Exercise suppresses IP6K3 to modulate BCAA metabolism and ferroptosis in MASLD

Abstract

Background

Metabolic dysfunction-associated steatotic liver disease (MASLD) is linked to branched-chain amino acid (BCAA) dysmetabolism and ferroptosis, and exercise is considered protective, yet the underlying mechanisms remain unclear.

Methods

Integrated bioinformatic analyses of public datasets were performed to identify key regulators of MASLD. Hepatic inositol hexakisphosphate kinase 3 (IP6K3) expression was examined in liver samples from patients with MASLD as well as mice. The functional role of IP6K3 was assessed by either hepatocyte-specific Ip6k3 deletion or exercise intervention in mice. Single-cell RNA sequencing (scRNA-seq), IP6K3 overexpression and knockdown, liquid chromatography–tandem mass spectrometry, and co-immunoprecipitation were used to explore the underlying mechanism by which IP6K3 regulates MASLD.

Results

Hepatic IP6K3 was significantly upregulated in both patients with MASLD as well as mice, and exercise markedly attenuated hepatic IP6K3 expression in MASLD mice. Hepatocyte-specific Ip6k3 deletion conferred resistance to MASLD-induced liver injury. scRNA-seq revealed that these protective effects were related to enhanced BCAA degradation and inhibition of ferroptosis. Further studies showed that BCAA degradation was dependent on branched-chain aminotransferase 2 (BCAT2). In vitro, IP6K3 overexpression exacerbated BCAA-promoted ferroptosis, which was rescued by ferrostatin-1. Conversely, IP6K3 knockdown prevented ferroptosis, and this effect was abolished by inhibiting BCAT2. Mechanistically, elevated IP6K3 in MASLD could bind to heterogeneous nuclear ribonucleoprotein K (HNRNPK) in hepatocytes. Exercise-downregulated IP6K3 facilitated HNRNPK release, thereby stabilizing BCAT2 mRNA, promoting BCAA catabolism, and ultimately preventing ferroptosis in MASLD.

Conclusions

Our study identifies IP6K3 as a key molecule in exercise-induced protection against MASLD, and suggests that IP6K3 inhibition might represent a potential pharmacological strategy for MASLD intervention.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s11658-026-00981-1.

Introduction

Metabolic dysfunction-associated steatotic liver disease (MASLD, formerly NAFLD), has emerged as a leading cause of end-stage liver disease worldwide [1, 2]. MASLD encompasses a broad clinicopathological spectrum, initiating as simple steatosis and potentially advancing to metabolic dysfunction-associated steatohepatitis (MASH), progressive fibrosis, and ultimately cirrhosis or hepatocellular carcinoma [3, 4]. Driven by excessive caloric intake and sedentary lifestyles, the global surge in obesity and metabolic syndrome has established a reciprocal pathogenic link between MASLD and systemic disturbances such as hyperglycemia, insulin resistance, visceral adiposity, diabetes, and dysregulated adipokine signaling [3, 5]. Given that abnormalities in energy metabolism underpin the entire trajectory of MASLD, identifying novel metabolic regulators remains a priority for the development of targeted therapies.

While hepatic lipid accumulation and chronic inflammation are defining hallmarks of MASLD [6], emerging evidence highlights branched-chain amino acid (BCAA: leucine, isoleucine, and valine) catabolic dysfunction and ferroptosis as critical drivers of disease progression [7–12]. Elevated circulating BCAA levels are strongly associated with obesity, insulin resistance, and metabolic dysfunction in rodents [9, 13]. Branched-chain amino acid transaminase 2 (BCAT2) serves as a rate-limiting enzyme initiating mitochondrial BCAA catabolism [14]. Recent studies suggest that BCAA accumulation may trigger neuronal lipid peroxidation and ferroptosis during cerebral ischemia–reperfusion injury [15], and contribute to heart failure [16, 17]. In the liver, impaired BCAA degradation is linked to obesity, diabetes, and the exacerbation of lipotoxicity and hepatic carcinoma via inhibition of autophagy [7–9, 18, 19]. Shin and colleagues found that hepatic branched-chain α keto-acid dehydrogenase, an enzyme in the BCAA degradation pathway, was reduced in obese and diabetic humans as well as high-fat-fed macaques [20]. BCAA were elevated in patients with non-alcoholic steatohepatitis [21], and BCAA could exacerbate hepatic lipotoxicity by reducing lipogenesis and inhibiting autophagy in mice [19]. Collectively, these findings suggest that the intersection of BCAA metabolism and ferroptosis represents a vital axis in MASLD pathogenesis.

Although recent years have witnessed progress in the development of pharmacotherapies for MASH, treatment options applicable to the broader patient population remain limited overall, and current clinical guidelines emphasize lifestyle interventions, specifically caloric restriction and structured physical activity [22]. Meta-analyses indicate that an adequate exercise dose (> 750 metabolic equivalents/week) is significantly more effective at reducing liver fat than standard care [23]. While exercise is known to improve glucose homeostasis and insulin sensitivity in patients with diabetes and obese mice, partially through the modulation of BCAA catabolism [24, 25], the precise molecular transducers through which exercise promotes BCAA degradation in the context of MASLD remain poorly understood.

Inositol hexakisphosphate kinase 3 (IP6K3) belongs to the inositol hexaphosphate kinase (IP6K) family, which governs the biosynthesis of inositol pyrophosphates. In eukaryotic cells, inositol pyrophosphates play critical roles in cellular development and biological processes as energy regulators, phosphodonors, steric/allosteric regulators, and G protein-coupled receptor messengers [26]. IP6Ks, rate-limiting enzymes in inositol pyrophosphate synthesis, can catalyze the addition of a phosphate group to the first, third, or fifth position of the inositol ring [27]. Through this process, IP6Ks regulate diverse cellular processes, including cell death and immune regulation [28, 29]. IP6K1 knockout mice were resistant to high-fat diet-induced obesity, cardiac ischemia–reperfusion injury, and type 2 diabetes [30–33]. IP6K2, potentially collaborated with IP6K1, plays a role in maintaining phosphate balance and energy regulation in mammals. Deletion of Ip6k1 and Ip6k2 in HCT116 cells (a human colon cancer cell line) could enhance adenosine triphosphate (ATP) levels as well as free intracellular phosphate [34]. Moritoh and colleagues found that Ip6k3 is highly expressed in muscle and that its deletion induces significant metabolic changes and impacts lifespan [35]. However, the physiological role of IP6K3 in the liver, especially its contribution to the progression of MASLD, remains largely undefined.

By integrating public transcriptomic datasets and analyzing liver biopsies from patients with MASLD and murine models, we identified a significant upregulation of hepatic IP6K3, which was notably reversed following bariatric surgery. We hypothesized that exercise suppresses IP6K3 expression to modulate BCAA metabolism and ferroptosis in MASLD. Our study demonstrated that hepatocyte-specific Ip6k3 deletion mimicked the protective effects of exercise against MASLD. Mechanistically, we showed that IP6K3 bound to HNRNPK, thereby destabilizing BCAT2 mRNA. This reduction in BCAT2 expression triggered BCAA accumulation and subsequent ferroptosis. Conversely, exercise-induced suppression of IP6K3 restored BCAA catabolism and inhibited ferroptotic cell death, providing a novel therapeutic target for the management of MASLD.

Materials and methods

For key resources (antibodies, biological samples, chemicals, commercial assays, cell lines, oligonucleotides, plasmids, and siRNA, etc.) used in this work, please see Supplementary Table S1.

Animal

Age-matched 6-week-old C57BL/6J (WT) mice, purchased from Shanghai SLAC Laboratory Animal Co. Ltd. (Shanghai, China), were used. Alb-Cre mice (B6.FVB(129)-Tg(Alb1-cre)1Dlr/J, #1016832) were purchased from the Jackson Laboratory. An Ip6k3^flox/flox mouse strain was generated by Cyagen Biosciences (Guangzhou, China) using CRISPR/Cas9 technology: two pairs of sgRNA sequences targeting exon 3 of the Ip6k3 gene were designed using the CRISPR/Cas9 system; Cas9 mRNA and gRNA were synthesized through in vitro transcription and microinjected into fertilized eggs of C57BL/6J mice to generate F0 mice (Ip6k3^flox/flox). Positive F0 mice were crossed with wild-type mice to obtain F1 heterozygous (Ip6k3^flox/wt) offspring. F1 mice were intercrossed to generate homozygous floxed F2 mice (Ip6k3^flox/flox), which were then bred with Alb-Cre mice to obtain Ip6k3^flox/flox Alb-Cre mice (hepatocyte-specific Ip6k3 deletion, Ip6k3^fl/fl Alb-Cre). All mice were bred and housed in a temperature-controlled environment (23 ± 2 °C) with free access to water and chow, under a 12 h/12 h light/dark cycle, and all efforts were made to minimize their use and discomfort. Daily monitoring included body weight, behavior, and responses to stimuli. All experimental procedures were approved by the Institutional Animal Care and Use Committee of Shanghai Tenth People’s Hospital of Tongji University (approval no. SHDSYY-2022-3019).

Transcriptomic analysis and pathway enrichment

Publicly available liver transcriptomic datasets related to non-alcoholic steatohepatitis (NASH) and bariatric intervention were obtained from the Gene Expression Omnibus (GEO) database, including GSE89632 (NASH cohort) and GSE48452 (bariatric intervention cohort). Differential expression analysis was performed independently in each dataset. Genes with an adjusted P value < 0.05 and |log₂ fold change|> 1 were defined as differentially expressed.

Genome-wide differential expression patterns were visualized using volcano plots. To identify genes exhibiting opposite regulatory patterns between NASH and bariatric intervention, log₂ fold changes from the two datasets were integrated and compared. Genes upregulated in NASH but downregulated following bariatric intervention, as well as genes showing the reverse pattern, were identified and visualized using scatter plots and intersection analysis, enabling the systematic identification of candidate genes potentially reversed by bariatric intervention.

To explore the biological relevance of these oppositely regulated genes, pathway enrichment analyses were conducted using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database. Enrichment analysis was performed separately for overlapping gene sets derived from the NASH and bariatric intervention cohorts. In addition, gene set enrichment analysis (GSEA) was applied to ranked gene lists from each dataset to evaluate coordinated changes in metabolic pathways, including inositol phosphate metabolism and phosphatidylinositol signaling pathways.

Human liver samples

Human liver samples, procured from patients admitted to Shanghai Tenth People’s Hospital between 2017 and 2021, were obtained from individuals (over 18 years old) who underwent hepatectomy owing to hepatolithiasis, liver trauma, or hepatic hemangioma, as described in our previous study [36]. Liver tissues were isolated from patients with MASLD and patients without MASLD at Shanghai Tenth People’s Hospital. Clinical characteristics of the subjects are presented in Supplementary Table S2 (n = 6 each group). Individuals meeting any of the following criteria were excluded from this study: excessive alcohol consumption (> 140 g for men or > 70 g for women per week), drug or toxin use, viral infection (e.g., acquired immunodeficiency syndrome (AIDS), hepatitis B virus, and hepatitis C virus), coronary heart disease, myocardial infarction, or cerebral apoplexy. The study was approved by the institutional research ethics committee of Shanghai Tenth People’s Hospital of Tongji University (approval no. SHSY-IEC-4.1/19-126/01) and conducted according to the ethical standards of the Helsinki Declaration. Informed consent was obtained from each recruited patient.

Mouse MASLD models and exercise training

The MASLD model was established by feeding 6-week-old mice with a HFHC diet (42% fat and 0.2% cholesterol; TP 26304; TrophicDiet, Nantong, China) for 16 weeks as described previously [37]. Exercise training was initiated after 8 weeks of high-fat and high-cholesterol (HFHC) diet feeding, and the exercise regimen consisted of 8 weeks of treadmill training on a motor-driven treadmill (JD-PT, Jide Biological Instruments, Shanghai, China). During the exercise period, the mice were maintained on the HFHC diet. The treadmill exercise program included 1 week of treadmill familiarization followed by 7 weeks (6 days per week) of treadmill training at 20 m/min for 60 min per day [38, 39].

Cell treatment

The HepG2 (HEPG2) cell line was obtained from the American Type Culture Collection (ATCC, HB-8065). Cells were maintained in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin at 37 °C in a humidified incubator with 5% CO₂. An in vitro MASLD model was created by treating HepG2 cells with 0.3 mM palmitic acid (PA) for 48 h to induce lipid stress [40]. Control groups were cultured in DMEM without PA. To prepare the PA solution, fatty acid-free bovine serum albumin (BSA) powder (Roche, #03117405001) was dissolved in 150 mM NaCl at 37 °C to make a 0.17 mM stock solution. PA powder (Sigma-Aldrich, #P9767) was dissolved in 150 mM NaCl at 70 °C, mixed with the BSA solution at 37 °C for 1 h, filtered through a 0.22-μm filter, and used as a 5 mM PA-BSA stock solution. For experiments related to BCAA, customized BCAA-free DMEM (Boster, China) supplemented with exogenous BCAA (weight ratio of leucine: valine: isoleucine = 2:1:1; HY-N0486, HY-N0717, and HY-N0771, respectively; MedChemExpress) was used to exclude the impact of endogenous BCAA contained in regular culture medium [41]. Since BCAAs are essential amino acids, the control group was cultured in BCAA-free DMEM supplemented with 0.4 mM exogenous BCAA, whereas the treatment group was exposed to BCAA-free DMEM medium containing 6 mM exogenous BCAA [42]. Cells were incubated with BCAA for 24 h prior to sample collection for subsequent analysis. For pharmacological inhibition of ferroptosis during BCAA overload, ferrostatin-1 (Fer-1; HY-100579, MedChemExpress) was used at a final concentration of 2 μM [43]. Fer-1 was added 2 h prior to BCAA stimulation (pre-treatment), followed by co-treatment with BCAA for an additional 24 h before sample collection. To block BCAA catabolism, the BCAT2 inhibitor (BCAT-IN-2, HY-141669, MedChemExpress) was used at a concentration of 20 μM and added to the culture medium simultaneously with BCAA for 24 h [44]. To assess Bcat2 mRNA stability, actinomycin D (HY-17559, MedChemExpress) was used to block transcription. HepG2 cells were treated with actinomycin D at a final concentration of 5 μg/mL [45], and total RNA was harvested at 0, 0.5, 1, 2, 4, and 8 h after actinomycin D addition for subsequent reverse transcription and quantitative polymerase chain reaction (PCR) analyses. All small molecules were dissolved in dimethyl sulfoxide (DMSO), and the final DMSO concentration was kept identical across groups (≤ 0.1%, v/v).

Lipid peroxidation assay using BODIPY 581/591 C11

Lipid peroxidation was assessed using the Lipid Peroxidation Assay Kit with BODIPY 581/591 C11 (Beyotime, S0043M) according to the manufacturer’s instructions with minor modifications. Briefly, BODIPY 581/591 C11 (2 mM stock) was diluted in phosphate-buffered saline (PBS) to prepare a working solution at a final concentration of 2 μM, which was freshly prepared and protected from light prior to use. HepG2 cells cultured in six-well plates were washed once with PBS and incubated with 1 mL of the BODIPY 581/591 C11 working solution at 37 °C for 10 min. After incubation, cells were washed twice with PBS to remove excess dye. Following staining, cells were maintained in PBS and immediately imaged using a fluorescence or confocal microscope. Oxidized BODIPY 581/591 C11 was detected by green fluorescence, whereas the reduced form emitted red fluorescence. Lipid peroxidation was quantified by calculating the ratio of green to red fluorescence intensity using ImageJ software.

Tissue and blood sampling

The mice were anesthetized with phenobarbital sodium (40 mg/kg, i.p.) and euthanized. Tissue and blood samples were collected. Blood was collected and left undisturbed at room temperature for at least 30 min to allow clotting, and then centrifuged at 1000×g for 10 min to obtain serum. The livers were isolated and washed in ice-cold PBS to remove blood. Then the liver tissues were dissected and rapidly frozen in liquid nitrogen for further examination.

Histological analysis

For hematoxylin–eosin (HE) and Masson’s trichrome staining, liver tissues were fixed and embedded in paraffin and sectioned at 4–5 μm thickness. Paraffin sections were subjected to a graded ethanol series for dehydration, followed by staining with HE (Sigma-Aldrich) or Masson’s trichrome solution (Solarbio Life Science, Beijing, China, #G1340) according to standard protocols. Stained slides were visualized under a light microscope (Leica DM2500). For Oil Red O staining, frozen liver tissues were collected and embedded in Jung’s tissue freezing medium (Leica, Nussloch, Germany). Sections (8 μm thick) were cut using a cryostat (Leica), mounted on poly-l-lysine-coated slides, dried overnight at 37 °C, and then stained with Oil Red O (#O0625, Sigma-Aldrich) dissolved in isopropanol. Periodic acid–Schiff (PAS) staining (#1.01646, Sigma-Aldrich) was performed as follows: Paraffin sections were deparaffinized and rehydrated (frozen sections were directly rehydrated in distilled water), oxidized with 0.5% periodic acid for 10 min, rinsed in distilled water for 10 min, immersed in Schiff’s reagent for 10 min, and washed in distilled water for 5 min. Nuclei were counterstained with hematoxylin for 3 min (excessive staining was destained in 1% HCl–alcohol). The sections were dehydrated through an ethanol series, cleared in xylene, and mounted with a coverslip.

MASLD activity score

To assess the severity of MASLD, liver histopathology in mice was systematically evaluated using the MASLD activity score, including systematic evaluation of hepatocellular ballooning, lobular inflammation, and fibrosis [40, 46]. Steatosis was quantified as the percentage of lipid droplet-containing hepatocytes via ImageJ software (NIH), with scores assigned as follows: 0 (< 5%), 1 (5–33%), 2 (> 33–66%), and 3 (> 66%). Hepatocyte ballooning was categorized as 0 (absent), 1 (mild; scattered cells), or 2 (severe; prominent ballooning). Lobular inflammation was scored based on the number of inflammatory foci per 200× field: 0 (none), 1 (< 2 foci), 2 (2–4 foci), or 3 (> 4 foci).

Immunofluorescence

Paraffin-embedded sections were deparaffinized and rehydrated. Antigen retrieval was performed using a standard protocol. Endogenous peroxidase activity was blocked by incubating sections in 3% hydrogen peroxide at room temperature in the dark for 25 min. Nonspecific binding was minimized with 10% goat serum (at room temperature for 30 min). Primary antibodies were applied to the sections and incubated overnight at 4 °C in a humidified chamber. Species-matched horseradish peroxidase (HRP)-conjugated secondary antibodies were then added and incubated for 50 min at room temperature under light-protected conditions. Tyramide Signal Amplification (TSA) reagents were applied and incubated for 20 min at room temperature. Nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) (10 min in the dark), and sections were mounted for fluorescence microscopy imaging (FV1000, Olympus). The antibodies used are listed in Supplementary Table S1.

Immunoblotting

Immunoblotting was performed as described in our previous work [47]. The antibodies used are listed in Supplementary Table S1. The original uncropped Western blot images are provided in the Supplementary Material 2.

Biochemical assays

Serum alanine aminotransferase (ALT) and aspartate aminotransferase (AST) activities were measured with commercial assay kits from Sigma-Aldrich (#MAK052 and #MAK055, respectively, St. Louis, MO). Levels of serum Triglycerides (TG) and total cholesterol (TC) were determined using an automatic biochemistry analyzer (Beckman Coulter, Fullerton, CA) according to the manufacturer’s instructions. Liver reduced glutathione (GSH), nicotinamide adenine dinucleotide (NAD)⁺/NADH, and superoxide dismutase (SOD) activity were measured with commercial assay kits from Beyotime Biotechnology (#S0053, #S0175, and #S01015, respectively, Haimen, China). Serum and liver BCAA level were measured with commercial assay kits from Abcam (#ab83374). The kits used are listed in Supplementary Table S1.

Glucose tolerance and insulin tolerance tests

Glucose tolerance test (GTT) and insulin tolerance test (ITT) were performed after 16 weeks of HFHC treatment. The Mouse Insulin Enzyme-Linked ImmunoSorbent Assay Kit (Beyotime Biotechnology) and a portable glucose meter (Abbot Laboratories) were applied to measure fasting serum insulin levels and blood glucose levels, respectively. After 16 weeks of HFHC feeding, all group of mice were fasted for 14 h or 4 h and then intraperitoneally injected with d-glucose (2.0 g/kg body weight) or insulin (0.75 U/kg body weight) for the GTT or ITT, respectively [48]. Tail vein blood was collected at baseline and at 15, 30, 60, and 120 min post-injection for glucose measurement.

Liquid chromatography–tandem mass spectrometry (LC–MS)

Plasma BCAA concentrations were measured using isotope dilution LC–MS. The plasma samples were pretreated by protein precipitation. LC–MS was performed with an Agilent Zorbax SB-C18 column and a Sciex TripleQuad 4500MD.

RNA quantification

For quantitative PCR analysis, total RNA was extracted using RNAiso Extract Reagent (Takara, Japan). The PCR reaction was performed with a two-step SYBR^® PrimeScript™ RT-PCR Kit (Takara, Japan) in a CFX96 system (Bio-Rad) using specific primers. The cycling conditions were as follows: an initial denaturation at 95 °C for 30 s, followed by 40 cycles of 95 °C for 5 s and 60 °C for 30 s. The primers are listed in Supplementary Table S1. Data were normalized to GAPDH expression.

scRNA-seq data analysis

The analysis and processing of the single‐cell RNA‐sequencing (scRNA‐seq) data were performed using R 4.2.3 (R Core Team, Vienna, Austria) and RStudio software. Doublet removal was conducted by identifying clusters of cells expressing gene expression patterns of two cell types simultaneously. Mitochondrial quality control was implemented by removing cells containing fewer than 200 detected transcripts and cells with more than 20% mitochondrial reads. The data were log-transformed and normalized to transcripts per million (TPM)-like values (scaling factor of 10,000), with subsequent analysis performed using the Seurat package. Following these filtering procedures, 13,784 high-quality cells were retained for further analysis. Principal component analysis (PCA) was conducted to reduce data dimensionality, with the top 20 principal components (PCs) subsequently utilized for uniform manifold approximation and projection (UMAP). Shared nearest neighbors (SNN) clustering was then applied, leading to the identification of 22 distinct clusters in the data. Differential expression analysis across the clusters was performed via the Wilcoxon rank-sum test in the FindAllMarkers function of the Seurat R package. Marker genes for known liver cell populations were used to annotate the clusters. Ten high‐quality cell types were identified, including hepatocytes, hepatic stellate cells, cholangiocytes, endothelial cells, dendritic cells, B cells, natural killer T (NKT) cells, neutrophils, macrophages, and monocytes. Enrichment KEGG categories for significantly expressed markers in each cell cluster were identified by enrichGO from the R package clusterProfiler (version 4.8.1). Cell–cell communication analysis was conducted using the R package CellChat (version 1.6.1) on single-cell and spatial transcriptomics data. Hepatocyte pseudotime trajectories were inferred using Monocle3 (version 1.3.7). After trajectory learning, cells were automatically divided into two independent partitions. The major trajectory (partition 1) was used for primary pseudotime analysis, while the other was analyzed separately and shown in the Supplementary Figures. Functional scores were calculated at the single-cell level and visualized along pseudotime. The raw scRNA-seq data from our study were deposited in the Gene Expression Omnibus (GEO) database under accession number GSE297883. The dataset GSE189600 (normal human liver and NASH human liver) was downloaded from the GEO database.

RNA interference

RNA interference was used to silence IP6K3, HNRNPK, and SYNCRIP in HepG2 cells using siRNA pool (100 nM). The siRNAs were synthesized by Sangon Biotech (Shanghai, China). Lipo8000™ Transfection Reagent (Beyotime Biotechnology) was used to transfect the siRNA pool into HepG2 cells following the manufacturer’s instructions. Knockdown efficiency was validated by immunoblotting or qPCR.

Plasmid constructs and overexpression in cells

Recombinant vectors encoding hemagglutinin (HA)-tagged IP6K3 and Flag-tagged HNRNPK were constructed and subcloned into pcDNA3.1 eukaryotic expression vectors (Invitrogen). Transfection was performed as described above using Lipo8000™ Transfection Reagent. All constructs were confirmed by DNA sequencing.

Nano-HPLC–MS/MS

HA-IP6K3-transfected or vector-control HepG2 cells were treated with PA for 48 h following transfection. The cells were then lysed and immunoprecipitated with a monoclonal anti-HA antibody (Sigma), followed by sodium dodecyl sulfate (SDS) polyacrylamide gel electrophoresis (PAGE) separation to identify IP6K3-associated protein bands through differential analysis. Target gel regions were subjected to in-gel tryptic digestion. Peptides were analyzed by nano-HPLC–MS/MS on an EASY-nanoLC 1200 system connected to a Q Exactive™ Plus mass spectrometer (Thermo Fisher Scientific, MA, USA). Tandem mass spectra were processed by PEAKS Studio version X⁺ (Bioinformatics Solutions Inc., Waterloo, Canada). PEAKS DB was set up to search the UniProt Homo sapiens database (version 2024, 20,608 entries) assuming trypsin as the proteolytic enzyme. Peptides with a 1% false discovery rate (FDR) and proteins with a 1% FDR containing at least one unique peptide, were retained.

Modeling protein structure and interaction with the DeepMind AlphaFold3 algorithm

The assembled hetero-tetrameric structure of IP6K3 (UniProt: Q96PC2) and HNRNPK (UniProt: P61978) was predicted using the AlphaFold 3 algorithm as described previously [49, 50]. The interaction between IP6K3 and HNRNPK and the potential association between amino acids residues were viewed and analyzed using PyMoL software (Schrodinger, New York, NY). The binding affinity of the IP6K3–HNRNPK complex was further estimated using the PRODIGY web server based on the AlphaFold3-predicted structure.

Co-immunoprecipitation (Co-IP)

Cellular lysates were prepared using ice-cold lysis buffer (20 mM Tris pH 7.5, 135 mM NaCl, 2 mM ethylenediaminetetraacetic acid (EDTA), 2 mM dithiothreitol (DTT), 25 mM β-glycerophosphate, 2 mM sodium pyrophosphate, 10% glycerol, 1% Triton X-100, 1 mM sodium orthovanadate, 10 mM NaF, 1 mM phenylmethylsulfonyl fluoride (PMSF)) supplemented with complete protease inhibitor cocktail (Sigma). Following a 10-min centrifugation at 12,000×g (4 °C), supernatants were subjected to overnight immunoprecipitation at 4 °C with 0.2 µg anti-HA or anti-Flag antibodies. Immune complexes were captured by incubation with pre-cleared Protein A/G Plus-agarose beads for 3 h, followed by five washes with lysis buffer. Bound proteins were eluted by boiling for 10 min and subjected to immunoblotting.

Statistical analysis

Data were analyzed with GraphPad Prism 9.5.1 statistical software (La Jolla, CA). All values are presented as the mean ± standard error on the mean (SEM). Shapiro–Wilk tests were used to examine the normal distribution of the data. The F test was used to test the homogeneity of variance of the data. For data with a nonnormal distribution, we conducted Mann–Whitney tests between two groups and Kruskal–Wallis tests for more than two groups. For data with a normal distribution and equal variance, a two-tailed unpaired t test was used between two groups, and one-way analysis of variance (ANOVA) with Tukey’s post hoc multiple comparisons test was conducted for more than two groups. For data with a normal distribution but unequal variance, an unpaired t test with Welch’s correction was used between two groups, and Brown-Forsythe ANOVA with Dunnett’s T3 post hoc multiple comparisons test was conducted for more than two groups. The body weight curve, glucose curve, and insulin concentration curve were analyzed with a two-way repeated-measures ANOVA with a mixed-effects model (REML), followed by Tukey’s multiple comparisons test. Sample sizes were established following field-specific guidelines, with exact numbers detailed in the corresponding figure legends. mRNA decay kinetics were analyzed by fitting the normalized RNA levels to a one-phase exponential decay model using nonlinear regression. The initial RNA level (Y₀) was constrained to 1, and the plateau was constrained to 0. Decay rate constants (k) and half-lives were calculated from the fit curves. Differences in decay rates between groups were assessed using extra sum-of-squares F tests. P < 0.05 was considered statistically significant.

Results

Identification of IP6K3 as a responsive gene in MASLD and exercise training

To explore exercise-responsive genes involved in MASLD, we analyzed transcriptomic datasets from a bariatric intervention cohort (GSE48452) and a NASH cohort (GSE89632). Differential expression analysis revealed a subset of genes characterized by reciprocal regulation between metabolic improvement and disease progression. Specifically, hepatic IP6K3 was downregulated following bariatric intervention and significantly upregulated in NASH (Fig. 1A–C). Intersection analysis confirmed IP6K3 as a core shared gene exhibiting inverse regulation between these two states (Fig. 1D). Functional enrichment analysis (KEGG/GSEA) of overlapping genes identified inositol phosphate metabolism and the phosphatidylinositol signaling system as the most consistently enriched pathways, further supporting a role for IP6K3 in MASLD pathogenesis (Fig. 1E–H).

Fig. 1.

Identification of IP6K3 as a responsive gene in MASLD and exercise training. A Volcano plot showing differentially expressed genes in the bariatric intervention cohort (GSE48452); B volcano plot showing differentially expressed genes in the NASH cohort (GSE89632); C scatter plot illustrating oppositely regulated genes between bariatric intervention and NASH datasets; D intersection analysis of genes upregulated or downregulated in bariatric intervention and NASH cohorts; E, G KEGG pathway enrichment analysis of overlapping genes in bariatric intervention (E) and NASH (G) datasets; F, H gene set enrichment analysis showing enrichment of inositol phosphate metabolism and phosphatidylinositol signaling system pathways; I immunoblot and qPCR analysis of IP6K3 expression in liver samples from patients with or without MASLD; J immunoblot and qPCR analysis of IP6K3 expression in livers from normal diet (ND) and HFHC-fed mice; K analysis of GSE189600 (publicly available single-nucleus RNA-sequencing data from the livers of patients with MASLD or healthy individuals) revealed that IP6K3 was abundantly expressed in hepatic parenchymal cells; L analysis of odds ratios (OR) and a plot of IP6K3 transcriptional levels in various cell types in GSE189600. The red arrow indicates hepatocytes; M immunofluorescent staining of IP6K3 (green) with Alb (red, a marker of hepatocytes) and CD68 (red, a marker of macrophages in human) in liver tissues of patients with MASLD. Immunofluorescent staining of IP6K3 (green) with Alb (red) and F4/80 (red, a marker of macrophages in mice) in the liver tissues of mice fed an HFHC diet. DAPI was used to stain nuclei (blue). Scale bar, 100 μm; N schematic illustration of the HFHC diet-induced MASLD model with or without exercise training; O immunoblot and qPCR analysis of IP6K3 expression in liver samples from HFHC mice with or without exercise training; P immunofluorescent staining of IP6K3 (green) with Actin (red) in the liver tissues from HFHC mice with or without exercise training. DAPI was used to stain nuclei (blue). scale bar, 100 μm. n = 4–6 per group, **P < 0.01, ****P < 0.0001

We next validated these findings in clinical and experimental samples. Hepatic IP6K3 mRNA and protein levels were significantly elevated in patients with MASLD and in mice fed a high-fat high-cholesterol (HFHC) diet compared with healthy controls (Fig. 1I, J). Analysis of single-nucleus RNA sequencing data from liver samples of patients with nonalcoholic steatohepatitis (GSE189600) revealed that IP6K3 was predominantly enriched in hepatic parenchymal cells, with negligible expression in non-parenchymal populations such as macrophages and endothelial cells (Fig. 1K, L). This hepatocyte-specific localization was further corroborated by immunofluorescence co-localization studies in both clinical and murine MASLD liver tissues (Fig. 1M).

Finally, we investigated whether exercise training attenuated the hepatic upregulation of IP6K3. In HFHC-fed mice, regular exercise training significantly attenuated the hepatic upregulation of IP6K3 at both the transcript and protein levels (Fig. 1N, O), a finding confirmed by immunofluorescence (Fig. 1P). Notably, exercise training did not alter the expression of other family members, IP6K1 and IP6K2, in liver tissues of HFHC-fed mice, suggesting that IP6K3 is a uniquely exercise-responsive isoform within the liver (Supplementary Fig. S1A–C). Collectively, these results identify IP6K3 as a hepatocyte-specific, exercise-responsive regulator that is aberrantly induced during MASLD and suppressed by exercise intervention.

Hepatocyte-specific Ip6k3 deletion attenuates HFHC-induced liver injury

To determine the functional contribution of IP6K3 to MASLD progression, we generated hepatocyte-specific Ip6k3-deficient mice (Ip6k3^fl/flAlb-Cre) and their wild-type (Ip6k3^fl/fl) littermates (Supplementary Fig. S2). Following 16 weeks of HFHC diet intake, Ip6k3^fl/flAlb-Cre mice exhibited significantly reduced weight gain, liver weight, and the ratio of liver weight-to-body weight in response to HFHC intake compared with the Ip6k3^fl/fl counterparts (Supplementary Fig. S3). Histological assessment revealed that the pronounced fibrosis, lipid accumulation, and impaired glycogen storage observed in HFHC-fed Ip6k3^fl/fl mice were largely prevented by hepatocyte-specific Ip6k3 deletion (Fig. 2A).

Fig. 2.

Hepatocyte-specific Ip6k3 deletion attenuates HFHC-induced liver injury. Four groups of Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice fed a normal diet (ND) or an HFHC diet were included, and the following indices were examined. A Representative images and quantitative analyses of HE, Masson’s trichrome, Oil Red O, and periodic acid–Schiff (PAS) staining in the liver tissues. Scale bar, 100 µm; B, C GTT and ITT assay; D serum ALT, AST, TG, and TC levels; E mRNA levels of pro‐inflammatory factors (Il-6, Il-1β, Tnf-α, Mcp-1), lipid synthesis (Acca, Fas, Srebp1c), lipid intake (Fabp1, Fatp1, Cd36, Ppar-γ), bile acid synthesis (Cyp7a1), and fatty acid oxidation (Ucp2, Pdk4, Ppara, Cpt1a, Acox1) in the liver tissues; F protein levels of α-SMA, TNF-α, IL-6, and IL-1β in the liver tissues; G protein levels of p-AKT and p-IRS1 in the liver tissues. n = 3–8 per group, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Consistent with the improved histological profile, Ip6k3^fl/flAlb-Cre mice displayed superior glucose homeostasis and insulin sensitivity under HFHC conditions, as evidenced by enhanced glucose and insulin tolerance tests (Fig. 2B, C). This systemic improvement was corroborated by augmented hepatic insulin signaling, as indicated by elevated p-AKT/AKT and p-IRS1/IRS1 ratios (Fig. 2G). Furthermore, biochemical analysis of serum markers revealed that ALT, AST, TG, and TC levels were significantly lower in HFHC-fed Ip6k3^fl/flAlb-Cre mice when compared with control group (Fig. 2D).

Molecular profiling of liver tissues showed that Ip6k3 deficiency prompted a favorable shift in the transcriptional landscape. Specifically, we observed a marked downregulation of genes involved in inflammation (Il-6, Il-1β, Tnf-α, Mcp-1), lipid synthesis (Acca, Fas, Srebp1c), lipid intake (Fabp1, Fatp1, Cd36, Ppar-γ), bile acid synthesis (Cyp7a1), and fatty acid oxidation (Ucp2, Pdk4, Ppara, Cpt1a, Acox1) in Ip6k3^fl/flAlb-Cre mice compared with Ip6k3^fl/fl mice under HFHC diet (Fig. 2E). Western blotting confirmed that hepatocyte-specific Ip6k3 deletion relieved hepatic fibrosis (evidenced by reduced α-SMA expression) and reduced pro-inflammatory cytokine production (TNF-α, IL-6, and IL-1β) under HFHC intake (Fig. 2F). These findings provided evidence that Ip6k3 in hepatocyte played an essential role in MASLD.

Single-cell RNA sequencing reveals that hepatocyte-specific Ip6k3 deletion reprograms ferroptosis and BCAA metabolism

To elucidate the potential mechanisms by which hepatocyte-specific Ip6k3 deletion protects against HFHC-induced liver injury, we performed single-cell RNA sequencing to profile the transcriptomic landscape of hepatic cells from HFHC-fed Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice (Fig. 3A). After stringent quality control, 13,784 cells were retained for downstream analysis and visualized via UMAP. Unsupervised clustering identified ten distinct cell populations, including hepatocytes, hepatic stellate cells, cholangiocytes, endothelial cells, dendritic cells, B cells, NKT cells, neutrophils, macrophages, and monocytes (Fig. 3B). The canonical marker genes defining each cluster are shown in the heatmap in Fig. 3C, and the relative proportions of these populations across genotypes are shown in Fig. 3D. Intercellular communication analysis using CellChat [51] revealed a substantial reduction in interaction strength between hepatocytes and inflammatory myeloid populations (monocytes, macrophages, or neutrophils) in Ip6k3^fl/flAlb-Cre mice compared with the control (Fig. 3E), suggesting that hepatocyte-specific Ip6k3 deletion might dampen inflammatory responses in the liver in MASLD.

Fig. 3.

Single-cell RNA sequencing reveals that hepatocyte-specific Ip6k3 deletion reprograms ferroptosis and BCAA metabolism. Single-cell RNA sequencing (scRNA-seq) analysis was performed with liver tissues from Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice fed an HFHC diet. A Schematic illustration of the experimental workflow for scRNA-seq; B UMAP visualization of ten cell clusters; C heatmap showing the expression of the indicated genes in ten different cell clusters; D relative proportions of the ten major liver cell populations in HFHC-fed Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice; E intercellular interactions among ten cell clusters; F volcano plot showing the downregulated (197) and upregulated (240) genes in Ip6k3^fl/flAlb-Cre mice fed an HFHC diet when compared with the Ip6k3^fl/fl counterpart; G KEGG enrichment analysis showing the upregulated (red) and downregulated (blue) genes in hepatocytes in Ip6k3^fl/flAlb-Cre mice fed an HFHC diet when compared with the Ip6k3^fl/fl counterpart. H UMAP visualization of hepatocyte subsets; I Trajectory inference of hepatocytes along the predominant transcriptional trajectory (partition 1) under HFHC conditions; J Pseudotime-dependent changes in iron-, ferroptosis resistance- and BCAA degradation-associated gene signature scores along the predominant hepatocyte trajectory (partition 1) in HFHC-fed Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice. Gene signature scores were calculated based on curated gene sets associated with iron handling, ferroptosis regulation, and BCAA degradation

Given that IP6K3 was predominantly expressed in hepatocytes, we focused our analysis on the hepatocyte cluster. Compared with HFHC-fed Ip6k3^fl/fl mice, HFHC-fed Ip6k3^fl/flAlb-Cre mice exhibited 240 upregulated and 197 downregulated genes (Fig. 3F). KEGG enrichment analysis identified significant transcriptomic shifts in the ferroptosis and BCAA (valine, leucine, and isoleucine) degradation pathways (Fig. 3G).

Trajectory inference analysis further unveiled the transcriptional heterogeneity of hepatocytes under HFHC stress, identifying five clusters across two distinct partitions (Fig. 3H, I). Partition 1, which encompassed the majority of hepatocytes, was selected for primary pseudotemporal analysis. Along this trajectory, hepatocytes from both genotypes exhibited a progressive decline in gene signature scores associated with iron homeostasis, ferroptosis resistance, and BCAA degradation. However, across the entire pseudotime trajectory, hepatocytes from Ip6k3^fl/flAlb-Cre mice showed higher ferroptosis resistance- and BCAA degradation-associated gene signature scores compared with the control (Fig. 3J). These findings indicate a gradual loss of redox and metabolic homeostasis under chronic HFHC stress for both genotypes of mice, although the Ip6k3^fl/flAlb-Cre mice exhibited a protective shift. Variation in the pseudotime trajectory for partition 2 was shown in Supplementary Fig. S4. Collectively, these data suggest that the hepatoprotective effects of Ip6k3 deficiency are driven by the inhibition of ferroptosis and the enhancement of BCAA catabolism, which together reshape the intrinsic metabolic state of the liver in MASLD.

Hepatocyte-specific Ip6k3 deletion and exercise training suppress ferroptosis and accelerate BCAA metabolism in MASLD

We next investigated whether hepatocyte-specific Ip6k3 deletion functionally suppressed ferroptosis in vivo. Ferroptosis is characterized by decreased expression of GPX4, accumulation of Fe²⁺, and excessive production of lipid peroxides, leading to uncontrollable redox reactions, and ultimately cell death [11, 52, 53]. Compared with HFHC-fed Ip6k3^fl/fl mice, HFHC-fed Ip6k3^fl/flAlb-Cre mice showed decreased protein levels of pro-ferroptotic factors (TFR, ACSL4, ALOX15) and increased protein levels of anti-ferroptotic factors (GPX4, DHODH, FSP1) in liver tissues. Changes in mRNA expression were consistent with the protein (Fig. 4A, B). In line with the suppressed ferroptotic signaling, HFHC-fed Ip6k3^fl/flAlb-Cre mice also exhibited increased hepatic GSH, the ratio of NAD⁺/NADH and activity of SOD, together with reduced reactive oxygen species (ROS) production and iron content (Fig. 4C–E). These findings were replicated in vitro using HepG2 cells. Under PA stimulation, si-IP6K3 downregulated pro-ferroptotic factors and upregulated anti-ferroptotic defense proteins while reducing ROS production and iron accumulation (Supplementary Fig. S5A–D). Conversely, IP6K3 overexpression by pcDNA3.1-IP6K3 led to increased expression of pro-ferroptotic proteins and decreased expression of anti-ferroptotic proteins (Supplementary Fig. S5E). The efficacy of si-IP6K3 (1401 was used in this work) and pcDNA3.1-IP6K3 was confirmed (Supplementary Fig. S5F, G), and an optimal PA concentration (0.3 mM) was selected (Supplementary Fig. S5H). These results suggest that hepatic IP6K3 plays an essential role in the regulation of ferroptosis in MASLD.

Fig. 4.

Hepatocyte-specific Ip6k3 deletion and exercise training suppress ferroptosis and accelerate BCAA metabolism in MASLD. A Four groups were designed: Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice fed a normal diet (ND) or an HFHC diet. mRNA levels of pro-ferroptotic (Tfr, Acsl4, Alox15) and anti-ferroptotic factors (Gpx4, Dhodh, Fsp1) in liver tissues of the four groups; B protein levels of pro-ferroptotic (TFR, ACSL4, ALOX15) and anti-ferroptotic factors (GPX4, DHODH, FSP1) in liver tissues of the four groups; C liver GSH, the ratio of NAD⁺/NADH and activity of SOD in liver tissues of the four groups; D, E ROS production and free iron in living cells (mostly Fe²⁺) assessed by 2′,7′-Dichlorodihydrofluorescein diacetate (DCFH-DA) or FerroOrange probes with flow cytometry in single‐cell suspensions prepared from mice liver tissues of the four groups; F, G serum BCAA levels measured by LC–MS, and liver BCAA levels assessed by commercial BCAA assay kit in the four groups; H three groups were designed: Ip6k3^fl/fl mice without exercise training, Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice with 8-week exercise training. All three groups were fed a 16-week HFHC diet; I liver GSH, NAD⁺/NADH, and activity of SOD in three groups; J, K mRNA levels of pro-ferroptotic (Tfr, Acsl4, Alox15) and anti-ferroptotic factors (Gpx4), and corresponding protein levels in liver tissues of the three groups; L, M Serum (detected by LC–MS) and liver (detected by a BCAA assay kit) BCAA levels in the three groups. n = 3–10 per group, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Recent studies found that increased circulating BCAA (leucine, isoleucine, and valine) was associated with a heightened risk of metabolic diseases, including MASLD [7–9], while BCAA accumulation prompts ROS production and ferroptosis [15, 54, 55]. In this work, both serum and hepatic BCAA levels were significantly increased in HFHC-fed Ip6k3^fl/fl mice compared with normal diet-fed counterparts, which was markedly dampened in Ip6k3^fl/flAlb-Cre mice (Fig. 4F, G). These findings indicate that hepatic IP6K3 could regulate BCAA metabolism during MASLD.

Exercise improves health outcomes in patients with MASLD and also reduces lipid droplet formation and decreases hepatic triglycerides in high-fat diet-treated mice [56–58]. To determine whether exercise-conferred beneficial effects were related to hepatocyte IP6K3, mice were subjected to an exercise intervention (Fig. 4H). Compared with the non-exercise Ip6k3^fl/fl mice fed an HFHC diet, exercise training in Ip6k3^fl/fl led to decreased body weight, liver weight, and liver-to-body weight ratio caused by HFHC; attenuated hepatic steatosis and improved glycogen metabolism; reduced serum levels of ALT, AST, TG, and TC; and suppressed the expression of inflammatory cytokines (TNF-α, IL-6, and IL-1β). Almost all these changes were further enhanced by hepatocyte-specific Ip6k3 deletion (Supplementary Fig. S6A–E). These findings demonstrate that exercise training is capable of attenuating HFHC-elicited liver injury, which was intensified by hepatocyte-specific Ip6k3 deletion.

We therefore examined whether exercise training inhibited ferroptosis and accelerated BCAA metabolism. Our results revealed that exercise training significantly increased hepatic antioxidant indices (GSH, NAD⁺/NADH, and SOD) (Fig. 4I), inhibited the mRNA expression of pro-ferroptotic factors (Tfr, Acsl4, and Alox15), and upregulated the anti-ferroptotic factor Gpx4 in liver tissues (Fig. 4J). The corresponding protein changes in liver tissues were consistent with the mRNA results, except for ALOX15 (Fig. 4K). Hepatocyte-specific Ip6k3 deletion further enhanced most of the benefits conferred by exercise training (Fig. 4H–K). It was also demonstrated that exercise training significantly reduced BCAA levels in either serum or hepatic tissues in Ip6k3^fl/fl mice fed an HFHC diet. However, exercise training did not further decrease the BCAA levels in Ip6k3^fl/flAlb-Cre mice when compared with the Ip6k3^fl/fl mice (Fig. 4L, M). These findings suggested that hepatocyte-specific Ip6k3 deletion could suppress ferroptosis and accelerate BCAA metabolism in MASLD, and the protection against liver injury in MASLD by exercise training was possibly mediated by IP6K3.

IP6K3 negatively regulated BCAT2 expression in MASLD

To explore how IP6K3 regulated BCAA degradation, we analyzed six BCAA catabolism-related genes (Bcat1, Bcat2, Ppm1k, Bckdk, Bckdhα, and Bckdhβ) [15] under different conditions. In vivo, compared with HFHC-fed Ip6k3^fl/fl mice, the mRNA level of Bcat2 was increased in HFHC-fed Ip6k3^fl/flAlb-Cre mice (Fig. 5A). In vitro, silencing of IP6K3 in HepG2 cells led to increased BCAT2, BCKDK, and BCKDHα mRNA expression under PA stimulation (Fig. 5B). Exercise training increased the mRNA level of Bcat2, Ppm1k, Bckdk, and Bckdhα in HFHC-fed Ip6k3^fl/fl mice compared with the non-exercise Ip6k3^fl/fl mice (Fig. 5C).

Fig. 5.

IP6K3 negatively regulated BCAT2 expression in MASLD. A and D mRNA levels of Bcat1, Bcat2, Ppm1k, Bckdk, Bckdhα, and Bckdhβ and protein level of BCAT2 in liver tissues of Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice fed a normal diet (ND) or an HFHC diet; B and E mRNA levels of BCAT1, BCAT2, PPM1K, BCKDK, BCKDHα, and BCKDHβ and protein level of BCAT2 in HepG2 cells treated with si-scramble or si-IP6K3 with or without PA incubation; C and F mRNA levels of Bcat1, Bcat2, Ppm1k, Bckdk, Bckdhα, and Bckdhβ and protein level of BCAT2 in liver tissues of Ip6k3^fl/fl, Ip6k3^fl/fl + exercise and Ip6k3^fl/flAlb-Cre + exercise mice fed with HFHC diet. n = 3 for protein, n = 6 for mRNA, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

On the basis of these findings, we focused on the gene BCAT2 and tried to establish the interplay between IP6K3 and BCAT2. Compared with HFHC-fed Ip6k3^fl/fl mice, HFHC-fed Ip6k3^fl/flAlb-Cre mice exhibited a significant increase in the protein level of BCAT2 (Fig. 5D). In vitro, IP6K3 knockdown in HepG2 cells increased the protein level of BCAT2 under PA stimulation (Fig. 5E). Importantly, exercise training also increased the protein level of BCAT2 in HFHC-fed Ip6k3^fl/fl mice compared with the non-exercise counterparts, and hepatocyte-specific Ip6k3 deletion further increased BCAT2 (Fig. 5F).

Collectively, these data demonstrate that the suppression of BCAT2 (the rate-limiting enzyme initiating BCAA catabolism) under HFHC or PA stress can be effectively reversed by either hepatocyte-specific Ip6k3 deletion, IP6K3 knockdown in HepG2 cells, or exercise training. These findings support the model in which IP6K3 serves as a negative regulator of BCAT2 expression in the context of MASLD. Furthermore, the observation that exercise restores BCAT2 expression suggests that the metabolic benefits of physical exercise are, at least in part, mediated by the suppression of hepatic IP6K3.

IP6K3 regulates BCAA-induced ferroptosis in a BCAT2-dependent manner

To further delineate the link among IP6K3, BCAA metabolism, and ferroptosis, we used excessive BCAA to induce ferroptosis in hepatocytes and explored whether this process was regulated by IP6K3 and BCAT2. The concentration of BCAA (6 mM) for ferroptosis induction was essentially on the basis of our experimental results (Supplementary Fig. S7A, B) together with relevant literature [41, 42].

Compared with the control, excessive BCAA incubation in HepG2 cells significantly increased lipid peroxidation as evidenced by an elevated green/red fluorescence ratio. Consistently, excessive BCAA significantly upregulated the expression of pro-ferroptotic factors (TFR, ACSL4, ALOX15) while suppressing the anti-ferroptotic factor GPX4 at both mRNA and protein levels. Importantly, these changes induced by excessive BCAA were further enhanced by IP6K3 overexpression (Fig. 6A–C). To confirm that the excessive BCAA-induced cell death was ferroptotic in nature, the ferroptosis inhibitor ferrostatin-1 (Fer-1) was applied. We found that excessive BCAA-induced ferroptosis was inhibited by Fer-1 at a final concentration of 2 μM (Supplementary Fig. S7C). Though BCAA-induced ferroptosis was significantly enhanced by IP6K3 overexpression, the ferroptosis-specific inhibitor Fer-1 almost completely abolished the effect evoked by IP6K3 overexpression (Fig. 6D–F).

Fig. 6.

IP6K3 regulates BCAA-induced ferroptosis in a BCAT2-dependent manner. A–C Representative images and quantification of lipid peroxidation assessed by C11-BODIPY (581/591) staining, mRNA levels of pro-ferroptotic (TFR, ACSL4, ALOX15) and anti-ferroptotic factors (GPX4) and the corresponding protein levels in HepG2 cells transfected with pcDNA3.1 without BCAA stimulation, and pcDNA3.1 or pcDNA3.1-IP6K3 with BCAA stimulation; D–F Representative images and quantification of lipid peroxidation assessed by C11-BODIPY (581/591) staining, mRNA levels of pro-ferroptotic (TFR, ACSL4, ALOX15) and anti-ferroptotic factors (GPX4) and the corresponding protein levels in HepG2 cells transfected with pcDNA3.1, or pcDNA3.1-IP6K3 with or without Fer-1 treatment (all three groups was under BCAA stimulation). G–I Representative images and quantification of lipid peroxidation assessed by C11-BODIPY (581/591) staining, mRNA levels of pro-ferroptotic (TFR, ACSL4,, ALOX15) and anti-ferroptotic factors (GPX4) and the corresponding protein levels in HepG2 cells transfected with si-scramble without BCAA stimulation, and si-scramble or si-IP6K3 with BCAA stimulation; J–L Representative images and quantification of lipid peroxidation assessed by C11-BODIPY (581/591) staining, mRNA levels of pro-ferroptotic (TFR, ACSL4, ALOX15) and anti-ferroptotic factors (GPX4) and the corresponding protein levels in HepG2 cells transfected with si-scramble with or without iBCAT2 (BCAT-IN-2, a BCAT2 inhibitor) treatment, or transfected with si-IP6K3 under BCAA stimulation with or without iBCAT2 treatment. n = 3 for protein, n = 6 for mRNA and C11-BODIPY (581/591) staining. Scale bar, 50 μm. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

While overexpression of IP6K3 exacerbated BCAA-induced ferroptosis (Fig. 6A–C), IP6K3 knockdown effectively prevented the ferroptotic cell death triggered by excessive BCAA loading in HepG2 cells (Fig. 6G–I). Given our earlier demonstration that IP6K3 negatively regulated BCAT2 expression (Fig. 5), the enzyme responsible for the critical initial step of BCAA catabolism [14], we sought to determine whether BCAT2 is the functional mediator of IP6K3’s effects. To this end, we employed BCAT-IN-2 (iBCAT2, a BCAT2 inhibitor) to block BCAA degradation. While iBCAT2 treatment alone did not overtly alter the baseline expression of ferroptosis-related factors, it profoundly influenced the protective effects of IP6K3 silencing. Specifically, the capacity of IP6K3 knockdown to prevent BCAA-induced ferroptosis was abolished upon pharmacological inhibition of BCAT2 (Fig. 6J–L). These results demonstrate that the protective effect of IP6K3 deficiency is contingent upon functional BCAA catabolism. Collectively, our findings establish an obligate role for BCAT2 in this regulatory axis, defining a novel “IP6K3–BCAT2–BCAA–ferroptosis” functional pathway in the pathogenesis of MASLD.

IP6K3 interacts with HNRNPK to destabilize BCAT2 mRNA

Direct interaction between IP6K3 and BCAT2 would provide a straightforward explanation for the inhibition of BCAA catabolism. However, the Co-IP assay failed to detect a physical association between these two proteins (data are not provided). To elucidate the mechanism by which IP6K3 modulates BCAT2 expression, we transfected HepG2 cells with an HA-tagged IP6K3 vector, subjected them to PA, and performed immunoprecipitation using an anti-HA antibody. The most prominent band (red box) was selected for further investigation, and nanoscale liquid chromatography–tandem mass spectrometry identified two potential interactors, with heterogeneous nuclear ribonucleoprotein K (HNRNPK) among the top candidates (Fig. 7A). Functional screening via RNA interference revealed that knocking down HNRNPK further suppressed PA-induced BCAT2 transcription in HepG2 cells. However, knockdown of the synaptotagmin-binding cytoplasmic RNA-interacting protein (SYNCRIP) did not alter PA-induced BCAT2 transcription. These findings specifically implicate HNRNPK in the regulation of BCAT2 (Supplementary Fig. S7D, E).

Fig. 7.

IP6K3 interacts with HNRNPK to destabilize BCAT2 mRNA. A HepG2 cells expressing HA-tagged-IP6K3 were treated with PA, lysed, immunoprecipitated with anti-HA, and separated via SDS-PAGE. The most prominent ~ 65 kDa band (indicated by red box) was excised and subjected to mass spectrometry. Molecules interacting with IP6K3 were identified using a nano liquid chromatography–tandem mass spectrometry, and HNRNPK was selected for further investigation; B AlphaFold3-predicted interaction pattern between IP6K3 and HNRNPK. Binding affinity estimation using PRODIGY predicted a strong interaction between IP6K3 and HNRNPK (ΔG ≈ −23.1 kcal/mol); C Co-IP and immunoblotting detected the interaction between IP6K3 and HNRNPK in HepG2 cells under PA stimulation; D actinomycin D chase assay for BCAT2 mRNA expression in HepG2 cells transfected with either pcDNA3.1 or pcDNA3.1-HNRNPK with or without PA stimulation; E, F mRNA levels of BCAT2 and the corresponding protein level of BCAT2 in HepG2 cells treated with either pcDNA3.1 or pcDNA3.1-HNRNPK with or without PA stimulation; G A working model of IP6K3-regulated ferroptosis in MASLD. n = 3 for protein, n = 6 for mRNA, *P < 0.05, ***P < 0.001, ****P < 0.0001

HNRNPK was first identified as a constituent of heterogeneous nuclear ribonucleoprotein (hnRNP) complexes. There are 20 major types of protein in the hnRNP family, including hnRNP A-U [59]. Among them, HNRNPK is believed to play a crucial role in the process of mRNA stability, splicing, and translational regulation [60]. AlphaFold3 analysis predicted the existence of a possible interaction between IP6K3 and HNRNPK and displayed 23 binding sites between the two proteins. Binding affinity estimation using PRODIGY suggested that the IP6K3-HNRNPK complex adopts a highly stable interaction conformation, with a predicted binding free energy of approximately −23.1 kcal/mol (Fig. 7B). Detailed PRODIGY output parameters are provided in Table S3 (Supplemental Materials). Co-IP assays confirmed the interaction between IP6K3 and HNRNPK in HepG2 cells under PA stimulation (Fig. 7C).

To determine whether HNRNPK regulates BCAT2 at the post-transcriptional level, we assessed the mRNA stability of BCAT2 following actinomycin D treatment. HNRNPK overexpression significantly attenuated the decay of BCAT2 mRNA in PA-stimulated HepG2 cells, indicating that HNRNPK acts as a stabilizing factor for the BCAT2 transcript (Fig. 7D). Consistently, HNRNPK overexpression elevated both BCAT2 mRNA and BCAT2 protein levels under lipotoxic stress (Fig. 7E, F). The efficacy of pcDNA3.1-HNRNPK was confirmed (Supplementary Fig. S7F). Collectively, these data suggest a model in which IP6K3 sequesters or interferes with HNRNPK under MASLD-like conditions (PA stimulation), thereby promoting the instability of BCAT2 mRNA and reducing the expression of the BCAT2 enzyme.

Discussion

MASLD is a primary driver of end-stage liver diseases globally, and although recent years have witnessed progress in the development of pharmacotherapies for MASH, treatment options applicable to the broader patient population remain limited overall [3–5, 22]. In this study, we identified IP6K3 as a previously unrecognized regulator responsible of BCAA metabolism and ferroptosis in MASLD. Our main findings demonstrate that (1) IP6K3 is upregulated in MASLD but can be downregulated through exercise training; (2) Both hepatocyte-specific Ip6k3 deletion and exercise training mitigate liver injury by promoting BCAA catabolism and suppressing ferroptosis in MASLD; (3) Mechanistically, IP6K3 regulates BCAA-induced ferroptosis in a BCAT2-dependent manner by binding to HNRNPK, which subsequently destabilizes BCAT2 mRNA. Collectively, these results delineate a functional IP6K3/HNRNPK–BCAT2–BCAA–ferroptosis axis in MASLD. We believe that the initial therapeutic step involves disrupting the IP6K3/HNRNPK complex to release HNRNPK, thereby stabilizing BCAT2 mRNA, enhancing BCAA metabolism, and ultimately attenuating ferroptosis in MASLD (Fig. 7G). Given that MASLD affects approximately 38% of the global adult population and involves complex metabolic dysregulation [3, 5, 61–65], identifying IP6K3 as a metabolic rheostat offers a promising strategy for the development of much-needed pharmacological treatments.

Since their discovery in 1993, inositol pyrophosphates have emerged as evolutionarily conserved signaling metabolites with versatile and increasingly recognized models of action [26, 66, 67]. The IP6K family consists of three isoforms (IP6K1–3) characterized by distinct tissue distributions and physiological roles. While IP6K1 has been implicated in insulin sensitivity [27, 30] and protection against myocardial ischemia–reperfusion injury [68] and IP6K2 is known to modulate apoptosis in ovarian carcinoma cells [69, 70], the role of IP6K3 in metabolic homeostasis is only beginning to be elucidated. Previous reports indicated that Ip6k3 is highly expressed in muscle and that its deletion induces significant metabolic changes and impacts lifespan [35]. Furthermore, pharmacological inhibition of IP6K family via UNC7467 has shown promise in ameliorating hepatic steatosis in obese mice [71]. In the present study, we utilized integrated transcriptomic analysis to identify IP6K3 as a key gene that is upregulated in non-alcoholic steatohepatitis but downregulated following bariatric surgery. We confirmed that IP6K3 is primarily localized in hepatocytes and is abundantly expressed in liver samples from patients with MASLD and HFHC-fed mice. Most importantly, we found that exercise training effectively suppresses hepatic IP6K3 expression in MASLD. By employing hepatocyte-specific Ip6k3 deletion mice, we demonstrated that IP6K3 is an essential driver of lipid accumulation, insulin resistance, and hepatic inflammation. Collectively, our data identify IP6K3 as a previously unrecognized metabolic rheostat that mediates the hepatoprotective effect of exercise training in MASLD.

To elucidate the potential mechanisms by which IP6K3 modulates liver injury, we conducted single-cell RNA sequencing to compare the transcriptomic profiles of Ip6k3^fl/fl and Ip6k3^fl/flAlb-Cre mice fed a HFHC diet. Our analysis revealed that the hepatoprotective effects of hepatocyte-specific Ip6k3 deletion are primarily mediated through the inhibition of ferroptosis and the augmentation of BCAA catabolism. Ferroptosis, a form of regulated cell death characterized by the downregulation of GPX4, accumulation of intracellular Fe²⁺, and excessive lipid peroxidation, has emerged as a critical driver of hepatocyte death and MASLD progression [11, 52, 53]. Consistent with prior studies showing that ferroptosis inhibition alleviates liver injury [10, 36], our data suggest that IP6K3 may act as a pro-ferroptotic regulator in the MASLD context.

Furthermore, we investigated the role of BCAA (isoleucine, leucine, and valine), essential amino acids primarily derived from diet [16]. A recent study have found that low-protein diets reduce hepatic lipid accumulation and liver damage, uncovering a previously unappreciated role of amino acids in hepatic lipogenesis and MASLD [72]. Landmark tracing experiments in mice demonstrated that the major sites for BCAA oxidation are skeletal muscle (59%), brown adipose tissue (19%), liver (8%), kidney (5%), and heart (4%), whereas the major sites for BCAA incorporation into protein include liver (27%), pancreas (24%), skeletal muscle (24%), kidney (9%), and brown adipose tissue (6%) [73]. Liu and colleagues found that BCAA accumulation in the liver promotes hepatic tumor in mice [18]. Our findings reinforce the “pathogenic” role of BCAA buildup in liver injury and position IP6K3 as a previously unrecognized gatekeeper of BCAA utilization in the liver.

Exercise training is a cornerstone of MASLD management, recognized for its ability to reduce hepatic steatosis [58, 74], enhance insulin sensitivity [75], and restore mitochondrial function [76]. Recent evidence suggests that the improvements in glucose homeostasis and insulin sensitivity in diabetic patients and obese murine models following exercise are closely linked to the systemic activation of BCAA catabolism [24, 25]. However, the specific molecular transducers that relay the benefits of exercise to hepatic BCAA metabolic pathways remain to be fully defined.

In the present study, we demonstrated that hepatocyte-specific Ip6k3 deletion effectively inhibits ferroptosis, alleviates liver injury, and reduces BCAA concentrations in both serum and in the liver of HFHC-fed mice. Crucially, we found that exercise training replicates the phenotypic benefits of hepatocyte-specific Ip6k3 deletion. This alignment is further supported by our observation that exercise training significantly attenuates the pathological upregulation of IP6K3 induced by an HFHC diet. Collectively, these findings position IP6K3 as a previously unrecognized metabolic rheostat that is induced by MASLD but suppressed by physical activity. Our study suggests that the downregulation of IP6K3 is a primary mechanism through which exercise promotes BCAA catabolism and suppresses ferroptosis to counteract the progression of MASLD.

To elucidate the mechanisms by which IP6K3 governs hepatic BCAA degradation in MASLD, we systematically evaluated the impact of IP6K3 and exercise training on six key BCAA catabolic genes (Bcat1, Bcat2, Ppm1k, Bckdk, Bckdhα, and Bckdhβ) [15]. Intracellular BCAA levels are primarily regulated by the initial two steps of the catabolic pathway: the reversible transamination catalyzed by branched-chain aminotransferase isozymes (cytosolic BCAT1 and mitochondrial BCAT2) followed by the irreversible oxidative decarboxylation mediated by the branched-chain α-keto acid dehydrogenase (BCKDH) complex [77]. Notably, BCAT2 has recently been identified as a critical metabolic checkpoint that modulates ferroptosis, apoptosis, and autophagy in various pathological contexts [54]. Our screening revealed that BCAT2 was the exclusive common target upregulated following either hepatocyte-specific Ip6k3 deletion or exercise training in HFHC-fed mice, as well as by IP6K3 knockdown in PA-stimulated hepatocytes. These findings suggest that IP6K3 acts as a negative regulator of BCAT2 expression in MASLD and that BCAT2 is the primary enzyme mediating BCAA degradation downstream of both Ip6k3 ablation and physical exercise.

To establish a direct functional link between the IP6K3–BCAT2–BCAA axis and ferroptosis, we conducted a series of in vitro experiments. In HepG2 cells, excessive BCAA stimulation induced significant ferroptotic cell death, which was markedly exacerbated by IP6K3 overexpression but effectively rescued by the ferroptosis inhibitor Ferrostatin-1 (Fer-1). Conversely, IP6K3 knockdown attenuated BCAA-induced ferroptosis; however, this protective effect was completely abolished by the pharmacological inhibition of BCAT2 (iBCAT2). Collectively, these results delineate a novel functional pathway, IP6K3–BCAT2–BCAA–ferroptosis, wherein IP6K3 inhibition is essential for maintaining BCAT2-mediated BCAA homeostasis and cellular survival under metabolic stress.

Finally, we investigated the regulatory mechanism by which IP6K3 controls BCAT2 expression in MASLD. Initial Co-IP assays ruled out a direct protein–protein interaction between IP6K3 and BCAT2. To identify potential intermediaries, we employed an integrated approach combining HPLC–MS, AlphaFold3 prediction, and Co-IP assays, which identified HNRNPK as a primary candidate interactor. HNRNPK is a versatile heterogeneous nuclear ribonucleoprotein (hnRNP) protein belonging to the hnRNP family [59], with a crucial role in mRNA stability, splicing, and translation. As an RNA-binding protein, HNRNPK functions through its three K homology domains [60]. AlphaFold3 analysis predicted the existence of a possible highly stable interaction between IP6K3 and HNRNPK, which was confirmed by Co-IP assays. Furthermore, HNRNPK overexpression in HepG2 cells elevated both BCAT2 mRNA and BCAT2 protein levels under PA stimulation. Actinomycin D chase assay indicated that such regulation occurred, possibly by stabilizing BCAT2 mRNA. These findings suggest that HNRNPK could stabilize BCAT2 mRNA to increase BCAT2 protein output.

Conclusions

We identify IP6K3 as a critical pathological driver of MASLD through the modulation of BCAA metabolism and ferroptosis. Under MASLD conditions, elevated IP6K3 binds to HNRNPK, leading to destabilize BCAT2 mRNA and the reduction of BCAT2-mediated BCAA catabolism, thus promoting hepatocyte ferroptosis and exacerbating liver injury. Exercise training suppresses hepatic IP6K3 to release HNRNPK, which stabilizes BCAT2 mRNA to increase BCAT2 protein expression, thus restoring BCAA metabolism and inhibiting ferroptosis. These findings delineate a novel IP6K3/HNRNPK–BCAT2–BCAA–ferroptosis signaling axis and suggest that targeting IP6K3 may represent a promising therapeutic strategy for MASLD.

Abbreviations

MASLD

Metabolic dysfunction-associated steatotic liver disease

NAFLD

Non-alcoholic fatty liver disease

HFHC

High-fat and high-cholesterol diet

Ip6k3^fl/flAlb-Cre

Hepatocyte-specific Ip6k3 deletion

IP6K3

Inositol hexakisphosphate kinase 3

BCAA

Branched-chain amino acid

BCAT2

Branched-chain aminotransferase 2

HNRNPK

Heterogeneous nuclear ribonucleoprotein K

SYNCRIP

Synaptotagmin-binding cytoplasmic RNA interacting protein

TFR

Transferrin receptor

ACSL4

Acyl-CoA synthetase long-chain family member 4

ALOX15

Arachidonate 15-lipoxygenase

GPX4

Glutathione peroxidase 4

FSP1

Ferroptosis suppressor protein 1

DHODH

Dihydroorotate dehydrogenase

Fer-1

Ferrostatin-1

scRNA-seq

Single-cell RNA sequencing

KEGG

Kyoto Encyclopedia of Genes and Genomes

GSEA

Gene set enrichment analysis

Co-IP

Co-immunoprecipitation

Author contributions

F.M.S. and Z.C.L. conceived and designed research; Z.C.L., F.F.X., Y.Z., C.R.S., X.Y.S., Y.P.X., H.G., J.T.F., and X.Y.T. performed experiments; Z.C.L., L.L., D.J.L., and H.F. analyzed data. Z.C.L., F.M.S., and J.R wrote the manuscript. D.J.L. and F.M.S. provided funding. All authors contributed to productive discussions and provided knowledge for the final version of this manuscript.

Funding

This work was supported by grants from the National Natural Science Foundation of China (82273922, 82274030, and 82304687).

Data availability

Processed single-cell RNA-seq data of this study was deposited in the Gene Expression Omnibus under accession codes GSE297883. To review GEO accession GSE297883, go to (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE297883) and enter token “clivqciqzzqxjkp” into the box.

Declarations

Ethics approval and consent to participate

The study protocol was performed in accordance with the Declaration of Helsinki, and written informed consent was obtained from all participants. The study was approved by the institutional research ethics committee of Shanghai Tenth People’s Hospital of Tongji University (approval no. SHSY-IEC-4.1/19-126/01) on 4 September 2019. All animal procedures were conducted in accordance with the recommendations of the International Council for Laboratory Animal Science (ICLAS) guidelines. All experimental procedures were approved by the Institutional Animal Care and Use Committee of Shanghai Tenth People’s Hospital of Tongji University (approval no. SHDSYY-2022-3019) on 6 January 2022.

Consent for publication

All authors gave their consent for publication.

Competing interests

The authors declare no competing interests.