYBR Green qPCR Master Mix (B21702, Selleck, TX, USA). Relative mRNA expression levels were calculated using the 2⁻ΔΔCt method, with β-actin as the internal reference gene. The primer sequences used in this study are listed in Table 1. Western blot analysis For cell samples, total proteins were extracted using RIPA lysis buffer (P0013B, Beyotime) supplemented with 1% protease–phosphatase inhibitor cocktail (P1045, Beyotime). The lysates were centrifuged at 13,000 × g for 15 min at 4 °C, and the supernatants were collected. Protein samples were mixed with 5× SDS-PAGE loading buffer (P0015, Beyotime) and denatured by boiling at 99 °C for 5 min. Nuclear proteins were isolated using a nuclear protein extraction kit (P0027, Beyotime) according to the manufacturer’s instructions. Bone tissue proteins were isolated using a bone tissue protein extraction kit (EX1270, Solarbio, Beijing, China). The proteins were separated by 4–20% SDS-PAGE and subsequently transferred onto 0.22 μm PVDF membranes (Merck-Millipore, Temecula, CA, USA). The membranes were blocked with 5% non-fat milk (P0216, Beyotime) at room temperature for 2 h and then washed three times with TBST, followed by incubation with primary antibodies at 4 °C overnight. After washing three times with TBST, the membranes were incubated with species-specific HRP-conjugated secondary antibodies (anti-rabbit or anti-mouse; Proteintech, Wuhan, China). After three additional washes with TBST, protein bands were visualized using an ultrasensitive enhanced chemiluminescence detection reagent (P10100, NCM Biotech, Suzhou, China) and imaged with an eBlot Touch Imager system (e-Blot, Shanghai, China). The source and dilution ratio of the primary antibody are shown in Table 2. Immunofluorescence assay Cells were seeded in 24-well plates or confocal dishes and subjected to the indicated treatments. After treatment, cells were washed three times with PBS and fixed with 4% formaldehyde (P0099, Beyotime) for 30 min at room temperature. Cells were then permeabilized with 0.25% Triton X-100 (P0096, Beyotime) in PBS for 10 min, followed by three washes with PBS. Subsequently, cells were blocked with 5% BSA at room temperature for 1 h and incubated with primary antibodies at 4 °C overnight. After washing three times with PBS, cells were incubated with fluorescently labeled secondary antibodies (anti-rabbit or anti-mouse; Proteintech). Nuclei were counterstained with DAPI for 15 min, followed by three PBS washes. Fluorescence images were captured using fluorescence microscope (DMi8, Leica Microsystems). The source and dilution ratio of the primary antibodies are shown in Table 3. Osteoblast differentiation and mineralization BMSCs were seeded into 48-well plates at a density of 5 × 10⁴ cells per well and cultured for 24 h. Subsequently, osteogenic induction medium (MUXMX-90021, Cyagen Biosciences, Suzhou, China) was applied to induce osteogenic differentiation. The medium was replaced every 3 days, and differentiation was maintained for up to 7 or 21 days. Alkaline phosphatase (ALP) activity and extracellular matrix mineralization were evaluated by staining with a BCIP/NBT alkaline phosphatase staining kit (C3206, Beyotime) and Alizarin Red S solution (C0148S, Beyotime), respectively. For quantitative analysis of ALP and ARS staining, the Alkaline Phosphatase Assay Kit (P0321S, Beyotime) and 10% hexa-decylpyridinium chloride monohydrate were used to measure ALP activity and ARS-stained calcification, respectively, following previously described methods [34]. C11-BODIPY and ferrous ion staining Both C11-BODIPY and ferrous ion staining were performed according to the manufacturers’ instructions. Briefly, after the indicated treatments, cells were washed with PBS and incubated with the working solutions of C11-BODIPY (S0043S, Beyotime) and ferrous ion (S1070S, Beyotime) at 37 °C for the recommended time. After incubation, the cells were washed with PBS and observed using a fluorescence microscope (DMi8, Leica Microsystems). Measurement of Fe²+, MDA and GSH levels According to the manufacturer’s instructions, the ferrous ion concentration was evaluated using the ferrous ion determination kit (S1066S, Beyotime), the content of MDA was detected using the MDA detection kit (S0131S, Beyotime), and the content of GSH was detected using the GSH detection kit (S0053, Beyotime). Glycolytic flux For the detection of glycolytic flux, we performed the assay using the relevant kit according to the manufacturer’s instructions. Glucose detection kit (S0201S, Beyotime), L-Lactate Assay Kit (S0208S, Beyotime), NAD+/NADH Assay Kit (S0175, Beyotime). Isolation of MiExos Fresh milk was centrifuged at 5,000 × g for 20 min at 4 °C to remove the fat layer and large particles. The supernatant was further centrifuged at 12,000 × g for 60 min at 4 °C to eliminate residual lipids and cell debris. The obtained supernatant was then adjusted to the isoelectric point of casein by adding acetic acid and incubated at room temperature for 10 min to induce casein precipitation. Subsequently, the sample was centrifuged at 7000 × g for 30 min at 4 °C to remove the casein precipitate. The resulting supernatant was filtered through a 0.22 μm membrane filter and subjected to ultracentrifugation at 100,000 × g for 90 min at 4 °C to pellet exosomes. The pellet was resuspended in sterile PBS and further washed by ultracentrifugation at 100,000 × g for another 90 min. Finally, milk-derived exosomes (MiExos) were collected for subsequent experiments. The morphology of exosomes was observed using transmission electron microscopy (HT7700, HITACHI, Tokyo, Japan). Particle size, concentration and zeta potential of exosomes were analyzed by ZetaView (Particle Metrix, Germany). The gastrointestinal stability of MiExos Based on prior research, simulated gastric fluid (SGF) and simulated intestinal fluid (SIF) were prepared [35]. To evaluate the stability of MiExos during oral administration, a MiExos suspension was incubated at 37 °C under shaking conditions in PBS, SGF, or SIF. Particle size and morphological integrity were assessed using TEM after incubation. Lentivirus construction and infection Lentiviral vectors for the knockdown of ALKBH5, METTL3 and SIRT6, as well as Lentiviral vectors for the overexpression of wild-type (WT) ALKBH5, mutant ALKBH5, METTL3, and SIRT6, were purchased from Obio (Shanghai, China). Cells were infected with lentiviruses at a multiplicity of infection (MOI) of 25 in the presence of polybrene (5 µg/mL). After 24 h, the infection medium was replaced with fresh complete medium. The infection efficiency was assessed by quantitative PCR (qPCR) 48 h post-infection. The efficiency of lentiviral-mediated knockdown and overexpression of METTL3, ALKBH5 and SIRT6 was rigorously validated via qRT-PCR (Fig. S31). The shRNA sequences used in this study are provided in Supplementary Table 4. Cellular uptake MiExos were labeled with PKH67 fluorescent dye according to the manufacturer’s instructions (C3635S, Beyotime), and cell nuclei were counterstained with DAPI (C1006, Beyotime). BMSCs were incubated with PKH67-labeled MiExos for different time points. The cellular uptake of MiExos was observed using fluorescence microscope (DMi8, Leica Microsystems). To investigate the mechanisms of MiExos internalization, BMSCs were pretreated with specific endocytosis inhibitors, including nystatin (Nys, 11 µM), chlorpromazine (Chl, 50 µM), and (N-ethyl-N-isopropyl) amiloride (EIPA, 50 µM), for 30 min at 37 °C. Subsequently, cells were incubated with PKH67-labeled MiExos. The uptake efficiency was evaluated by fluorescence microscope (DMi8, Leica Microsystems). Micro-CT analysis Micro-computed tomography (micro-CT) analysis was performed on the tibiae of mice to evaluate the trabecular bone microarchitecture beneath the tibial plateau growth plate. The scanning was conducted with a current of 0.08 mA, a voltage of 80 kV, and a voxel resolution of 6 μm. The region of interest (ROI) was defined in the proximal tibial metaphysis, starting 2 mm distal to the growth plate and extending 1.9 mm along the longitudinal axis of the tibia. The ROI was restricted to the trabecular bone compartment, excluding cortical bone. Three-dimensional images were reconstructed using Recon software. Quantitative parameters were analyzed, including trabecular thickness (Tb.Th, mm), trabecular number (Tb.N, mm⁻¹), trabecular separation (Tb.Sp, mm), bone volume fraction (BV/TV), and bone mineral density (BMD). Histological analysis Histological analysis was performed as previously described [36]. Briefly, tibial specimens were fixed in 4% formaldehyde and decalcified for 2 months, followed by paraffin embedding and sectioning. Subsequently, 5 μm-thick sections were subjected to HE staining, immunohistochemical staining, and immunofluorescence analysis. Biocompatibility assay To evaluate the in vivo toxicity of MiExos, an independent cohort of mice receiving MiExos alone was established, since diabetes modeling itself may induce damage to internal organs. These mice were maintained under the same housing conditions and treatment duration as the experimental groups. At the end of the intervention period, fresh blood samples were collected for routine serum biochemical analysis. Meanwhile, major organs, including the heart, liver, spleen, lungs, and kidneys, were harvested and subjected to HE staining to assess potential systemic toxicity. For biochemical analysis, albumin (ALB), total bilirubin (TBIL), alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyltransferase (γ-GT), urea nitrogen (BUN), and creatinine (CRE) were measured. Statistics Statistical analyses were performed using GraphPad Prism software (version 10.2.0, San Diego, CA, USA). Data are presented as mean ± standard deviation (SD). Comparisons between two groups were performed using an unpaired two-tailed Student’s t-test. For comparisons among multiple groups, the Brown–Forsythe test was used to assess homogeneity of variances, and the Shapiro–Wilk test was applied to evaluate normality of data distribution. For data satisfying parametric assumptions, one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test was performed. For data not meeting parametric assumptions, the Kruskal–Wallis test followed by Dunn’s post hoc test was used as a non-parametric alternative. The sample size for each experimental group was n ≥ 3. Specific values of n are provided in the figure. Results 1. Diabetic microenvironment exacerbates age-related bone deterioration and induces ferroptosis To evaluate the effects of diabetes on bone quality, a mouse model of type 2 diabetes mellitus (T2DM) was established using a high-fat diet combined with low-dose streptozotocin (STZ) injection (Fig. 1a). After 12 weeks, diabetic mice (DM) showed significantly increased body weight and blood glucose levels compared with controls (Con) (Fig. S1a-b), accompanied by aggravated insulin resistance (Fig. S1c), confirming successful model establishment. Diabetic microenvironment exacerbates age-related bone deterioration and induces ferroptosis. a Schematic illustration of the type 2 diabetes mellitus (T2DM) mouse model established by high-fat diet (HFD) feeding combined with low-dose streptozotocin (STZ) injection (Created with BioGDP.com). b–c Representative micro–CT images of the proximal tibia and quantitative analysis of BV/TV and BMD. DM: Diabetes. Scale bar: 1 mm. d Representative HE staining of tibial sections. Scale bar: 200 μm. e-g Correlation analyses of HbA1c with metabolic (glucose) and iron-related parameters (transferrin receptor and ferritin) in participants from the NHANES 2021–2023 dataset. h–i Ferrous iron (Fe²⁺) concentration in serum and malondialdehyde (MDA) levels in bone tissue. j–k Representative immunohistochemical staining for 4-hydroxynonenal (4-HNE) in tibial sections and corresponding quantitative analysis. Scale bar: 100 μm. l mRNA expression levels of ferroptosis-related genes ACSL4, GPX4, and TFRC in bone tissue. m–n Protein expression levels of ACSL4 and GPX4 in bone tissue and quantitative analysis. o–p Representative immunofluorescence images showing LepR and 4-HNE staining in bone tissue. Scale bar: 50 μm. Data are presented as mean ± SD, n ≥ 3 Micro-CT analysis and HE staining revealed marked deterioration of trabecular bone architecture in the tibia of aged mice, including decreased bone volume fraction (BV/TV), bone mineral density (BMD), trabecular number (Tb.N), and trabecular thickness (Tb.Th), together with increased trabecular separation (Tb.Sp). Notably, these age-associated changes were further aggravated in the presence of diabetes (Fig. 1b-d, Fig. S2). Consistently, qRT-PCR and western blot analyses showed increased expression of the aging markers P53、P21 and γH2AX, along with reduced expression of the osteogenic markers COL1A1、OPN and RUNX2 (Fig. S3), indicating that diabetes is associated with accelerated bone aging and impaired osteogenic potential. To explore potential mechanisms, publicly available transcriptomic data from peripheral blood mononuclear cells of patients with diabetes were re-analyzed. GO and KEGG enrichment analyses indicated significant enrichment of oxidative stress–related pathways, and several ferroptosis-associated genes, including ALOX5 and PTGS2, were upregulated (Fig. S4). Analysis of NHANES data further revealed positive correlations of glycated hemoglobin (HbA1c) with both glucose and transferrin receptor levels, whereas no significant association was observed between HbA1c and ferritin, suggesting altered iron handling in diabetes (Fig. 1e–g). Experimental validation showed that diabetic-aged mice exhibited increased serum ferrous iron levels and elevated lipid peroxidation markers (MDA and 4-HNE) in bone tissue (Fig. 1h–k). Tissue staining revealed that pro-ferroptotic proteins ACSL4 was markedly elevated in diabetic tissues, whereas the anti-ferroptotic protein GPX4 was significantly decreased (Fig. S5). Consistently, qRT-PCR and western blot analyses confirmed upregulation of the pro-ferroptotic genes ACSL4 and TFRC, along with downregulation of GPX4, supporting the induction of ferroptosis in the diabetic microenvironment (Fig. 1l–n). It is noteworthy that although ferroptosis-related indicators showed an increasing trend in aged individuals, the differences did not reach statistical significance. This observation suggests that aging alone may be insufficient to robustly induce ferroptosis. Immunofluorescence staining revealed enhanced 4-HNE signals in bone marrow cavity cells under combined aging and diabetes, which partially overlapped with LepR⁺ cells (Fig. 1o–p), indicating that BMSCs may be susceptible to diabetes associated ferroptotic stress. 2. Limited effects of ferroptosis inhibition on osteogenesis in diabetic senescent BMSCs BMSCs were exposed to high glucose (25.5 mM) in combination with palmitic acid (PA) to mimic the diabetic microenvironment (HGPA) [37]. HGPA treatment reduced cell viability in a dose-dependent manner, with a significant decrease observed after 24 h at 200 µM PA (Fig. 2a). Therefore, this concentration was selected for subsequent experiments. Limited effects of ferroptosis inhibition on osteogenesis in diabetic senescent BMSCs. a Cell viability of BMSCs cultured under control glucose (5.5 mM) or high glucose (25.5 mM) with increasing concentrations of palmitic acid (HGPA) for the indicated time points, as assessed by the CCK-8 assay. b-c mRNA and protein expression levels of senescence-associated markers (P53, P21) and the proliferation marker PCNA in BMSCs under different treatments, determined by qRT–PCR and western blot, respectively. d Intracellular GSH and MDA levels quantified using GSH and MDA assay kits. e-f mRNA and protein expression levels of the ferroptosis-related genes ACSL4 and GPX4 in BMSCs under different treatments, analyzed by qRT–PCR and western blot. g-h BMSCs were treated with Control, HGPA, HGPA + ferrostatin-1 (Fer-1), or HGPA + liproxstatin-1 (Lip-1). Representative images of lipid peroxidation staining and ferrous iron staining are shown and quantitative analysis. Scale bar: 50 μm. i-j Immunofluorescence staining showing the protein expression of the senescence markers γH2AX and P53 in BMSCs under different treatments and quantitative analysis. Scale bar: 50 μm. k mRNA expression levels of osteogenic-related genes (RUNX2, COL1A1, OPN, and OCN) determined by qRT–PCR. l Protein expression levels of COL1A1 and OPN assessed by western blot analysis. m Immunofluorescence staining of RUNX2 protein expression in BMSCs under different treatments and quantitative analysis. Scale bar: 50 μm. n-o Osteogenic differentiation capacity of BMSCs evaluated by alkaline phosphatase (ALP) staining and Alizarin Red S staining and quantitative analysis. Data are presented as mean ± SD, n = 3 We first evaluated the impact of the HGPA microenvironment on BMSCs senescence. qRT-PCR and western blot analyses demonstrated a marked upregulation of senescence-associated markers P53 and P21, accompanied by a significant downregulation of the proliferation marker PCNA, indicating accelerated cellular aging under diabetic conditions (Fig. 2b-c). Given previous data implicating ferroptosis in diabetic BMSCs injury, we next sought to determine the dominant mode of programmed cell death. A panel of cell death inhibitors was applied, and CCK-8 assays revealed that only the ferroptosis inhibitor ferrostatin-1 (Fer-1) significantly rescued cell viability, identifying ferroptosis as the primary mechanism mediating HGPA-induced BMSCs death (Fig. S6). Consistently, Metabolic assays revealed a significant increase in malondialdehyde (MDA) levels accompanied by a marked reduction in glutathione (GSH) content (Fig. 2d). qRT-PCR and western blot analyses showed increased expression of the pro-ferroptotic gene ACSL4 and suppressed expression of the anti-ferroptotic gene GPX4 under HGPA conditions (Fig. 2e-f). To further clarify the contribution of ferroptosis to BMSCs senescence and osteogenic dysfunction, two canonical ferroptosis inhibitors, Fer-1 and liproxstatin-1 (Lip-1), were applied. CCK-8 rescue assays determined the optimal concentrations to be 10 µM for Fer-1 and 5 µM for Lip-1 (Fig. S7). Lipid peroxidation and ferrous iron staining confirmed that both inhibitors effectively suppressed ferroptotic activity (Fig. 2g-h). Functionally, young BMSCs, ferroptosis inhibition markedly attenuated cellular senescence, as evidenced by a significant reduction in γH2AX and P53 immunofluorescence signals (Fig. S8). Concurrently, qRT-PCR analysis demonstrated robust upregulation of osteogenic genes (RUNX2, COL1A1, OPN, and OCN) (Fig. S9a). Immunofluorescence staining further confirmed the restoration of the osteogenic master transcription factor RUNX2 and the extracellular matrix protein COL1A1 (Fig. S9b). In line with these molecular changes, Alizarin Red S staining revealed a pronounced recovery of matrix mineralization capacity (Fig. S9c). Together, these findings indicate that in young BMSCs exposed to a diabetic microenvironment, ferroptosis plays a critical role in driving osteogenic impairment, and its inhibition effectively restores osteogenic potential, consistent with previous reports [37]. We next examined whether ferroptosis inhibition exerts comparable effects in senescent BMSCs. In contrast to young cells, immunofluorescence analysis showed that suppression of ferroptosis failed to significantly reduce the senescence-associated markers γH2AX and P53, exhibiting only a modest trend toward improvement (Fig. 2i-j). During osteogenic induction, qRT-PCR revealed only limited recovery of osteogenic gene expression (RUNX2, COL1A1, OPN, and OCN) (Fig. 2k), which was corroborated by western blot and immunofluorescence analyses (Fig. 2l-m). Functional assays, including ALP and Alizarin Red S staining, similarly demonstrated minimal restoration of ALP activity and matrix mineralization capacity (Fig. 2n-o). To further validate these observations, a replicative senescence model was established using passage 8 (P8) BMSCs. Although Fer-1 treatment effectively reduced intracellular ferrous iron accumulation (Fig. S10), it failed to largely attenuate the expression of senescence markers P53 and γH2AX (Fig. S11a-b), and only marginally affected the expression of osteogenic markers RUNX2 and COL1A1 (Fig. S11c-d). Collectively, these results demonstrate that, unlike in young BMSCs, inhibition of ferroptosis alone is insufficient to reverse osteogenic impairment, suggesting that additional mechanisms cooperate with ferroptosis to drive functional decline in senescent BMSCs. 3. m6A-mediated glycolysis–ferroptosis crosstalk regulates osteogenic capacity of BMSCs in senescence and diabetes As the most prevalent internal post-transcriptional RNA modification in eukaryotic cells, N⁶-methyladenosine (m6A) plays a critical role in regulating cell fate decisions [38, 39]. We therefore sought to determine whether m6A homeostasis is altered in BMSCs under diabetic conditions. young BMSCs exhibited a selective upregulation of the m6A demethylase ALKBH5, whereas other m6A regulators remained unchanged. In contrast, senescent BMSCs displayed a distinct m6A dysregulation pattern characterized by concurrent ALKBH5 upregulation and marked downregulation of the m6A methyltransferase METTL3 (Fig. 3a-b). This senescence-dependent divergence was consistently recapitulated in a replicative senescence model, with ALKBH5 elevation alone in early-passage (P3) cells but combined ALKBH5 increase and METTL3 loss in late-passage (P8) BMSCs (Fig. 3c-d, Fig. S12). m6A-mediated glycolysis–ferroptosis crosstalk regulates osteogenic capacity of BMSCs in senescence and diabetes. a–b Protein expression levels of m6A regulatory proteins in BMSCs under different treatments were assessed by western blot with quantitative analysis. c–d METTL3 protein expression was examined by immunofluorescence staining and quantitatively analyzed. Scale bar: 50 μm. e Schematic illustration depicting the experimental design, intervention strategies, and grouping of BMSCs (Created with BioGDP.com). f qRT–PCR analysis of senescence-associated genes P53 and P21. g-h Immunofluorescence staining of P53 protein expression in BMSCs under different treatments and quantitative analysis. Scale bar: 50 μm. i Immunofluorescence staining of osteogenic markers RUNX2 and COL1A1. Scale bar: 50 μm. j Intracellular GSH and MDA levels were quantified using GSH and MDA assay kits. k qRT–PCR analysis of ferroptosis-related genes GPX4 and ACSL4. l Protein expression levels of ferroptosis-related proteins were assessed by western blot with quantitative analysis. m-n Representative images of lipid peroxidation staining in BMSCs and quantitative analysis. Scale bar: 50 μm. o-p Protein expression levels of ferroptosis-related proteins were assessed by western blot with quantitative analysis. q–r Immunofluorescence staining and quantitative analysis of senescence-associated proteins P53 and γH2AX. Scale bar: 50 μm. s–t Immunofluorescence staining and quantitative analysis of osteogenic markers RUNX2 and COL1A1. Scale bar: 50 μm. Data are presented as mean ± SD, n = 3 These divergent m6A signatures led us to hypothesize that the distinct responses of young and senescent BMSCs to ferroptosis inhibition under diabetic conditions arise from differential m6A dysregulation. To test this hypothesis, ALKBH5 was first silenced in young BMSCs. Western blot analysis showed that ALKBH5 knockdown markedly attenuated diabetic microenvironment–induced ferroptosis, as evidenced by downregulation of ACSL4 and upregulation of GPX4, and significantly alleviated cellular senescence, indicated by reduced expression of P53 and γH2AX (Fig. S13a-b). Concomitantly, osteogenic capacity was substantially restored, with immunofluorescence staining revealing significant upregulation of the osteogenic markers RUNX2 and COL1A1 (Fig. S13c). Collectively, these results indicate that diabetes-induced ALKBH5 upregulation acts as a key driver of ferroptosis and senescence in young BMSCs, thereby contributing to impaired osteogenic differentiation. We next investigated senescent BMSCs to clarify the role of m6A in diabetes-associated osteogenic decline (Fig. 3e). Unexpectedly, METTL3 overexpression alone further aggravated cellular senescence, as evidenced by increased expression of P53, P21, and γH2AX, and concomitantly impaired osteogenic differentiation, with reduced RUNX2 and COL1A1 expression, in senescent BMSCs exposed to diabetic stress. In contrast, ALKBH5 knockdown alone exerted only a modest protective effect on senescence and osteogenic capacity. Strikingly, simultaneous METTL3 overexpression and ALKBH5 silencing markedly suppressed senescence-associated markers and robustly restored osteogenic differentiation, as indicated by decreased P53, P21, and γH2AX levels and increased RUNX2 and COL1A1 expression (Fig. 3f–i, Fig. S14). Mechanistically, METTL3 overexpression in senescent BMSCs under diabetic conditions significantly exacerbated ferroptosis, characterized by rised ferrous iron, elevated lipid peroxidation (increased MDA), depleted antioxidant capacity (reduced GSH), upregulation of ACSL4, and downregulation of GPX4. Notably, this pro-ferroptotic effect was completely abolished when ALKBH5 was concurrently silenced (Fig. 3j–l, Fig. S15). These observations suggest that METTL3 and ALKBH5 exert context-dependent synergistic effects on ferroptosis regulation. To elucidate the mechanistic basis of this interplay, we revisited our previous finding that activation of the METTL3–glycolysis axis rescues senescence-associated osteogenic defects [8]. We therefore hypothesized that glycolysis acts as a key intermediary linking METTL3 to ferroptosis. Consistently, METTL3 overexpression markedly enhanced glycolytic activity in BMSCs under diabetic conditions, as evidenced by increased GLUT1 and LDHA expression, restoration of the glycolytic regulator c-Myc (Fig. S16a-b). Further analyses demonstrated that although ALKBH5 knockdown increased glycolytic activity, this effect was completely abolished by concomitant METTL3 silencing (Fig. S16c-d), indicating that ALKBH5 does not directly regulate glycolysis but instead functions in a METTL3-dependent manner. Given the differential expression of METTL3 in young versus senescent BMSCs under diabetic stress, we therefore infer that attenuation of the METTL3–glycolysis axis underlies the limited osteogenic benefit of ferroptosis inhibition in diabetic senescent BMSCs, with glycolysis serving as a critical mediator linking METTL3 to ferroptosis activity. Subsequent studies have confirmed METTL3 overexpression alone did not induce ferroptosis but instead enhanced osteogenic differentiation. However, upon ferroptosis induction with erastin, METTL3 overexpression markedly exacerbated ferroptotic injury, as indicated by increased lipid peroxidation, ACSL4 upregulation, and GPX4 suppression (Fig. 3m-p), accompanied by elevated senescence markers P53 and γH2AX and reduced osteogenic markers RUNX2 and COL1A1. Importantly, glycolysis inhibition effectively reversed these effects, indicating that METTL3 amplifies ferroptosis in a glycolysis-dependent manner under diabetic stress (Fig. 3q-t). The CCK-8 results confirmed the excellent biocompatibility of 2-DG at the selected concentrations, ruling out the possibility that the observed experimental phenotypes were influenced by direct cellular damage (Fig. S17). Together, these findings indicate that two interconnected dysregulated axes—ALKBH5/ferroptosis and METTL3/glycolysis—coexist in senescent BMSCs under diabetic conditions and jointly determine BMSCs osteogenic potential. The subsequent sections will focus on dissecting the intrinsic regulatory mechanisms of the ALKBH5–ferroptosis axis. 4. ALKBH5 regulates ferroptosis through the SIRT6-NRF2 axis Given the well-established role of the SIRT6–NRF2 axis in maintaining redox homeostasis and limiting ferroptosis [40, 41] (Fig. 4a), we next investigated whether ALKBH5 exerts its effects on ferroptosis through the SIRT6–NRF2 pathway. Western blot and immunofluorescence analyses demonstrated that HGPA exposure markedly suppressed the SIRT6–NRF2 axis, accompanied by significant downregulation of its downstream antioxidant genes HO-1, NQO1, and GPX4. In contrast, ALKBH5 knockdown robustly restored the expression of SIRT6–NRF2 signaling components and their downstream targets (Fig. 4b-c, Fig. S18). Consistently, immunofluorescence co-localization assays revealed that HGPA significantly disrupted the spatial association between SIRT6 and NRF2, whereas silencing ALKBH5 effectively reinstated their co-localization (Fig. 4d-e), indicating that ALKBH5 negatively regulates the SIRT6–NRF2 antioxidant axis. ALKBH5 Regulates Ferroptosis Through the SIRT6-NRF2 Axis. a Schematic diagram of SIRT6-NRF2 axis (Created with BioGDP.com). b Protein expression levels of SIRT6-NRF2 axis proteins in BMSCs under different treatments were assessed by western blot with quantitative analysis. c Immunofluorescence staining of GPX4 protein expression in BMSCs under different treatments. Scale bar: 50 μm. d-e Representative immunofluorescence images showing the co-localization of NRF2 and SIRT6 and Pearson’s R value. Scale bar: 10 μm. f Protein expression levels of NRF2 axis proteins in BMSCs under different treatments were assessed by western blot with quantitative analysis. g-h Representative images of lipid peroxidation staining and ferrous iron staining are shown. Scale bar: 50 μm. i Immunofluorescence staining of NRF2 protein expression in BMSCs under different treatments. Scale bar: 50 μm. j qRT–PCR analysis of ferroptosis-related genes GPX4 and ACSL4. k Protein expression levels of ferroptosis-related proteins were assessed by western blot with quantitative analysis. l Protein expression levels of SIRT6-NRF2 axis proteins in BMSCs under different treatments were assessed by western blot with quantitative analysis. m Protein expression levels of SIRT6, NRF2 and GPX4 proteins in BMSCs under different treatments were assessed by Immunofluorescence staining with quantitative analysis. Scale bar: 50 μm. n-o Knockdown of ALKBH5 reduced the degradation rate of SIRT6 mRNA, and WT-ALKBH5 facilitated the degradation rate of SIRT6 mRNA but MUT-ALKBH5 had no significant effect. Data are presented as mean ± SD, n = 3 To further validate this regulatory relationship, ALKBH5 was overexpressed in normal BMSCs. Western blot revealed a pronounced reduction in NRF2 and its downstream antioxidant proteins upon ALKBH5 overexpression, which was largely reversed by concomitant SIRT6 overexpression (Fig. 4f). Mechanistically, ubiquitination assays showed that ALKBH5 overexpression significantly increased NRF2 ubiquitination, whereas reintroduction of SIRT6 abolished this effect, indicating that ALKBH5 suppresses NRF2 stability by impairing SIRT6-mediated protection against ubiquitin-dependent degradation (Fig. S19). Collectively, these data suggest that ALKBH5 inhibits intracellular antioxidant defense by disrupting SIRT6–NRF2 interaction and promoting NRF2 ubiquitination. To determine whether this pathway functionally contributes to ferroptosis regulation, SIRT6 was further silenced in ALKBH5-deficient BMSCs. Lipid peroxidation and ferrous iron staining demonstrated that SIRT6 knockdown significantly reversed the ferroptosis-suppressive effects conferred by ALKBH5 silencing, and the immunofluorescence of NRF2 further confirmed this (Fig. 4g-i). Consistently, qRT-PCR and protein analyses revealed that the ALKBH5 knockdown–induced suppression of the ferroptosis marker ACSL4 and restoration of GPX4 expression were abrogated upon SIRT6 depletion (Fig. 4j-k). Moreover, immunofluorescence staining showed that SIRT6 silencing reversed the inhibitory effects of ALKBH5 knockdown on senescence-associated markers P53 and γH2AX (Fig. S20), indicating that SIRT6 is indispensable for ALKBH5-mediated regulation of ferroptosis and cellular aging. To assess whether these effects depend on the m6A demethylase activity of ALKBH5, a catalytically inactive mutant ALKBH5 (H204A) was introduced. Unlike wild-type ALKBH5, the mutant form exerted no significant effect on the SIRT6–NRF2 axis (Fig. 4l-m). Subsequently, ferroptosis was further induced using erastin. The results showed that ALKBH5 overexpression markedly increased intracellular ferrous iron accumulation, whereas overexpression of the ALKBH5-MUT exerted no significant effect. Consistently, immunofluorescence staining revealed that ALKBH5 overexpression further suppressed the SIRT6–NRF2 axis, while this effect was abolished in cells expressing ALKBH5-MUT, thereby corroborating the m6A-dependent regulatory role of ALKBH5 in ferroptosis (Fig. S21). RNA stability assays further revealed that ALKBH5 knockdown significantly prolonged the half-life of SIRT6 mRNA, whereas overexpression of wild-type, but not mutant, ALKBH5 accelerated SIRT6 mRNA degradation (Fig. 4n-o). Together, these results demonstrate that ALKBH5 suppresses SIRT6 expression by reducing SIRT6 mRNA stability through site-specific m6A demethylation, thereby impairing NRF2-dependent antioxidant defense and promoting ferroptosis. 5. Coordinated regulation of METTL3 and ALKBH5 restores bone regenerative capacity in aging diabetic mice Based on the in vitro findings, we next performed in vivo validation in aged mice. METTL3 was overexpressed in the bone marrow cavity, and a parallel group received the glycolysis inhibitor 2-deoxy-D-glucose (2-DG) to assess glycolysis dependency. Micro-CT analysis demonstrated that METTL3 overexpression significantly increased bone volume fraction and bone mineral density in aged mice, whereas these improvements were largely abrogated by 2-DG administration (Fig. 5a–b). Consistently, immunofluorescence staining showed that METTL3 overexpression markedly reduced the senescence marker P53 and enhanced the osteogenic marker BMP2, both of which were reversed upon glycolysis inhibition (Fig. 5c–d). These findings indicate that METTL3 promotes bone regeneration in aged mice through activation of glycolytic metabolism. qRT–PCR analysis of BMSCs further confirmed that METTL3 overexpression significantly upregulated key glycolytic enzymes GLUT1 and LDHA, while 2-DG treatment suppressed their expression (Fig. 5e). Notably, the expression of ferroptosis-related genes ACSL4 and GPX4 remained unchanged regardless of METTL3 overexpression or glycolysis inhibition (Fig. 5f), supporting that METTL3-mediated osteogenic enhancement in aged mice occurs independently of ferroptosis regulation. Coordinated regulation of METTL3 and ALKBH5 restores bone regenerative capacity in aging diabetic mice. a-b representative micro–CT images of the proximal tibia and quantitative analysis of bone volume fraction (BV/TV) and bone mineral density (BMD) in aging mice under different treatments. Scale bar: 1 mm. c-d representative immunofluorescence images and quantitative analysis of P53 and BMP2 expression in bone tissue. Scale bar: 100 μm. e qRT–PCR analysis of GLUT1 and LDHA expression in BMSCs isolated from each treatment group. f qRT–PCR analysis of GPX4 and ACSL4 expression in BMSCs isolated from each treatment group. g-h representative micro–CT images of the proximal tibia and quantitative analysis of bone volume fraction (BV/TV) and bone mineral density (BMD) in aging combined with diabetes mice under different treatments. Scale bar: 1 mm. i representative immunofluorescence images and quantitative analysis of P53 and BMP2 expression in bone tissue. Scale bar: 100 μm. j qRT–PCR analysis of GLUT1 and LDHA expression in BMSCs isolated from each treatment group. k qRT–PCR analysis of GPX4 and ACSL4 expression in BMSCs isolated from each treatment group. Data are presented as mean ± SD, n ≥ 3 Furthermore, we extended our in vivo validation to mice with combined aging and diabetes. Micro-CT analysis revealed that METTL3 overexpression alone further exacerbated bone loss, as indicated by a significant reduction in bone volume fraction and bone mineral density. In contrast, ALKBH5 knockdown partially restored these parameters. Notably, simultaneous METTL3 overexpression and ALKBH5 silencing resulted in a robust recovery of bone volume fraction and bone mineral density (Fig. 5g–h). Consistently, immunofluorescence staining demonstrated that METTL3 overexpression further elevated the senescence marker P53 and suppressed the osteogenic marker BMP2. ALKBH5 knockdown partially attenuated P53 expression and restored BMP2 levels, whereas combined regulation produced a more pronounced suppression of P53 and enhancement of BMP2 expression (Fig. 5i, Fig. S22). Mechanistically, although METTL3 overexpression alone activated glycolytic metabolism in BMSCs, it concurrently aggravated ferroptosis under diabetic conditions. In contrast, ALKBH5 knockdown markedly suppressed ferroptosis, accompanied by a modest compensatory activation of glycolysis. Notably, coordinated regulation of METTL3 and ALKBH5 simultaneously inhibited ferroptosis and robustly enhanced glycolytic flux (Fig. 5j–k), providing a metabolic basis for the synergistic improvement in bone regeneration observed in vivo. 6. Milk-derived exosomes display robust ferroptosis resistance Given their intrinsic stability, biocompatibility, and antioxidant-enriched cargo [42], milk-derived exosomes (MiExos) were investigated for their capacity to protect BMSCs from ferroptosis under diabetic stress. MiExos were isolated using a combination of acid precipitation and ultracentrifugation, yielding freeze-dried white powders. Upon resuspension in aqueous solution, MiExos formed a transparent suspension exhibiting the tyndall effect, indicating good dispersion (Fig. 6a). Transmission electron microscopy (TEM) revealed typical double-membrane vesicular structures, consistent with canonical extracellular vesicle morphology (Fig. 6b). Nanoparticle tracking analysis showed an average particle size of 129.9 nm (Fig. 6c), and zeta potential measurement indicated a value of −16.2 ± 0.8 mV (Fig. 6d). We further evaluated the gastrointestinal stability of MiExos. TEM confirmed that the exosomes maintained an intact bilayer membrane structure in simulated gastric fluid (SGF) and simulated intestinal fluid (SIF), with particle size variations of less than 10% throughout the digestion process (Fig. 6e-f). Milk-derived exosomes display robust ferroptosis resistance. a Images of re-dissolved MiExos after freeze-drying and laser irradiation. b Transmission electron microscope (TEM) image of MiExos. Scale bar: 50 nm. c Histogram showing the particle size distribution of MiExos. d Zeta potential of MiExos. e-f TEM images and size analysis of MiExos after incubation in PBS, SGF, and SIF. Scale bar: 50 nm. g Representative fluorescence images of BMSCs incubated with PKH67-labeled MiExos for various time points. Scale bar: 10 μm. h-i Representative fluorescence images of BMSCs incubated with different concentrations of PKH67-labeled MiExos for separately pre-treated with 4 °C or different endocytosis-related inhibitors and quantitative analysis. Scale bar: 10 μm. j-k Representative fluorescence staining images of Fe2+ within BMSCs and quantitative analysis. Scale bar: 50 μm. l qRT–PCR analysis of TFRC, PTGS2, GPX4 and ACSL4 expression in BMSCs under different treatments. m qRT–PCR analysis of P53, P21, PCNA and CCNA2 expression in BMSCs under different treatments. n Representative images of P53 and γH2AX immunofluorescence staining of BMSCs in different treatment groups. Scale bar: 50 μm. o Representative images of RUNX2 and COL1A1 immunofluorescence staining of BMSCs in different treatment groups. Scale bar: 50 μm. Data are presented as mean ± SD, n = 3 Prior to functional evaluation, the biocompatibility and cellular uptake of MiExos were assessed. CCK-8 and live/dead assays demonstrated no significant cytotoxicity in BMSCs or HUVECs even at high concentrations (400 µg/mL), indicating excellent cell compatibility (Fig. S23a-d). Hemolysis assays confirmed that MiExos exhibited negligible blood toxicity, with hemolysis rates consistently below 5% (Fig. S23e). Fluorescent labeling with PKH67 further revealed that BMSCs efficiently internalized MiExos in a time-dependent manner (Fig. 6g). Mechanistic studies using specific endocytosis inhibitors showed that chlorpromazine, a clathrin-mediated endocytosis inhibitor, reduced MiExos uptake by ~80%, whereas nystatin and EIPA had small effects, suggesting that clathrin-dependent endocytosis is the primary internalization route. Uptake was also significantly inhibited at low temperatures, supporting a contribution of membrane fusion in MiExos internalization (Fig. 6h-i). To investigate the anti-ferroptotic potential of MiExos, intracellular Fe2+ levels and ferroptosis-related gene expression were evaluated. Treatment with MiExos significantly reduced intracellular Fe2+ accumulation and suppressed pro-ferroptotic genes, including ACSL4, TFRC, and PTGS2, while upregulating the anti-ferroptotic gene GPX4 (Fig. 6j-l, Fig. S24a). Immunofluorescence analysis of the SIRT6–NRF2–HO-1 axis revealed that MiExos did not activate SIRT6 but significantly enhanced NRF2 and HO-1 expression (Fig. S24b). Co-localization analysis further demonstrated that MiExos inhibited NRF2 ubiquitination (Fig. S24c), suggesting stabilization of NRF2 as a key mechanism for ferroptosis resistance. Functionally, MiExos rescued diabetes-induced senescence and osteogenic impairment in BMSCs. qRT-PCR analysis showed significant downregulation of senescence markers P53 and P21, and upregulation of proliferation-related genes PCNA and CCNA2 (Fig. 6m). Immunofluorescence confirmed reduced P53 and γH2AX signals (Fig. 6n, Fig. S25a). Osteogenic markers RUNX2 and COL1A1 were significantly restored (Fig. 6o, Fig. S25b), and Alizarin Red S staining indicated improved mineralization capacity (Fig. S26). Notably, MiExos exerted a negligible impact on glycolytic activity, as evidenced by the stable levels of glucose uptake, NAD+/NADH ratio, and lactate levels, as well as the unchanged expression of key glycolytic markers GLUT1 and LDHA (Fig. S27). Collectively, these results demonstrate that MiExos exhibit excellent biocompatibility, are efficiently internalized by target cells, and confer robust anti-ferroptotic and pro-regenerative effects in BMSCs. 7. Milk-derived exosomes coordinate macrophage polarization and ferroptosis Macrophage polarization critically regulates the intramedullary oxidative microenvironment [43]. We therefore investigated the effects of milk-derived exosomes on macrophage phenotypic modulation. LPS stimulation markedly increased the expression of M1-associated marker-CCR7, which was further amplified under HGPA conditions, whereas treatment with MiExos significantly attenuated CCR7 expression in a concentration-dependent manner (Fig. 7a). Conversely, IL-4 induced robust upregulation of the M2 marker CD206, which was markedly suppressed by HGPA exposure; this inhibitory effect was effectively reversed by MiExos treatment (Fig. 7b). Consistently, qRT–PCR analysis demonstrated that pro-inflammatory cytokines, including TNF-α, IL-1β, and IL-6, were significantly reduced following MiExos administration compared with the HGPA group, while anti-inflammatory mediators such as IL-1RA and IL-10 exhibited a progressive increase (Fig. 7c). Milk-derived exosomes coordinate macrophage polarization and ferroptosis. a-b Representative immunofluorescence images and quantitative analysis of macrophage polarization markers CCR7 (M1) and CD206 (M2) under different treatments. 200/400: 200/400 µg/mL MiExos. Scale bar: 50 μm. c qRT–PCR analysis of pro-inflammatory cytokines (IL-1β, IL-6, TNF-α) and anti-inflammatory cytokines (IL-1RA, IL-10) in macrophages following different treatments. d Representative images of DHE staining showing intracellular ROS levels in macrophages under different conditions. Scale bar: 50 μm. e qRT–PCR analysis of ferroptosis-related genes TFRC, PTGS2, ACSL4, and GPX4 in macrophages subjected to different treatments. f Western blot analysis and corresponding quantification of ACSL4 and GPX4 protein expression in macrophages under different treatments. g Representative immunofluorescence images and quantitative analysis of NRF2 expression in macrophages. Scale bar: 50 μm. h Images of BMSCs cultured with conditioned media from differently treated macrophages (Created with BioGDP.com). i Representative immunofluorescence images and quantitative analysis of senescence markers P53 and γH2AX in BMSCs following macrophage-conditioned treatments. Scale bar: 50 μm. j Alizarin Red S staining of BMSCs after osteogenic induction under different treatment conditions and quantitative analysis. Data are presented as mean ± SD, n = 3 Given the pivotal role of reactive oxygen species (ROS) in sustaining M1 polarization and inflammatory signaling, intracellular ROS levels were subsequently assessed [44]. DHE staining showed MiExos markedly suppressed ROS accumulation in macrophages (Fig. 7d). As ferroptosis represents a key downstream oxidative process within the diabetic microenvironment, we further examined ferroptosis-related gene expression. HGPA exposure induced a dysregulated ferroptotic profile, characterized by elevated expression of pro-ferroptotic genes PTGS2, TFRC, and ACSL4, accompanied by downregulation of the anti-ferroptotic factor GPX4, whereas MiExos treatment significantly reversed these alterations (Fig. 7e). Western blot analysis of ACSL4 and GPX4 proteins corroborated these transcriptional findings (Fig. 7f). Moreover, MiExos markedly enhanced the expression of the redox-regulatory transcription factor NRF2 in macrophages (Fig. 7g), suggesting activation of antioxidant defense pathways. To evaluate the functional interplay between macrophages and BMSCs, conditioned media from differently treated macrophages were applied to BMSCs (Fig. 7h). Immunofluorescence staining of P53 and γH2AX further confirmed the attenuation of cellular senescence (Fig. 7i). Upon osteogenic induction, conditioned media from MiExos–treated macrophages markedly restored osteogenic capacity, as evidenced by increased expression of RUNX2 and COL1A1 (Fig. S28) and enhanced matrix mineralization detected by Alizarin Red S staining (Fig. 7j). Collectively, these findings demonstrate that MiExos alleviate oxidative stress–driven macrophage polarization by suppressing ferroptosis, thereby creating a pro-regenerative microenvironment that protects BMSCs from senescence and restores osteogenic potential. 8. Milk-derived exosomes synergize with anaerobic exercise to promote bone regeneration in aging–diabetes comorbidity The above findings demonstrated the potent anti-ferroptotic properties of MiExos. To further enhance bone regeneration, we sought to integrate glycolytic activation into this therapeutic strategy. Extensive previous studies have established intense anaerobic exercise as an effective approach to stimulate glycolytic metabolism within bone tissue [33, 45]. Accordingly, mice were administered MiExos orally via drinking water on a daily basis and subjected to a structured anaerobic exercise regimen. After 12 weeks of intervention, tibiae were harvested for micro-CT analysis and histological evaluation (Fig. 8a). Milk-derived exosomes synergize with anaerobic exercise to promote bone regeneration in aging–diabetes comorbidity. a Experimental design of the anaerobic exercise (EX) model mice treated with MiExos by oral administration. b Representative micro–CT images of the proximal tibia and quantitative analysis of BV/TV and BMD, Tb.Sp and Tb.N. Scale bar: 1 mm. d Representative HE staining of tibial sections. Scale bar: 500 μm. e-f Representative immunofluorescence images and quantitative analysis of P53 and BMP2 expression in bone tissue. Scale bar: 100 μm. g-h Representative immunofluorescence images and quantitative analysis of GPX4 and ACSL4 expression in bone tissue. Scale bar: 100 μm. i-j Representative immunofluorescence images and quantitative analysis of LDHA expression in bone tissue. Scale bar: 100 μm. Data are presented as mean ± SD, n = 6 Micro-CT reconstruction revealed that administration of MiExos alone moderately increased bone volume fraction and bone mineral density compared with diabetic-aged controls, accompanied by partial restoration of trabecular architecture; Notably, the combination of MiExos with anaerobic exercise resulted in a marked enhancement of bone volume fraction and bone mineral density, with trabecular integrity nearly restored to control levels (Fig. 8b-c). These findings were further corroborated by HE staining (Fig. 8d). Immunofluorescence analyses demonstrated that MiExos treatment alone partially suppressed the senescence marker P53 and moderately increased the osteogenic marker BMP2, whereas the combined intervention produced a more pronounced reduction in P53 expression and robust upregulation of BMP2 (Fig. 8e-f). Assessment of ferroptosis-related markers revealed that MiExos significantly decreased ACSL4 expression and increased GPX4 levels, indicating effective ferroptosis inhibition, while the addition of anaerobic exercise exerted minimal further influence on GPX4 and ACSL4 expression (Fig. 8g-h). In contrast, evaluation of the glycolytic marker LDHA showed that MiExos partially restored glycolytic capacity, which was substantially amplified following anaerobic exercise training (Fig. 8i-j). Moreover, analysis of the intramedullary immune microenvironment demonstrated that the combined strategy markedly reduced M1 macrophage polarization while promoting M2 macrophage differentiation, suggesting an alleviation of inflammatory stress within bone tissue (Fig. S29). Importantly, histopathological examination of major organs revealed preserved tissue architecture without evident inflammatory infiltration or structural damage following MiExos administration (Fig. 30a). Serum biochemical analyses further confirmed the absence of significant systemic toxicity (Fig. 30b). Collectively, these results indicate that orally administered MiExos exhibit favorable biosafety and, when combined with glycolysis activation via anaerobic exercise, synergistically promote bone regeneration in the context of diabetes–aging comorbidity, highlighting a promising and translational therapeutic strategy. Discussion This study systematically elucidated the multi-level molecular basis underlying bone regeneration impairment induced by the combined effects of diabetes and aging, and proposes a novel therapeutic paradigm centered on the precise coordinated regulation of metabolic homeostasis and cell death fate. Unlike previous studies that primarily attributed diabetes-associated bone damage to chronic inflammation, oxidative stress accumulation, or dysregulation of individual signaling pathways, our findings demonstrate that the progressive decline in regenerative capacity fundamentally arises from the collapse of intracellular homeostatic regulatory networks within BMSCs. This dysfunction is particularly characterized by the m6A epigenetic imbalance-driven functional decoupling between glycolytic metabolic programs and ferroptosis pathways. In recent years, ferroptosis has been increasingly recognized as a critical contributor to diabetes-related tissue injury and has been implicated in the progression of liver [46], cardiovascular [47], neural [48], renal [49], and skeletal disorders [50]. Accumulating evidence indicates that the diabetic microenvironment induces ferroptosis in stem cells and bone-associated cells, and that pharmacological inhibition of ferroptosis can ameliorate impaired bone regeneration [7, 51, 52]. Meanwhile, metabolic reprogramming—especially the age-associated decline in glycolytic activity—has been identified as a fundamental driver of stem cell dysfunction [53, 54], with glycolytic activation markedly enhancing osteogenic potential [8, 55]. However, these pathological processes have largely been investigated in isolation, and their integrated regulatory mechanisms have remained poorly understood. Here, we reveal for the first time, from an age-dependent perspective, that diabetes induces distinct yet synergistic patterns of m6A regulatory imbalance in BMSCs across different aging stages. In young BMSCs, the diabetic microenvironment predominantly activates ALKBH5-mediated demethylation, attenuating the SIRT6–NRF2 signaling axis, thereby impairing antioxidant defense and markedly increasing ferroptosis susceptibility. During this phase, METTL3 expression remains relatively preserved, enabling rapid restoration of glycolytic metabolism upon ferroptosis inhibition and effective recovery of osteogenic function. In contrast, aging BMSCs exhibit not only sustained ferroptosis activation but also pronounced METTL3 dysfunction, resulting in reduced m6A modification of key glycolytic genes and constrained metabolic flux. The ensuing energy insufficiency compromises the high metabolic demands of osteogenic differentiation, explaining why ferroptosis inhibition alone fails to fully rescue the regenerative capacity of aged stem cells. The m6A modification system emerges as a central “homeostatic gatekeeper” orchestrating this regulatory network. Extensive evidence has demonstrated that METTL3-mediated m6A methylation enhances the expression of critical glycolytic enzymes and glucose transporters [56, 57]. Conversely, ALKBH5-mediated demethylation participates in oxidative stress responses and ferroptosis sensitivity under pathological stimuli. Previous studies have shown that ALKBH5 suppress SLC7A11 and GPX4 expression, directly facilitating ferroptosis [58, 59], and modulates antioxidant capacity via NRF2 signaling [60]. SIRT6, a key regulator of metabolic homeostasis and redox balance, closely interacts with NRF2 to regulate its stability and transcriptional activity [61, 62]. Our data further demonstrate that ALKBH5 destabilizes SIRT6 mRNA, leading to its downregulation and subsequent collapse of antioxidant defense, ultimately driving uncontrolled ferroptosis. Collectively, these findings indicate that bone regeneration impairment induced by diabetes and aging represents a coordinated failure of energy metabolism and redox homeostasis governed by m6A dysregulation. In this pathological context, ferroptosis inhibition alone alleviates oxidative damage but cannot restore metabolic support, whereas isolated glycolytic activation may exacerbate lipid peroxidation under high oxidative stress conditions. This systemic imbalance fundamentally limits the efficacy of single-target interventions. Milk-derived exosomes (MiExos) have recently attracted significant interest due to their outstanding biocompatibility, gastrointestinal stability, and broad biological activities in inflammation modulation, metabolic regulation, and oxidative stress alleviation [23, 42]. However, their roles in bone marrow homeostasis and ferroptosis regulation remain largely unexplored. Our study provides the first evidence that milk-derived exosomes enhance antioxidant defense by inhibiting NRF2 ubiquitination-mediated degradation, thereby effectively suppressing ferroptosis and significantly improving diabetic microenvironment-induced BMSCs senescence and osteogenic dysfunction. Meanwhile, abundant evidence indicates that anaerobic exercise serves as a potent physiological activator of glycolysis, markedly enhancing metabolic flux in bone tissue through upregulation of key regulators such as PKM and LDHA, and promoting osteogenesis [45, 63, 64]. In aging and metabolic disease contexts, exercise-induced glycolytic enhancement is regarded as a critical metabolic foundation for restoring regenerative capacity. Based on the mechanistic understanding of metabolic–ferroptosis uncoupling, we developed a precise homeostasis reconstruction strategy integrating milk exosome therapy with anaerobic exercise. Milk exosomes primarily restore redox homeostasis and inhibit ferroptosis, while anaerobic exercise replenishes metabolic energy supply. These interventions target complementary pathological nodes driven by m6A imbalance, forming a highly synergistic molecular mechanism. The combined strategy markedly outperformed single interventions, nearly completely reversing bone loss and microarchitectural deterioration in diabetic aging mice. Furthermore, we observed that milk exosomes modulate macrophage polarization, reduce ROS accumulation, and improve the bone marrow immune microenvironment, further supporting BMSC functional recovery. These findings underscore that bone regeneration impairment is a systemic disorder driven by metabolic dysregulation, aberrant cell death, and immune microenvironment remodeling. Although our study highlights the ALKBH5/METTL3-mediated intrinsic dysfunction of BMSCs, it is important to acknowledge that in vivo bone regeneration is a coordinated process involving the neuro-vascular-immune niche. Future studies utilizing cell-specific ALKBH5/METTL3 knockout mice would further delineate the relative contribution of BMSC-specific m6A signaling. Furthermore, moving from the elucidation of molecular mechanisms to clinical translation requires addressing the inherent limitations of experimental scaling. While our study provides a solid mechanistic foundation using a murine model, we recognize that mouse bone metabolism and mechanical loading differ from those of humans. Future investigations using large animal models, such as porcine or ovine diabetic models, will be crucial to evaluate the long-term safety, optimal dosage, and biomechanical outcomes of the MiExos-exercise intervention before clinical application. In conclusion, this study identifies m6A imbalance-mediated metabolic–ferroptosis uncoupling as the central mechanism underlying bone regeneration failure induced by diabetes and aging, and proposes a precise homeostasis reconstruction strategy as a novel therapeutic avenue. This work provides a systemic theoretical framework and translational foundation for treating complex metabolic-associated bone degenerative diseases. Conclusion This study elucidates the key molecular basis underlying bone regeneration impairment under the combined effects of diabetes and aging, identifying m6A modification as a central “gatekeeper” coordinating metabolic homeostasis and cell death fate. The diabetic microenvironment primarily triggers ALKBH5 mediated ferroptosis activation, while aging further imposes METTL3-dependent glycolytic suppression, jointly disrupting metabolic supply and redox balance and establishing a pathological state highly resistant to single-target interventions. Building on this mechanistic framework, we demonstrate that milk-derived exosomes (MiExos), as natural, safe, and highly stable biological nanocarriers, effectively suppress ferroptosis and remodel the bone marrow redox microenvironment, thereby alleviating diabetes-induced cellular senescence and osteogenic dysfunction. Notably, the combination of exosome therapy with anaerobic exercise–mediated metabolic activation enabled coordinated regulation of ferroptosis and energy metabolism, markedly enhancing bone regeneration in diabetic aged mice. This dual-modality strategy overcomes the limitations of conventional single-target therapies in complex metabolic disorders and highlights the translational potential of exosome-based homeostasis reconstruction for regenerative medicine, offering a promising therapeutic avenue for aging- and diabetes-associated bone degenerative diseases. Supplementary Information Supplementary Material 1. Author contributions Zengguang Wang: Writing – original draft, Methodology, Investigation, Data curation, Conceptualization. Project administration. Haiyang Lin: Writing – original draft, Methodology, Investigation, Conceptualization. Project administration. Yuntao Li: original draft, Methodology, Investigation, Conceptualization. Project administration. Yiming Li: Visualization, Formal analysis, Software. Chen Xu: Writing – review & editing, Supervision, Resources, Validation. Yaokai Gan: Writing – review & editing, Supervision, Resources, Validation. Funding acquisition. Funding This work was funded by the National Key Research and Development Program of China (2023YFC2414102). Data availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Zengguang Wang, Haiyang Lin and Yuntao Li contributed equally to this work. Contributor Information Chen Xu, Email: 116126@9hospital.org.cn. Yaokai Gan, Email: ganyk1137@sh9hospital.org.cn. References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The data that support the findings of this study are available from the corresponding author upon reasonable request.