Bariatric surgery resolves MASH by enhancing MAT1A-dependent one-carbon metabolism

Abstract

Bariatric surgery shows variable efficacy in resolving metabolic dysfunction-associated steatohepatitis (MASH), with mechanisms underlying therapeutic heterogeneity remaining unclear. Using multiple male murine MASH models, we demonstrate that vertical sleeve gastrectomy (VSG) ameliorates hepatic steatosis and fibrosis through mechanisms that extend beyond weight loss. Multi-omic profiling reveals VSG enhances hepatic one-carbon metabolism (1CM) via upregulation of methionine adenosyltransferase 1 A (MAT1A), increasing S-adenosylmethionine (SAM) availability and the SAM/SAH ratio. This metabolic reprogramming restores hepatic phosphatidylcholine (PC) /phosphatidylethanolamine (PE) homeostasis, alleviating endoplasmic reticulum stress and mitochondrial dysfunction. Critically, dietary depletion of one-carbon substrates or genetic ablation of MAT1A abolishes VSG’s therapeutic effects, while betaine supplementation rescues surgical efficacy. In humans, paired liver biopsies from patients with MASH (n = 4) show MAT1A upregulation correlating with histological improvement, while plasma metabolomics from a separate cohort (n = 42) reveals that patients with poor hepatic response to VSG display compromised one-carbon metabolism signatures, with SAM levels correlating with therapeutic response. These findings establish MAT1A-mediated methylation as an essential mechanism for optimal VSG efficacy in preclinical models and identify one-carbon metabolites as candidate biomarkers for predicting surgical response, suggesting that preoperative metabolic profiling could guide methyl donor supplementation to optimize VSG outcomes in MASH.

Bariatric surgery resolves metabolic dysfunction-associated steatohepatitis (MASH). Here the authors showed vertical sleeve gastrectomy alleviated MASH by enhancing MAT1A-driven one-carbon metabolism and S-adenosylmethionine production.

Introduction

Metabolic dysfunction-associated steatotic liver disease (MASLD), affecting over one-third of adults globally, represents a major chronic liver disease epidemic. Its progressive form, metabolic dysfunction-associated steatohepatitis (MASH), is characterized by ballooning degeneration and necroinflammation, with or without fibrosis that can advance to life-threatening complications, including cirrhosis, hepatocellular carcinoma, and end-stage liver failure^1,2.

Bariatric surgery remains the most effective intervention for achieving durable weight loss and metabolic improvements³. Vertical sleeve gastrectomy (VSG), now the predominant bariatric procedure worldwide, induces rapid glycemic improvements that often precede significant weight loss⁴, suggesting profound weight-independent metabolic mechanisms. These observations have prompted its application as metabolic surgery for type 2 diabetes in patients with a body mass index (BMI) ≥ 27.5⁵.

Despite these metabolic benefits, MASH is not currently an approved indication for VSG. Evidence from randomized controlled trials examining VSG’s efficacy specifically for MASH remains limited. While retrospective cohorts report improvements in hepatic steatosis following VSG^6–9, the extent of MASH improvement varies considerably¹⁰, with some patients showing fibrosis progression at long-term follow-up, particularly those with baseline fibrosis¹¹. Notably, some patients achieve sustained weight loss yet fail to show significant improvement. These inconsistent outcomes underscore that MASH resolution after VSG depends on multiple patient-specific factors¹².

The mechanisms underlying this clinical heterogeneity remain poorly understood. A fundamental question persists: do VSG’s benefits in MASH derive from weight-independent metabolic reprogramming? Moreover, the molecular determinants of therapeutic response remain undefined. Elucidating these mechanisms is essential for optimizing patient selection and developing strategies to enhance therapeutic efficacy.

To address these critical knowledge gaps, we designed a comprehensive translational study utilizing multiple preclinical MASH models and clinical cohorts. We systematically disentangle the weight-dependent versus weight-independent effects of VSG on MASH murine models. Unbiased multi-omic profiling was employed to identify molecular pathways mediating VSG efficacy. To investigate clinical relevance, we stratified distinct VSG patient cohorts to examine their metabolic signatures. Based on these mechanistic insights, we explored potential therapy in preclinical models that could enhance VSG efficacy. Our systematic approach reveals fundamental metabolic requirements for surgical efficacy and provides a foundation for optimizing VSG outcomes in MASH.

Results

VSG Ameliorates MASH Progression Through Weight-independent Mechanisms

To investigate the therapeutic efficacy of VSG on MASH, we first established a robust disease model. Mice fed the Gubra-Amylin-NASH (GAN) diet for 30 weeks developed hallmark features of human MASH¹³, including significant weight gain, elevated plasma alanine aminotransferase (ALT), aspartate aminotransferase (AST), total cholesterol (TC) and severe histopathological damage —reflected by a NAFLD Activity Score (NAS) of 4–6 and established liver fibrosis (ISHAK score = 1-2) (Supplementary Fig. 1A–F).

To dissect the weight-dependent from weight-independent effects of VSG, we designed four experimental groups: (1) sham-operated mice on ad libitum chow diet (sham-Chow/Ad); (2) sham-operated mice on ad libitum GAN diet (sham-GAN/Ad); (3) sham-operated mice on food-restricted GAN diet to match VSG-induced weight loss (sham-GAN/WM) and (4) VSG-operated mice on ad libitum GAN diet (VSG-GAN/Ad) (Fig. 1A).

Fig. 1. VSG ameliorates MASH progression independent of weight loss.

A–H GAN diet-induced MASH model (A) Schematic illustration of GAN diet-induced MASH model with VSG/sham interventions. (n = 10, 13, 13, 10 mice for sham-Chow/Ad, sham-GAN/Ad, VSG-GAN/Ad and sham-GAN/WM groups, respectively) (B, C) Quantification of plasma (B) alanine aminotransferase (ALT) and (C) aspartate aminotransferase (AST) levels across groups. D Representative hepatic histopathology showing H&E staining (upper panels) and Masson’s trichrome staining (lower panels). Scale bar = 200 μm. E NAFLD Activity Score (NAS) assessed on (H&E-stained liver sections across experimental groups. F Quantification of hepatic triglyceride content across groups. G Quantification of hepatic fibrosis area by Masson’s trichrome staining across groups. H Hepatic hydroxyproline content across groups. I–P GAN+CCl₄ induced MASH model (I) Schematic illustration of GAN + CCl₄ induced MASH model with VSG/sham interventions, (n = 8, 9, 13 mice per group for sham-Chow, sham-GAN + CCl₄, VSG-GAN + CCl₄ groups, respectively). J, K Quantification of plasma (J) ALT and (K) AST levels across groups. L Representative hepatic histopathology showing H&E staining (upper panels) and Masson’s trichrome staining (lower panels). Scale bar = 200 μm. M NAFLD Activity Score (NAS) quantification across groups. N Hepatic triglyceride content across groups. O Quantification of hepatic fibrosis area by Masson’s trichrome staining across groups. P Hepatic hydroxyproline content across groups. All data are presented as mean ± SEM. Statistical significance was determined by two-way ANOVA with multiple comparisons. p-values from statistical comparisons between indicated groups are shown above the bars.

Prior to surgical intervention, all GAN-fed groups displayed comparable metabolic dysfunction, with similarly elevated plasma ALT, AST, TC, and fasting glucose levels compared to chow-fed controls (Supplementary Fig. 2A–D). Following surgery, both VSG-GAN/Ad and sham-GAN/WM groups maintained sustained weight loss throughout the 12-week observation period, whereas sham-GAN/Ad mice regained weight (Supplementary Fig. 3A). Weight matching was confirmed by equivalent reductions in visceral adipose tissue mass and adipocyte size between VSG-GAN/Ad and sham-GAN/WM groups (Supplementary Fig. 3B–E). Notably, while the sham-GAN/WM group required ~ 15% caloric restriction to achieve weight loss comparable to VSG, the VSG-GAN/Ad mice maintained ad libitum feeding, demonstrating that VSG induces weight loss independent of reduced caloric intake (Supplementary Fig. 3F).

Consistent with the known benefits of weight loss, the weight-matched control group (Sham-GAN/WM) showed significant improvement in MASH pathology compared to ad libitum-fed controls (Sham-GAN/Ad), as evidenced by reductions in plasma ALT/AST, NAS score, and fibrosis markers (Fig. 1B–H), confirming that weight loss per se confers substantial therapeutic benefit. VSG demonstrated superior therapeutic efficacy compared to caloric restriction-induced weight loss alone. Twelve weeks post-intervention, VSG-GAN/Ad mice achieved near-complete normalization of liver enzymes, while the sham-GAN/WM mice showed only partial improvement ( ~ 60% reduction in plasma ALT and AST) (Fig. 1B, C). Histological analysis revealed that although both interventions (VSG-GAN/Ad, sham-GAN/WM) reduced NAS scores, the improvement was significantly more pronounced following VSG (Fig. 1D, E). Remarkably, VSG treatment dramatically resolved hepatic steatosis (p < 0.001 vs sham-GAN/Ad group), reducing liver triglyceride content to levels comparable to healthy Chow-fed controls (p > 0.05 vs. sham-Chow/Ad) (Fig. 1F). Most strikingly, VSG significantly reduced the mild-to-moderate liver fibrosis present in this model. This comprehensive reversal was evident across multiple validated metrics: quantitative collagen area (Fig. 1G), total hepatic hydroxyproline content (Fig. 1H), and histological scoring (Supplementary Fig. 4A)—all showing significantly greater improvements than the modest effects observed in weight-matched controls. Consistent with this phenotypic reversal, VSG potently suppressed hepatic expression of key pro-fibrotic genes (Acta2, Col1a1) and pro-inflammatory mediators (Il6, Ccl2, Tnf) (Supplementary Fig. 4B).

To further demonstrate that VSG’s hepatic benefits are independent of weight loss, we employed a combined GAN diet plus carbon tetrachloride (CCl₄) model (GAN + CCl₄), which induces severe MASH without obesity (Fig. 1I and Supplementary Fig. 5). In this lean MASH model, VSG maintained its therapeutic efficacy despite no weight change (Supplementary Fig. 6A, B), significantly improving glucose homeostasis (Supplementary Fig. 6C–E), reducing liver injury markers (Fig. 1J, K), and ameliorating steatosis, inflammation and fibrosis (Fig. 1L–P and Supplementary Fig 7). Similarly, VSG in healthy lean mice induced metabolic changes, including improved glucose tolerance and altered bile acid profiles, without affecting body weight (Supplementary Fig. 8).

Collectively, these data from three complementary mouse models demonstrate that VSG reverses established MASH pathology through mechanisms that extend beyond weight loss alone. While weight reduction undoubtedly contributes to metabolic improvements, our findings reveal that VSG activates additional metabolic pathways — particularly evident in the lean MASH model, where therapeutic benefits occurred without any weight change—suggesting that intrinsic metabolic reprogramming plays an important role in VSG’s therapeutic efficacy, the molecular basis of which we sought to elucidate.

VSG Enhances Hepatic One-Carbon Metabolism and Methylation Capacity in MASH

To systematically elucidate the metabolic alterations underlying the mechanism by which VSG ameliorates MASH, untargeted metabolic profiling was conducted on liver tissue from VSG-GAN + CCl₄ and sham-GAN + CCl₄ mice. Principal component analysis (PCA) showed a clear separation between groups (Fig. 2A), with 1270 differentially abundant metabolites identified (Fig. 2B), demonstrating the broad impact of VSG on hepatic metabolism.

Fig. 2. VSG enhances hepatic one-carbon metabolism through MAT1A upregulation in MASH.

A PCA score plot based on liver metabolite profiles in sham-GAN + CCl₄ and VSG-GAN + CCl₄ groups (n = 9/11). B Volcano plot of differential abundant metabolites between sham-GAN + CCl₄ and VSG-GAN + CCl₄ groups. Blue dots: downregulated metabolites; red dots: upregulated metabolites (C) Pathway enrichment analysis (SMPDB) of differentially abundant metabolites between sham-GAN+CCl₄ and VSG-GAN+CCl₄ groups. D Schematic of one-carbon metabolism showing methyl unit flow from various sources (methionine metabolism, betaine metabolism, glycine and serine metabolism), to downstream products including phosphatidylcholine biosynthesis and methylhistidine metabolism. E Heatmap showing Z-scores of methylated metabolites in sham-GAN+CCl₄ vs VSG-GAN+CCl₄ groups. F SAM/SAH ratio in liver tissue from sham-GAN + CCl₄ and VSG-GAN+CCl₄ groups (n = 9/11). G KEGG pathway annotation of differentially expressed genes between sham-GAN + CCl₄ and VSG-GAN+CCl₄ groups. H Enrichment analysis of one-carbon metabolism-related pathways RNA-seq data in GAN + CCl₄ model. I Heatmap showing differential expression (Z-score normalized) of genes involved in transsulfuration, folate and methionine pathways (n = 5 mice per group). Gene abbreviation. Cbs (cystathionine-beta synthase), Cth (cystathionine gamma-lyase), Gclc (glutamate-cysteine ligase catalytic subunit), Gclm (glutamate-cysteine ligase, modifier subunit), Gls1 (glutaminase 1), Gss (glutathione synthetase), Mat1a (methionine adenosyltransferase 1A), Ms (methionine synthetase). Shapes represent experimental groups: circles for the sham-GAN + CCl4 group and squares for the VSG-GAN+CCl4 group. Color intensity corresponds to the Z-score magnitude of mRNA expression. DHF, dihydrofolic acid; THF, tetrahydrofolate MTHF, L-methylfolate; SAM, S-adenosylmethionine; SAH, S-adenosylhomocysteine) (J) Western blots of MAT1A and GAPDH in liver samples from the sham-GAN + CCl₄ group and VSG-GAN+CCl₄ groups. The graph shows relative intensity quantification. K Western blot analysis of MAT1A and GAPDH in paired liver biopsies from patients pre- and post-VSG (Clinical Cohort 1). The graph shows relative intensity quantification. L MAT1A mRNA expression in public datasets (GSE113822, GSE68812, GSE125946) comparing sham vs bariatric surgery groups (n = 4/5, 5/4, 5/6 respectively). All data are presented as mean ± SEM (standard error of mean). Statistical significance was determined by an unpaired Student’s t test. P-values from statistical comparisons between indicated groups are shown above the bars. Representative images are shown from three independent experiments with similar results.

Pathway enrichment analysis using the small molecule pathway database (SMPDB) revealed striking changes in one-carbon metabolism related pathways (Fig. 2C). Among the top ten most significantly enriched pathways were glycine and serine metabolism, methionine metabolism, and betaine metabolism—all critical components of one-carbon metabolism. In addition, phosphatidylcholine biosynthesis and methylhistidine metabolism, which are associated with methylated end-products of S-adenosylmethionine (SAM)-dependent methylation reactions, were prominently enriched (Fig. 2D).

A substantial upregulation of various methylated metabolites was observed in the livers of VSG-GAN+CCl₄ mice, indicating an enhanced global methylation capacity (Fig. 2E). This enhancement was mechanistically linked to increased availability of one-carbon units for substrate methylation via SAM. The cycling between SAM and S-adenosylhomocysteine (SAH) represents the central hub of one-carbon metabolism and methyl transfer process¹⁴. Quantitative analysis revealed that VSG significantly elevated hepatic SAM contents and increased the SAM/SAH ratio (p = 0.0041), a critical indicator of methylation potential (Fig.2F and Supplementary Fig. 9A, B). Notably, metabolites in the folate and choline cycles—including 5,10-methylenetetrahydrofolate, 5-methyltetrahydrofolate, choline, betaine, serine and glycine—remained unchanged (Supplementary Fig. 9C–I), suggesting that enhanced methylation capacity primarily results from methionine cycle activation rather than altered folate or choline metabolism.

To validate these findings in humans, we analyzed paired liver biopsies from 4 patients with histologically-confirmed MASH obtained before and 12 months after VSG (Clinical Cohort 1, Supplemental Data 1). These patients showed marked clinical improvement post-VSG, with mean BMI reduction from 39.52 ± 3.81 to 29.99 ± 3.62 kg/m², normalization of liver enzymes (ALT: 63.25 ± 7.60 to 18.25 ± 5.98 U/L), and histological improvement (NAS: 4.50 ± 0.50 to 1.75 ± 0.25). Consistently with our mouse data, the liver samples from patients demonstrated a nearly two-fold increase in SAM content and SAM/SAH ratio post-VSG (Supplementary Fig. 10A–D), while showing no significant changes in cysteine, 5,10-methylenetetrahydrofolate (5,10-CH₂-THF) and 5-methyltetrahydrofolate (5-Methyl THF) levels (Supplementary Fig. 10E–G). These human data from the small cohort (n = 4) are consistent with our preclinical findings, though larger studies with paired biopsies are needed for definitive validation of these mechanisms in patients.

While paired liver biopsies are limited by feasibility, systemic metabolic changes can provide additional insights. To further validate the enhanced methylation capacity and examine its systemic manifestations, we analyzed plasma samples from a separate cohort of 11 patients with MASLD who underwent VSG (Clinical Cohort 2, Supplemental Data 2). These patients showed remarkable clinical improvements, including significant weight loss (BMI: 41.00 ± 2.31 to 31.02 ± 1.98 kg/m²), normalization of liver enzymes (ALT: 76.00 ± 24.81 to 16.73 ± 3.24 U/L), and complete resolution of hepatic steatosis as confirmed by ultrasound (11/11 patients). Plasma analysis from these patients showed significant increases in phospholipid biosynthesis intermediates and methylated compounds, including phosphatidylcholines (PC (38:4): + 27%, PC (20:5): + 35%) and methylated purines (1-methylguanine: + 12%) (Supplementary Fig. 11). These coordinated changes across species in both liver and circulation supported that VSG enhanced methylation capacity primarily through methionine cycle activation.

To identify the molecular mechanism driving these metabolic changes, we performed RNA sequencing analysis. KEGG pathways annotation revealed that VSG substantially modulated amino acid metabolism pathways in addition to lipid metabolism (Fig. 2G), with significant enhancement of one-carbon metabolism-related pathways (Fig. 2H). Transcriptomic profiling demonstrated selective upregulation of methionine cycle enzymes in VSG-GAN + CCl₄ mice, with notably pronounced increases in Mat1a, while genes in the folate cycle and transsulfuration pathway showed minimal or inconsistent changes (Fig. 2I).

Among the upregulated enzymes, methionine adenosyltransferase 1 A (MAT1A) —the rate-limiting enzyme catalyzing SAM synthesis—exhibited the most pronounced increase, which was confirmed at the protein level by Western blotting (Fig. 2J). Consistently with our murine data, MAT1A upregulation was also detected in human liver tissue from Clinical Cohort 1, suggesting that this mechanism may also operate in humans (Supplemental Data 1). In these paired liver biopsies, MAT1A expression showed consistent upregulation at both mRNA (Supplementary Fig. 12) and protein (Fig. 2K) levels following VSG, suggesting that the molecular mechanism identified in our mouse model may be operative in human MASH resolution, though large cohorts are needed for confirmation. Further, we analyzed publicly available transcriptomic datasets from the GEO repository: GSE113822, GSE68812 and GSE125946, which consistently showed MAT1A upregulation (Fig. 2L). Taken together, these integrated metabolomic and transcriptomic analyses from both mouse models and human clinical samples indicate that VSG enhances hepatic one-carbon metabolism through MAT1A-mediated activation of the methylation cycle, resulting in increased methylation capacity that may contribute to its therapeutic effects in MASH.

VSG Restores hepatic PC/PE Homeostasis to Alleviate ER Stress and Mitochondrial Dysfunction

The methylation of phosphatidylethanolamine (PE) to phosphatidylcholine (PC) represents a critical SAM-dependent reaction catalyzed by phosphatidylethanolamine N-methyltransferase (PEMT). This pathway maintains endoplasmic reticulum (ER) membrane integrity and mitochondrial function, with disruption strongly implicated in MASH pathogenesis^15,16 (Fig. 3A). Given our finding that VSG enhances SAM availability through MAT1A upregulation, we investigated whether VSG could restore PC/PE homeostasis and thereby ameliorate ER stress and mitochondrial dysfunction.

Fig. 3. VSG restores hepatic PC/PE homeostasis and alleviates ER stress and mitochondrial dysfunction.

A Schematic illustration showing how PC/PE imbalance due to impaired SAM-dependent methylation leads to ER stress and mitochondrial dysfunction. The illustration was created in BioRender. Zang, Y. (2026) https://BioRender.com/juq60sz. B Heatmap of individual PC/PE species ratios from hepatic lipidomic analysis comparing sham-GAN + CCl₄ and VSG-GAN + CCl₄ groups (n = 9, 10, respectively). C Total hepatic PC/PE ratio in sham-GAN + CCl₄, and VSG-GAN+CCl₄ groups (n = 9, 10 respectively). D Western blot analysis of ER stress markers (P-PERK, PERK, P-eIF2α, eIF2α) in liver samples from sham-Chow, sham-GAN+CCl₄ and VSG-GAN + CCl₄ groups (n = 3 per group). E Densitometric quantification of p-PERK/PERK and p-eIF2α/eIF2α ratios from panel (D). F Quantification of mitochondrial ROS (mROS) production in isolated hepatocytes from indicated groups (n = 3 per group). G Quantification of hepatic malondialdehyde (MDA) levels as a marker of lipid peroxidation (n = 6, 8, 8 for sham-Chow, sham-GAN + CCl₄, and VSG-GAN + CCl₄, respectively). H Quantification of mitochondrial DNA to nuclear DNA ratio indicating mitochondrial content in liver samples (n = 6, 8, 8 for sham-Chow, sham-GAN + CCl₄, and VSG-GAN + CCl₄, respectively). I Quantification of hepatic ATP content (n = 6, 8, 8 for sham-Chow, sham-GAN + CCl₄, and VSG-GAN + CCl₄, respectively). All data are presented as mean ± SEM (standard error of mean). Statistical significance was determined by two-way ANOVA with multiple comparisons. P-values from statistical comparisons between indicated groups are shown above the bars.

Lipidomic profiling of hepatic tissues from VSG-GAN+CCl₄ and sham-GAN+CCl₄ mice revealed significant alterations in phospholipid composition. Liquid chromatography-mass spectrometry analysis demonstrated that VSG substantially elevate the total hepatic PC/PE ratio (Fig. 3B, C). This finding was validated in liver samples from Clinical Cohort 1 (described above, Supplemental Data 1), where lipidomic analysis revealed increased total PC/PE ratios and specific PC/PE species ratios following VSG (Supplementary Fig. 13A, B), demonstrating that the restoration of phospholipid homeostasis observed in our mouse model occurs in parallel with clinical improvement in human patients with MASH.

Decreased hepatic PC/PE ratios compromise ER membrane integrity, triggering ER stress through activation of the protein kinase RNA-like ER kinase (PERK)/eukaryotic initiation factor 2 alpha (eIF2α) pathway and impairing mitochondria-associated membrane (MAM) function, ultimately leading to mitochondrial dysfunction and hepatocellular injury^17,18. To evaluate whether VSG restoration of PC/PE homeostasis alleviates these pathological processes, we assessed ER stress markers and mitochondrial function in liver tissue from sham-Chow, sham-GAN+CCl₄, and VSG-GAN+CCl₄ mice.

VSG treatment significantly suppressed ER stress signaling, as evidenced by marked reduction in both PERK phosphorylation (p-PERK/PERK ratio) and eIF2α phosphorylation (p-eIF2α/eIF2α ratio) compared to sham-GAN+CCl₄ (Fig. 3D, E). Consistently, VSG also attenuated the IRE1A/XBP1 and ATF6 signaling pathways (Supplementary Fig. 14A–D). These molecular changes were accompanied by improved cellular redox homeostasis. VSG substantially decreased mitochondrial reactive oxygen species (mROS) production (p = 0.0287 vs sham-GAN+CCl₄; Fig. 3F) and reduced hepatic malondialdehyde (MDA) levels, a marker of lipid peroxidation (p < 0.0001 vs sham-GAN+CCl₄; Fig. 3G).

Furthermore, this restoration of mitochondrial function was supported by an improvement in hepatic cellular redox balance, as VSG treatment tended to restore the diet-induced depletion of the NAD⁺/NADH ratio (p = 0.0523) and significantly attenuated the elevation of the NADP⁺/NADPH ratio (p = 0.0109) (Supplementary Fig. 15A, B). Metabolic flux analysis via the Seahorse XF Glycolysis Stress Test revealed that VSG decreased cellular glycolytic function, significantly decreasing non-glycolytic acidification, basal glycolysis, glycolytic capacity, and glycolytic reserve compared to the sham-GAN + CCl₄ group (Supplementary Fig. 15C–G). Collectively, these data demonstrate that VSG co-ordinately ameliorates ER stress, rebalances cellular redox states, and improves bioenergetic flexibility in the injured liver.

Remarkably, VSG not only prevented mitochondrial damage but actively enhanced mitochondrial biogenesis and function. Hepatic mitochondrial DNA (mtDNA) content showed a trend toward increase following VSG (p = 0.0908 vs sham-GAN + CCl₄; Fig. 3H), indicating enhanced mitochondrial biogenesis. This was functionally reflected in elevated ATP production (p = 0.0001 vs sham-GAN + CCl₄; Fig. 3I), with both parameters restored to levels comparable to healthy sham-Chow controls. This improvements in mitochondrial bioenergetics suggest that VSG promotes metabolic recovery beyond simple prevention of damage.

Collectively, these findings establish a mechanistic link between VSG-induced enhancement of one-carbon metabolism and restoration of cellular homeostasis. By increasing SAM availability through MAT1A upregulation, VSG normalizes the hepatic PC/PE ratio, thereby alleviating ER stress via suppression of the PERK/eIF2α axis and restoring mitochondrial function. This coordinated improvement in organellar health may contribute to VSG’s therapeutic efficacy in MASH.

MAT1A-Mediated one-carbon metabolism is essential for VSG’s therapeutic effects in MASH

To establish the causal role of one-carbon metabolism (1CM) in VSG’s therapeutic efficacy, we employed complementary dietary and genetic approaches. First, we utilized a choline-deficient, 0.1% Methionine, high-fat diet (CDAHFD) model that depletes essential 1CM substrates. Untargeted metabolomic profiling revealed global metabolic reprogramming in CDAHFD-fed mice, as evidenced by clear group separation in PCA analysis (Fig. 4A) and substantial numbers of differentially abundant metabolites (Fig. 4B). Pathway enrichment analysis identified significant alterations in multiple one-carbon-associated pathways, including purine metabolism, cysteine and methionine metabolism, and glycerophospholipid biosynthesis, reflecting a systemic disruption of 1CM flux, evident in both liver and plasma samples (Fig. 4A–C and Supplementary Fig. 16A, B). Key methylation-dependent metabolites, including phosphocholine and 1-methylguanine, were significantly reduced in both hepatic tissue (Supplementary Fig. 16C, D) and plasma (Supplementary Fig. 16E, F), indicating impaired hepatic methylation capacity.

Fig. 4. MAT1A-mediated one-carbon metabolism is required to improve MASH following VSG.

A–C Hepatic metabolomic profiling of diet-induced MASH. A PCA score plot of hepatic metabolite profiles comparing Chow diet vs CDAHFD groups. B Volcano plot of differentially abundant hepatic metabolites (Chow diet vs CDAHFD. ( | log2FC | > 2, p < 0.05). C Metabolic pathway enrichment analysis of differentially abundant hepatic metabolites (Chow diet vs CDAHFD). Bubble size represents pathway impact; color intensity represents -log₁₀(P-value). D–J VSG fails to ameliorate CDAHFD-induced MASH (n = 10, 10, 13 mice for sham-Chow, sham-CDAHFD, VSG-CDAHFD, respectively). D, E Quantification of plasma levels of (D) ALT and (E) AST. F Representative hepatic histopathology of H&E (upper) and Masson’s trichrome staining (lower), Scale ba = 200 μm. G NAFLD Activity Score (NAS) quantification across groups. H Hepatic triglyceride content across groups. I Quantification of hepatic fibrosis area by Masson’s trichrome staining across groups. J Quantification of hepatic hydroxyproline content across groups. (K–M) Hepatic metabolomic and ER stress analysis reveals limited effects of VSG under CDAHFD. K PCA plots comparing sham-CDAHFD vs VSG-CDAHFD hepatic metabolites. L Volcano plots of differential metabolites (sham-CDAHFD vs VSG-CDAHFD. |log2FC | > 1, p < 0.05). M Western blot analysis in liver tissues lysates. N–S Genetic knockdown of MAT1A abrogates the therapeutic benefits of VSG in the GAN+CCl₄ MASH model (n = 8 mice per group). N Representative hepatic histopathology showing H&E staining (upper panels) for general morphology and inflammation, and Masson’s trichrome staining (lower panels) for fibrosis assessment in AAV-Ctrl and AAV-shMAT1A treated mice following sham or VSG surgery. Scale bar = 200 μm. O NAS quantification across groups. P Quantification of hepatic fibrosis area by Masson’s trichrome staining across groups. Q Quantification of hepatic hydroxyproline content across groups. R Quantification of plasma ALT levels across groups. S Western blot analysis in liver tissues lysates. All data are presented as mean ± SEM. Statistical significance was determined by two-way ANOVA. P-values from statistical comparisons between indicated groups are shown above the bars. Representative images are shown from three independent experiments with similar results.

After two weeks of CDAHFD feeding to establish disease, mice underwent either VSG or sham surgery (Supplementary Fig. 17A). Strikingly, VSG failed to provide therapeutic benefit in the 1CM-deficient model. Biochemical and molecular markers confirmed VSG’s lack of efficacy in the CDAHFD model. Plasma ALT (Fig. 4D) and AST (Fig. 4E) levels remained equally elevated in both surgical groups. Markers of hepatocyte apoptosis, including TUNEL-positive cells (Supplementary Fig. 17B, C) and caspase 3 activity (Supplementary Fig. 17D), showed no improvement with VSG.

Histological examination revealed severe MASH pathology in CDAHFD-fed mice that was unresponsive to VSG intervention. H&E staining showed marked hepatocellular ballooning and macrovesicular steatosis in both VSG and sham groups (Fig. 4F). Blinded semi-quantitative scoring confirmed these observations, with VSG failing to improve the NAS scores or any individual histological parameter, including steatosis grade, lobular inflammation, and hepatocellular ballooning (Fig. 4G). Consistent with the histological findings, biochemical analysis revealed that hepatic triglyceride content, while markedly elevated compared to chow-fed controls, showed no difference between VSG and sham-operated mice under CDAHFD conditions (Fig. 4H). The inflammatory response to CDAHFD was similarly unaffected by VSG. Hepatic macrophage infiltration, assessed by F4/80 immunostaining, remained comparable between VSG and sham-operated CDAHFD groups (Supplementary Fig. 17E, F). Pro-inflammatory genes— including Tnf, Il6, Ccl2, and Il1β — was equally evaluated in both surgical groups (Supplementary Fig. 17G). Furthermore, VSG failed to attenuate CDAHFD-induced fibrosis. Both VSG and sham groups showed similarly elevated collagen deposition (Fig. 4I) and hydroxyproline content (Fig. 4J) compared to chow-fed controls, along with upregulated expression of fibrotic genes including Acta2, Col1a1, Fn1, Tgfb1, and Ctgf (Supplementary Fig. 17H).

Untargeted metabolomic profiling of liver and plasma samples further revealed minimal differences between VSG-CDAHFD and sham-CDAHFD groups, as indicated by overlapping PCA plots (Fig. 4K and Supplementary Fig. 18A) and sparse differential metabolites in volcano plot analysis (Fig. 4L and Supplementary Fig. 18B). Western blot analysis of hepatic stress markers (e.g., p-eIF2α) confirmed that VSG did not alleviate hepatic ER stress in the CDAHFD model, with p-PERK, p-eIF2α levels remaining elevated (Fig. 4M). Notably, despite the lack of therapeutic efficacy in this model, VSG retained its characteristic ability to elevate circulating bile acids (Supplementary Fig. 19), indicating that VSG was technically successful but therapeutically ineffective in the absence of intact 1CM^19,20.

To specifically test whether MAT1A is required for VSG’s therapeutic effects, we performed genetic loss-of-function experiments using the GAN + CCl₄ induced MASH model. Adeno-associated virus (AAV)-mediated delivery of shRNA achieved approximately 80% knockdown of hepatic MAT1A expression (Supplementary Fig. 20). MAT1A knockdown abolished the therapeutic effects of VSG across multiple parameters. In control mice treated with scrambled shRNA (Scr-AAV), VSG significantly ameliorated MASH pathology as expected, reducing both histological damage and NAS scores. However, in MAT1A-deficient mice (shMAT1A-AAV), VSG failed to improve hepatic histopathology or reduce NAS scores, which remained as severe as sham-operated controls (Fig. 4N, O). Quantitative analysis of Masson’s trichrome staining (Fig. 4P) and hepatic hydroxyproline content (Fig. 4Q) revealed that VSG did not attenuate collagen deposition—a key marker of fibrosis—in GAN + CCL4- MAT1A-deficient mice. Beyond histological parameters, MAT1A deficiency prevented VSG’s key molecular and biochemical benefits. The VSG’s ability to reduce plasma ALT levels was abrogated in shMAT1A mice (Fig. 4R). Gene expression analysis revealed that MAT1A knockdown abolished VSG’s suppressive effects on inflammatory (Il6, Ccl2, Tnf) and fibrotic (Acta2, Col1a1) markers (Supplementary Fig. 21). Furthermore, MAT1A deficiency prevented VSG from alleviating ER stress, as p-PERK and p-eIF2α levels remained elevated following VSG in shMAT1A mice (Fig. 4S).

These complementary approaches—dietary depletion of 1CM substrates and genetic ablation of the rate-limiting enzyme—demonstrate that MAT1A-dependent one-carbon metabolism is essential for VSG’s therapeutic efficacy in MASH. The loss of VSG’s benefits in both models identifies enhanced methylation capacity as a critical mechanism mediating VSG’s therapeutic effects.

To definitively test whether the therapeutic benefits of VSG are mediated through the functional product of MAT1A, we performed a rescue experiment in MAT1A-knockdown mice. Genetic knockdown of MAT1A via AAV-shMAT1A abolished the therapeutic effects of VSG on liver injury, steatosis, and fibrosis in the GAN + CCl₄ model. Crucially, supplementing these MAT1A-deficient mice with S-adenosylmethionine (SAM) largely restored their responsiveness to VSG. The combination of VSG and SAM in AAV-shMAT1A mice significantly improved all measured parameters—including plasma ALT/AST levels, hepatic triglyceride content, NAFLD Activity Score, and fibrosis markers—to an extent comparable to the efficacy of VSG in control (AAV-Scramble) mice (Supplementary Fig. 22). This rescue experiment provides direct genetic evidence that the functional output of MAT1A, namely adequate SAM production, is the essential mediator through which VSG exerts its hepatoprotective effects.

Dysregulated one-carbon metabolism is associated with impaired hepatic improvement following VSG in human responders

To translate our mechanistic findings to clinical relevance, we investigated whether 1CM status is associated with hepatic outcomes in patients undergoing VSG for obesity. We analyzed a third cohort (Clinical Cohort 3, Supplementary Data 3) of 42 patients with obesity and hepatic disease who underwent VSG. Based on ultrasonographic findings at 12 months post-VSG, patients were classified as responders (n = 20, showing resolution of hepatic steatosis) or poor-responders (n = 22, showing persistent steatosis) (Fig. 5A).

Fig. 5. Patients with MASLD who respond poorly to VSG exhibit preoperative impairment of one-carbon metabolism.

A Clinical characteristics of responders (n = 20) and poor-responders (n = 22) at baseline and 12 months post-VSG. Both groups achieved similar weight loss ( ~ 23-24% BMI reduction) but differed significantly in hepatic outcomes. B PLS-DA score plot of pre-VSG plasma metabolites profiles distinguishing responders (blue) from poor-responders (red). C SMPDB pathway enrichment analysis of differentially abundant metabolites between pre-VSG plasmas from responders and poor-responders. Circle size: enrichment ratio; color intensity: -log₁₀(P-value). Red text indicates one-carbon metabolism-related pathways. D Linear regression analysis of baseline plasma SAM concentrations versus CK18 ratio (Post-VSG/ Pre-VSG), a potential noninvasive biomarker for MASH. E–I Plasma concentrations of one-carbon metabolites at pre-VSG: (E) SAM, (F) choline, (G) phosphocholine, (H) betaine, and (I) 1-methylguanine in responders vs poor-responders (n = 20/22). J Venn diagram showing overlapping dysregulated metabolic pathways between responder vs poor-responder (human) and chow diet vs CDAHFD (mouse). K Correlation between plasma and hepatic metabolism profiles in mouse models, validating plasma metabolomics as a surrogate for hepatic metabolic status (R² = 0.645, p < 0.0001). All data are presented as mean ± SEM. Statistical significance was determined by a two-tailed Unpaired Student’s t test. P-values from statistical comparisons between indicated groups are shown above the bars.

Baseline characteristics were comparable between the two groups, including age distribution, sex distribution, BMI, glycated hemoglobin, fasting glucose levels, and liver enzymes (ALT, AST), ensuring baseline comparability prior to intervention (Fig. 5A and Supplementary Data 3).

Importantly, both groups achieved comparable weight loss following VSG: BMI decreased from 41.91 ± 1.56 to 30.14 ± 1.43 kg/m2 in responders and from 38.50 ± 1.71 to 29.67 ± 1.46 kg/m² in poor-responders (p > 0.05 for between-group comparison), representing ~ 24% and ~ 23% BMI reduction, respectively. Despite this similar weight loss, hepatic outcomes differed dramatically. Responders showed significant improvement in liver enzymes with ALT decreasing from 115.3 ± 13.4 to 24.15 ± 7.21 U/L and AST from 63.10 ± 6.76 to 18.05 ± 2.17 U/L (both p < 0.05), while poor-responders maintained elevated levels (ALT: 116.0 ± 13.35 to 77.36 ± 3.38 U/L; AST: 61.68 ± 4.84 to 48.55 ± 3.55 U/L). The difference between responders and poor-responders at 12 months was significant for both ALT and AST (p < 0.0001) (Fig. 5A).

To identify baseline metabolic determinants that predict divergent treatment response despite similar weight loss, we performed comprehensive metabolomic profiling on pre-operative plasma samples from both groups. Partial least squares discriminant analysis (PLS-DA) revealed distinct metabolic signatures separating responder from poor-responder patients (Fig. 5B), indicating fundamental metabolic differences underlying therapeutic response. Pathway enrichment analysis identified striking alterations in 1CM pathways among poor-responder patients, with Methylhistidine Metabolism, Glycine and Serine Metabolism, Phosphatidylcholine Biosynthesis, and Betaine Metabolism ranking among the most significantly dysregulated pathways (Fig. 5C).

Critically, specific one-carbon metabolites showed associations with hepatic response to VSG. Pre-VSG plasma SAM levels showed a significant negative correlation with the post- to pre-operative CK18 ratio, a potential non-invasive MASH biomarker²¹, suggesting that higher baseline one-carbon metabolite levels are associated with better therapeutic outcome (Fig. 5D). Consistent with this, pre-VSG plasma from poor-responder patients exhibited significantly lower levels of key one-carbon metabolites compared to responder patients, including SAM (p = 0.0104; Fig. 5E), choline (p = 0.0325; Fig. 5F) and phosphocholine (p = 0.0466; Fig. 5G). In addition, poor-responder patients showed trend-level reductions in betaine (p = 0.0790) (Fig. 5H) and methylated purines (1-methylguanine p = 0.0735) (Fig. 5I).

To validate the clinical relevance of our preclinical findings, we compared metabolic perturbations between the plasma of CDAHFD mice (as shown in Supplementary Fig. 16A, B) and human poor-responder patients. Both species exhibited overlapping pathway dysregulation across ten metabolic pathways (Fig. 5J), with prominent involvement of one-carbon metabolism components, including glycine and serine metabolism, methionine metabolism, and betaine metabolism. To confirm that plasma metabolomics accurately reflects hepatic metabolic status, we correlated plasma and liver metabolite profiles in our mouse models, revealing a strong statistical association (R² = 0.645; p < 0.0001) (Fig. 5K). This finding, consistent with previous human studies^17,18, validates plasma metabolomics as a reliable surrogate for hepatic metabolic assessment.

Taken together, these clinical findings reveal that patients who fail to achieve hepatic improvement following VSG display a metabolic signature characterized by impaired one-carbon metabolism, despite achieving similar weight loss to responders. The reduced levels of key methylation metabolites (SAM, choline, phosphocholine) in poor-responders mirror the metabolic deficiencies observed in our CDAHFD mouse model, where dietary depletion of these same substrates abolished VSG’s therapeutic efficacy. These parallel findings across species provide compelling translational evidence that adequate one-carbon metabolic capacity is essential for VSG’s hepatic benefits, suggesting that baseline or persistent deficiencies in methylation metabolism may identify patients less likely to achieve liver improvement despite successful weight loss surgery.

Methyl donor restores VSG efficacy in CDAHFD-fed mice

Having established that one-carbon metabolism deficiency abolishes VSG’s therapeutic effects, we investigated whether supplementation with methyl donors could restore VSG efficacy. We selected betaine, a stable methyl donor that directly feeds into the methionine cycle²², to test whether replenishing one-carbon units could rescue VSG’s therapeutic efficacy in the CDAHFD model. Mice were treated according to the protocol outlined in Fig. 6A, receiving either betaine alone, VSG alone, or combined VSG + betaine treatment.

Fig. 6. Methyl donor betaine supplementation restore efficacy of VSG.

A–I mice received chow or CDAHFD for 2 weeks, underwent VSG or sham surgery, then received betaine supplementation (100 mg/kg in saline) or vehicle for 8 weeks post-operatively (n = 8, 10, 10, 7, 7 for sham-Chow, sham-CDAHFD, sham-Betaine-CDAHFD, VSG-CDAHFD, VSG-Betaine-CDAHFD group, respectively). A Experimental design. B, C Plasma levels of (B) ALT, and (C) AST at 8 weeks post-surgery. D Representative H&E (upper panels) and Masson’s trichrome staining (lower panels) of liver sections. Scale bar = 200 μm. E NAFLD Activity (NAS) Score quantification across groups. F Hepatic triglyceride content across groups. G Quantification of hepatic fibrosis area from Masson’s trichrome staining. H Hepatic hydroxyproline content as a quantitative measure of collagen deposition. I Hepatic mRNA expression of fibrosis-related genes (Acta2, Col1a1, Fn1) and inflammatory cytokines (Tnf, Il6). Expression normalized to Gapdh (n = 6/10/7/7/7, respectively). All data are presented as mean ± SEM. Statistical significance was determined by two-way ANOVA. P-values from statistical comparisons between indicated groups are shown above the bars.

Striking therapeutic synergy was observed when betaine supplementation was combined with VSG. While VSG alone failed to improve markers of liver injury in CDAHFD-fed mice, the combined treatment significantly reduced plasma ALT and AST levels (Fig. 6B, C). Crucially, this effect significantly exceeded the partial improvements from betaine monotherapy, confirming that the combination therapy achieved significantly greater efficacy than betaine monotherapy (p < 0.05 for ALT, VSG + betaine vs. betaine alone).

The VSG + betaine treatment reduced the NAS scores to an average of 2, representing substantial histological improvement, while betaine alone partially improved scores to ~ 3.5 and VSG alone remained ineffective ( ~ 5, similar to CDAHFD control) (Fig. 6D, E). Individual NAS components—steatosis, inflammation, and ballooning—all showed significant improvement with combination therapy, demonstrating comprehensive histological rescue (Fig. 6D, E). Hepatic triglyceride content demonstrated a clear treatment hierarchy (Fig. 6F). VSG alone had minimal effect, betaine alone provided moderate reduction, but the VSG+betaine combination achieved the greatest decrease in hepatic lipid accumulation.

Most notably, the combination therapy produced a substantial reduction in hepatic fibrosis. Quantitative analysis of Masson’s Trichrome staining (Fig. 6D, G) showed that VSG + betaine combination (~ 3%) significantly reduced fibrotic area compared to VSG alone or CDAHFD controls (~ 7%), approaching the baseline levels of healthy chow-fed controls (~ 2%). Similarly, hepatic hydroxyproline content, a quantitative measure of collagen deposition, was markedly decreased by combination therapy, confirming the anti-fibrotic effects (Fig. 6H). Gene expression analysis revealed that VSG + Betaine treatment effectively normalized the key pathogenic markers examined. The combination therapy reduced the expression of pro-fibrotic genes (Acta2, Col1a1, Fn1) and inflammatory mediators (Il6, Tnf) to near-baseline levels, an effect not achieved by betaine monotherapy (Fig. 6I).

We next hypothesized that directly supplementing S-adenosylmethionine (SAM), the end-product of MAT1A, could bypass the methyl-donor deficiency and restore the therapeutic response to VSG. Indeed, in CDAHFD-fed mice, where VSG alone showed limited benefit, co-administration of SAM largely restored its efficacy. This combination therapy (VSG + SAM) ameliorated MASH pathology, resulting in significant reductions in plasma ALT and AST levels (Supplementary Fig. 23A, B), hepatic triglyceride content (Supplementary Fig. 23C), and histological disease activity (NAS score) (Supplementary Fig.23D, E). Furthermore, it profoundly attenuated hepatic fibrosis, as evidenced by decreased collagen deposition (Masson-positive area) and hydroxyproline content (Supplementary Fig.23F, G). These data demonstrate that in the context of impaired one-carbon metabolism, restoring the hepatic SAM pool can largely unlock the full therapeutic potential of VSG, directly implicating that the functional output of the MAT1A pathway is the critical mediator.

These findings demonstrate that methyl donor supplementation can rescue VSG efficacy in the CDAHFD model, where one-carbon substrates are depleted. The synergy between betaine/SAM and VSG—transforming an ineffective surgery into an effective intervention—confirms the requirement for intact methylation capacity in VSG’s mechanism of action.

Discussion

The beneficial effects of bariatric surgeries against obesity-related diseases are often simplistically attributed to mechanical restriction and nutrient malabsorption. However, accumulating evidence suggests these metabolic surgeries exert profound effects on multiple physiological functions, including bile acid–FXR activation²³, altered lipid absorption²⁴, and modulation of gastrointestinal hormones²⁵. Prior metabonomics studies have also reported that VSG decreased circulating branched-chain amino acids (BCAAs) and increased levels of licoricidin, short-chain fatty acids (SCFAs) and bile acids.

Our study reveals that VSG ameliorates MASH through mechanisms that appear to extend beyond simple weight reduction. Using paired weight-matching experiments in murine models, we demonstrate that VSG achieves superior therapeutic efficacy compared to caloric restriction alone, particularly in ameliorating fibrosis—a notoriously treatment-resistant pathological feature. In the GAN diet model, while both VSG and weight-matched controls showed comparable weight loss and adipose tissue reduction, VSG achieved superior normalized liver enzymes and more pronounced resolution of hepatic steatosis and fibrosis. We further note that VSG significantly increased whole-body energy expenditure compared to sham and weight-matched controls (Supplementary Fig. 3H, I). Independently, the present study identifies the LCA–VDR–MAT1A one-carbon metabolism axis as a candidate hepatic mechanism: AAV-mediated MAT1A knockdown significantly attenuated VSG’s hepatic benefits, while SAM supplementation largely restored them. The GAN+CCl₄ model provided additional evidence, as VSG improved MASH pathology despite unchanged body weight throughout the experimental period. Recent work by Fredrickson et al. also demonstrated weight-independent benefits of VSG using a 14-week HFHC diet model, identifying TREM2+ macrophages as important mediators²⁶. Our study, employing longer-duration GAN feeding (30 weeks) and the GAN+CCl₄ combination model, sought to identify the metabolic mechanisms underlying VSG’s superior therapeutic efficacy.

Our data suggest that enhanced methylation capacity, particularly through MAT1A upregulation, may contribute to VSG’s weight-independent therapeutic effects. While we cannot definitively establish MAT1A as the sole mechanism—other pathways, including bile acid signaling, likely play important roles—it is also possible that pharmacological strategies to upregulate or activate hepatic MAT1A could represent a promising non-surgical approach to improve MASH. The convergence of evidence from metabolomics, loss-of-function studies, and betaine rescue experiments indicates that intact one-carbon metabolism appears necessary for optimal VSG efficacy. The distinct one-carbon metabolic signatures observed in human responders versus poor responders are consistent with this mechanism, suggesting its potential clinical relevance. The enhanced SAM availability may restore hepatic PC/PE homeostasis through PEMT-mediated phospholipid methylation, potentially alleviating ER stress and restoring mitochondrial function—processes critically linked to MASH progression. Notably, the role of MAT1A/SAMe is context-dependent: its reduction during fasting permits enhanced fatty acid oxidation in healthy liver¹⁵, whereas its VSG-induced restoration in MASH repairs disrupted methylation capacity and mitochondrial function. Our findings are in genuine conflict with a previous report showing that MAT1A inhibition is beneficial in diet-induced or genetically obese models²⁷. This conflict is not easily resolved by invoking differences in the disease model. The clinical observation that patients who underwent VSG exhibited higher hepatic SAM levels after surgery further underscores the complexity of this discrepancy. The underlying basis remains to be determined.

Both dietary depletion of one-carbon substrates and genetic MAT1A knockdown abolished VSG’s benefits, while betaine supplementation rescued efficacy, suggesting that methylation capacity may be both necessary and potentially targetable. These mechanistic insights could help explain the heterogeneous clinical responses we observed: despite nearly identical weight loss (24% vs 23% BMI reduction), responders and poor-responders showed divergent hepatic outcomes. Their distinct one-carbon metabolic signatures suggest that the capacity for this metabolic adaptation, rather than weight loss per se, might influence therapeutic success. This distinction from simple caloric restriction could guide patient selection and therapeutic optimization, though larger studies are needed to confirm these mechanisms.

While weight loss undoubtedly contributes to metabolic improvements following VSG, our integrated preclinical and clinical data suggest that one-carbon metabolism may represent an additional, essential mechanism. This weight-independent pathway could be particularly relevant for patients with compromised methylation capacity who might otherwise show suboptimal responses despite weight loss.

The variable clinical response to bariatric surgery has long puzzled clinicians. Our findings provide a possible molecular explanation: patients with compromised one-carbon metabolism may have reduced capacity to mount the methylation response that appears important for surgical efficacy. Supporting this hypothesis, poor-responders displayed distinct one-carbon metabolic signatures, with baseline plasma SAM levels inversely correlating with post-VSG hepatic steatosis. The marked differences in liver enzyme normalization between responders and poor-responders, despite similar weight loss, suggest that intact one-carbon metabolism may be important for resolving hepatic injury. This finding highlights the potential relevance of our mechanistic insights across the spectrum of MASH pathogenesis.

Multiple factors may contribute to compromised one-carbon metabolism in patients with MASH: dietary choline deficiency (affecting up to 25% of Americans), genetic polymorphisms in PEMT (rs7946), MTHFD1 (rs2236225), and MAT1A (rs3851059), and surgery-induced micronutrient deficiencies. These factors might converge to create a “methylation-deficient” phenotype potentially resistant to surgical intervention. Our betaine supplementation experiments provide proof-of-concept for addressing this limitation—betaine bypasses these deficiencies to potentially restore methylation capacity, transforming ineffective surgery into a robust therapeutic response in mice.

These findings suggest a potential precision medicine framework warranting clinical investigation. Preoperative assessment of one-carbon metabolic status might help identify patients who could benefit from supplementation, though several important caveats must be acknowledged. Betaine supplementation may interact with post-surgical micronutrient absorption and could serve as a substrate for de novo lipogenesis, potentially exacerbating hepatic steatosis. Any clinical application must therefore be preceded by well-controlled trials to assess both efficacy and safety.

Our study further delineates a potential upstream pathway by which VSG may signal to the liver to enhance one-carbon metabolism. To dissect the VSG-specific component of MAT1A induction, we compared MAT1A expression across all four experimental groups. VSG induced significantly greater MAT1A upregulation than food restriction alone (Supplementary Fig. 24A, B), and this additional induction was not attributable to glucagon, as plasma glucagon levels were significantly lower in VSG than in weight-matched controls despite higher MAT1A expression (Supplementary Fig. 24C). We propose a lithocholic acid (LCA)-vitamin D receptor (VDR) axis as a candidate transcriptional mechanism linking VSG to MAT1A induction. This hypothesis is supported by the prediction of VDR binding sites within the MAT1A promoter (Supplementary Fig. 24D) and by VSG-induced elevation of portal LCA levels (Supplementary Fig. 24E) — consistent with the established role of LCA as a VDR ligand. The biological effects of bile acids are highly concentration-dependent, and this dichotomy — where a metabolite exerts opposing effects depending on concentration and pathophysiological context — is well-established in bile acid biology. The MAT1A suppression reported in cholestatic liver injury, where bile acid concentrations reach cytotoxic levels, is therefore pathophysiologically distinct from the modest portal LCA elevation observed following VSG. Although VSG did not significantly alter hepatic Cyp8b1 expression in our model (data not shown) — likely because Cyp8b1 is already markedly suppressed in the context of established MASH at baseline — the elevated portal LCA levels observed after VSG may reflect enhanced intestinal reabsorption, consistent with alterations in bile acid transport reported following bariatric surgery²⁸. We acknowledge, however, that direct in vivo experimental evidence formally demonstrating LCA-mediated VDR activation of MAT1A transcription is currently lacking. The definitive validation of this proposed axis, including assessment of VDR activity in vivo and its causal relationship to MAT1A induction, remains an important goal for future investigation.

Several important limitations warrant consideration. First, our mechanistic insights derive primarily from murine models, and the optimal timing, dosing, and duration of methyl donor supplementation cannot be directly extrapolated to humans. Second, while we studied three distinct human cohorts containing 56 patients, the cohort with paired liver biopsies is small (n = 4), limiting the statistical power for mechanistic validation in humans. While baseline MASH was confirmed by biopsy in a subset of Clinical Cohort 3, post-operative assessment was performed using ultrasonography due to ethical constraints. Although this approach cannot evaluate inflammation or hepatocellular ballooning, the marked differences in steatosis resolution and liver enzyme normalization between responders and poor-responders provide clinically relevant endpoints. It should also be noted that energy expenditure was not directly measured in AAV-shMAT1A mice in the present study. The potential influence of hepatic MAT1A deficiency on VSG-induced whole-body thermogenesis cannot be formally excluded and warrants further investigation. Third, we cannot exclude contributions from other VSG-induced changes—including altered gastric hormones, bile acid signaling, or microbiome shifts—that may interact with one-carbon metabolism. Future studies incorporating non-invasive markers of inflammation and fibrosis, larger cohorts with extended follow-up, and controlled supplementation trials are needed to validate and translate these findings.

Future studies should address several critical questions. The relative contribution of one-carbon metabolism compared to established VSG mechanisms needs quantitative assessment. Whether other bariatric procedures similarly depend on one-carbon metabolism requires systematic evaluation. The potential for epigenetic modifications downstream of SAM-dependent methylation to mediate long-term benefits also warrants investigation. Most importantly, prospective clinical trials should test whether preoperative metabolic profiling can identify patients who would benefit from methyl donor supplementation.

In conclusion, we identify MAT1A-driven one-carbon metabolism as a potentially important, weight-independent mechanism for VSG efficacy in MASH. The convergence of evidence—from MAT1A upregulation in human liver biopsies to distinct metabolic signatures associated with clinical response—suggests that one-carbon metabolism may serve as both a candidate biomarker and therapeutic target. While our preclinical models demonstrate that methyl donor supplementation can rescue failed surgical responses, translation to clinical practice awaits validation. These findings may help explain why VSG succeeds or fails in different patients, providing a potential metabolic framework for personalized surgical approaches to MASH therapy.

Methods

Human study protocol

The study protocol was approved by the Ethics Committee of Huashan Hospital, Fudan University, with the approval number KY2020-878. The study was registered at ClinicalTrials.gov under the identifier NCT04366999. Written informed consent was obtained from all participants in each cohort before enrollment and sample collection, in accordance with the approved study protocol.

Patients undergoing primary VSG were recruited from the Center for Obesity and Metabolic Surgery, Huashan Hospital of Fudan University. All patients met the criteria for metabolic and bariatric surgery as per the 2019 Chinese Society for Metabolic and Bariatric Surgery Guidelines. Those with a history or current clinical evidence of drug-induced liver injury, alcoholic liver injury, viral hepatitis infection, genetic liver disease, and autoimmune liver disease were excluded from the study. All participants provided written informed consent before enrollment.

In the Cohort 1 study, we evaluated 4 liver tissues form subjects with histological-proven MASH before and after VSG. The clinical characteristics of the 4 subjects are listed in Supplemental Data 1.

In the Cohort 2 study, we included 11 plasma samples from individuals with MASLD collected before and 12 months after VSG. MASLD was diagnosed by hepatic ultrasound results, excluding significant alcohol consumption. The clinical characteristics of the Cohort 2 study are listed in Supplemental Data 2.

In the Cohort 3 study, we included 42 participants with obesity and MASLD who underwent VSG. Based on ultrasonographic findings at 12 months post-VSG, patients were classified as responder (n = 20, showing resolution of hepatic steatosis) or poor-responder (n = 22, showing persistent steatosis despite significant weight loss). Both groups achieved comparable weight loss (approximately 23 ~ 24% BMI reduction). The clinical characteristics of the Cohort 3 study are listed in Supplemental Data 3.

Peripheral blood samples were collected from all participants at 7am after overnight fasting prior to the surgery and at 12 months post-surgery. During laparoscopic VSG, liver biopsy specimens of 1 × 1 x 1 cm were obtained when clinically indication for MASLD/MASH pathological diagnosis, and MASH activity was graded by a single expert pathologist. For Cohort 3, plasma CK18 levels were measured as a non-invasive biomarker to assess the ratio of post-operation to pre-operation values for correlation with treatment response.

All relevant regulations concerning the collection and use of data and samples from human participants were followed in accordance with the criteria of the Declaration of Helsinki.

Animal study

The Animal Ethics Committee of the Shanghai Institute of Materia Medica approved all animal experiments and protocols conducted in this study with the approval number 2021-12-LJ-12. The mice were housed in isolated ventilated cages in an animal barrier facility at the Shanghai Institute of Materia Medica (Shanghai, China), where they were maintained on a 12/12 h light/dark cycle at 22–26 °C with access to sterile pellet food and water ad libitum. Adult male mice were used for the respective experiments as per their age. To establish the mouse MASH model, the following several modeling methods were employed,

GAN Mice model of MASH²⁹

Male C57BL/6 J mice (6–8 weeks old) were purchased from Zhejiang Vital River Laboratory Animal Technology. The mice were fed a GAN diet containing 40 kcal% Fat (Mostly Palm Oil), 20 kcal% Fructose and 2% Cholesterol (D09100310, Research Diet Inc., USA). Starting at week 30, the mice fed the GAN diet were randomly assigned to either the sham or VSG group. The amount of GAN diet fed to the sham-GAN/WM group was the mean consumption ( 85%) of the VSG-GAN/Ad group from the preceding day.

GAN + CCl₄-Induced MASH Mouse Model^30,31

Male C57BL/6 J mice, aged 6 weeks, the mice were fed a GAN diet for 4 weeks, followed by intraperitoneal injections of carbon tetrachloride (CCl₄) once weekly for 8 weeks to induce MASH. The CCl₄ dose was initially 0.05 mL/kg for once, and this was increased to 0.1 mL/kg for the subsequent four administrations. From weeks 5 − 9, mice were sorted into sham/VSG groups

CDAHFD Induced MASH mouse model³²

Male mice aged 6-8 weeks were fed a CDAHFD diet (45% fat with low methionine and no choline; A06071309; Research Diets, New Brunswick, USA). Mice fed with the normal control diet (Q031, Shanghai Shilin Biologic Science & Technology, Shanghai, China) were used as the control group. After 2 weeks of feeding, the mice fed the CDAHFD diet were randomly assigned to either the sham or VSG group. The betaine administration experiment was initiated after the second week of CDAHFD feeding. Betaine was dissolved in saline and administered daily at a dose of 100 mg/kg. SAM was dissolved in saline and administered daily at a dose of 30 mg/kg.

The VSG and sham surgeries were performed on all mice using a previously described^33,34. Briefly, mice were fasted for 6 hours before surgery. A titanium clip (Johnson Ethicon, Somerville, NJ) was used parallel to the smaller curvature, transecting 80% of the stomach, including the entire glandular part. Suturing was performed using 6-0 Prolene (Johnson Ethicon, Cincinnati, OH, USA). Sham surgery involved exposing the stomach and performing blunt clamping without any incision.

Cell culture

Primary mouse hepatocytes were isolated using Selgen’s two-step perfusion method and maintained in low-glucose DMEM supplemented with 10% FBS at 37 °C in a humidified incubator in a 95% air and 5% CO₂ atmosphere.

AAV experiment

For liver-specific gene modulation, male C57BL/6 J mice received tail-vein injections of adeno-associated virus serotype 8 (AAV8) vectors. For MAT1A knockdown, male C57BL/6 J mice first received AAV8-TBG-shMAT1A before one week on the GAN diet. Control groups received AAV8-scrambled-shRNA (AAV-shCtrl). All viral vectors were sourced from Genomeditech.

Histological analyses

Liver sections were processed for histological analysis to evaluate lipid accumulation, inflammation, fibrosis, and injury. To visualize the pattern of lipid accumulation and inflammatory status, paraffin-embedded liver sections were stained with hematoxylin and eosin (H&E) (Wuhan Servicebio Technology). Masson staining was performed by Wuhan Servicebio Technology Co., Ltd. The collagen-stained area quantification was performed with a positive pixel count algorithm using PerkinElmer software. Liver injury was measured by TUNEL staining using a kit from Sigma-Aldrich in Germany. The histological features of the tissues were observed under a light microscope from Olympus and imaged. The severity of liver steatosis, inflammation, and fibrosis was blindly assessed by peers. The NAS scoring system was used to quantify liver steatosis (scale of 0-3), lobular inflammation (scale of 0-3), and hepatocellular ballooning (scale of 0-2).

Untargeted metabolomics and lipidomics profiling

Full description can be found in the supplementary information. Briefly, liver tissues were firstly homogenized in 5-fold volume of water, and then sonicated for 10 min. Aliquots from each plasma or liver homogenate sample were pooled together as quality control (QC) for plasma or liver, respectively. For untargeted metabolomic analysis, 120 μL methanol/acetonitrile (1:1, v/v) was added to each 40 μL of plasma, liver homogenate or QC sample, and then the mixture was vortexed for 10 min. Subsequently, the mixture was centrifuged at 15000 rpm for 10 min at 4 °C. Eighty microliters of the supernatant were lyophilized and stored at − 80 °C prior to UPLC-Q-orbitrap mass spectrometer (MS) analysis. Lyophilized samples were reconstituted in acetonitrile/water (1:1, v/v) for the HSS T3 column and acetonitrile/water (9:1, v/v) for the BEH HILIC column. For lipidomics analysis, 300 μL of cold methanol containing lipid IS (consisting of d27-FA(14:0) LPE(17:1), LPC(17:0), PC(19:0/19:0), PE(17:0/17:0), PG(17:0/17:0), SM(d18:1/17:0), Cer(d18:1/17:0), DG(12:0/12:0/0:0), and TG(15:0/15:1/15:0), each at the concentration of 500 ng/mL), was added to each 50 μL of plasma, liver homogenate or QC sample, followed by the addition of 1 mL of MTBE. And then the mixture was vortexed for 10 min. Subsequently, 300 μL of ultrapure water was added to the obtained mixture and vortexed to form a two-phase system. After equilibration, the mixture was centrifuged at 15000 rpm for 10 min at 4 °C. Four hundred microliters of the supernatant were lyophilized and stored at − 80 °C prior to UPLC-Q-orbitrap MS analysis. Lyophilized samples were reconstituted in acetonitrile/isopropanol/water (65:30:5, v/v/v). Chromatographic separation of metabolomics analysis was performed on an Acquity UPLC HSS T3 column (100 × 2.1 mm; 1.8 μm), and an Acquity UPLC BEH HILIC column (100 × 2.1 mm; 1.7 μm) (Waters, Milford, MA, USA) maintained at 40 °C. For the HSS T3 column, the mobile phases A was water with 5 mM ammonium acetate and 0.1% formic acid for positive ion mode, and water with 5 mM ammonium acetate for negative ion mode. The mobile phases B was acetonitrile with 5 mM ammonium acetate and 0.1% formic acid for positive ion mode, and acetonitrile with 5 mM ammonium acetate for negative ion mode. The flow rate was 0.4 mL/min with the following mobile phase gradient: 0–1 min 0.5% (B); 1 − 20 min 0.5% − 95% (B); 20 − 23 min 95% (B); 23 − 26 min 0.5% (B). For BEH HILIC column, the mobile phases A was acetonitrile/water (95:5, v/v) 10 mM ammonium acetate and 0.1% formic acid for positive ion mode, and acetonitrile/water (95:5, v/v) with 10 mM ammonium acetate for negative ion mode. The mobile phase B was acetonitrile/water (1:1, v/v) with 10 mM ammonium acetate and 0.1% formic acid for positive ion mode, and acetonitrile/water (1:1, v/v) with 10 mM ammonium acetate for negative ion mode. The flow rate was 0.4 mL/min with the following mobile phase gradient: 0–1 min 2% (B); 1 − 12 min 2% − 45% (B); 12 − 14 min 45% − 90% (B); 14 − 16.5 min 90% (B); 16.5 − 20 min 2% (B). Chromatographic separation of lipidomics analysis was performed on an Acquity UPLC CSH C18 column (100 × 2.1 mm; 1.7 μm) (Waters, Milford, MA, USA) maintained at 40 °C. The mobile phases A was acetonitrile/water (6:4, v/v) with 10 mM ammonium formate and 0.1% formic acid for positive ion mode, and acetonitrile/water (6:4, v/v) with 10 mM ammonium formate for negative ion mode. The mobile phase B was isopropanol/acetonitrile (9:1, v/v) with 10 mM ammonium formate and 0.1% formic acid for positive ion mode, and isopropanol/acetonitrile (9:1, v/v) with 10 mM ammonium formate for negative ion mode. The flow rate was 0.36 mL/min with the following mobile phase gradient: 0–3 min 40% − 43% (B); 3 − 3.1 min 43% − 50% (B); 3.1 − 13 min 50% − 54% (B); 13 − 13.1 min 54%-70% (B); 13.1 − 20 min 70% − 99% (B); 20 − 20.1 min 99% − 40% (B); 20.1 − 25 min 40% (B). The untargeted metabolomics data were processed with Compound Discoverer software (Thermo Fisher Scientific, San Jose, CA, USA). The untargeted lipidomics data were processed with LipidSearch and Xcalibur software (Thermo Fisher Scientific, San Jose, CA, USA). The quantitation of the lipid IS was performed by Xcalibur (Thermo Fisher Scientific, San Jose, CA, USA). Principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA) was performed by SIMCA 14.1software (Umetrics, Umea, Sweden). The correlation analysis was visualized using the R package (corrplot). Univariate Analysis, cluster Analysis, enrichment analysis and pathway analysis were conducted on the MetaboAnalyst website³⁵.

Targeted metabolomics analysis

The PC/PE detection method is carried out using the following approach. 50 mg liver tissues were homogenated by 250 μL PBS. Liver homogenate was ultrasonic treated for 15 min in ice-cold water. 40 μL of each standard solution and liver homogenate were pipetted with 200 μL of IPA/MEOH (2/1, v/v), samples were homogenized again for three cycles (each cycle: 6800 rpm for 30 s, repeat three times), and then centrifuged for 5 min at 4000 rpm. The samples were analyzed by the UPLC-ESI-MS system. The analyses were performed on a Waters ACQUITY UPLC I-Class system (Waters, Milford, MA, USA) coupled to a Exactive mass spectrometer (Thermo Fisher Scientifi c Inc., Hemel Hempstead, UK) equipped with an electrospray ionization (ESI) source was used for the analysis. Chromatographic separation was performed using an Acquity UPLC CSH C18 (1.7 μm 2.1*100 mm) column with a mobile phase composed of solvent A (0.1% formic acid and 10 mM NH₄OAC in ACN/H₂O = 60/40) and solvent B (0.1% formic acid and 10 mM NH₄OAC in IPA/ACN = 90/10).

The one-carbon metabolites detection method is carried out using the following approach. 50 mg liver tissues were homogenated by 250 μL PBS. Liver homogenate was ultrasonic treated for 15 min in ice-cold water. 50 μL of each standard solution and liver homogenate were pipetted with 150 μL of 0.3 M perchloric acid in water into a 96-well plate, and the plate were shaken for 10 min and then centrifuged for 5 min at 4000 rpm. The analyses were performed on a Waters ACQUITY UPLC H-Class system (Waters, Milford, MA, USA) coupled to a triple quadrupole 6500 mass spectrometer (AB SCIEX, Framingham, MA, USA) equipped with an electrospray ionization (ESI) source was used for the analysis. Chromatographic separation was performed using an Acquity UPLC BEH C18 (1.7 μm 2.1*100 mm) column with a mobile phase composed of solvent A (water containing 0.1% formic acid and 5 mM NH₄OAC) and solvent B (acetonitrile containing 0.1% formic acid).

The bile acids detection method is carried out using the following approach. Liver tissues were firstly homogenized in 5-fold volume of water, and then sonicated for 10 min. A 40 µL of plasma or liver homogenate was mixed with 120 µL methanol and 20 µL of the internal standards (IS) containing 400 nM of d5-GCA, d5-GUDCA, d4-LCA, and d5-TCDCA for each. The mixture was vortexed for 10 min and then centrifuged at 15000 rpm for 10 min at 4 °C. A 40 µL of supernatant and 40 µL water were mixed, vortexed for 10 min, and then centrifuged at 4000 rpm for 10 min prior to UPLC-QQQ MS analysis. The mobile phase A was water containing 1 mM ammonium acetate and 0.1% acetic acid. The mobile phase B was isopropanol/acetonitrile (1/9, v/v) containing 0.1% formic acid. The flow rate was 0.4 mL/min with the following mobile phase gradient: 0–1 min 30% (B); 1 − 4.2 min 30% − 50% (B); 4.2 − 7 min 50% (B); 7 − 8.6 min 95% (B); 8.6 − 10.5 min 30% (B). The MS was operated in negative ion mode with dynamic multiple reaction monitoring. The optimal MS conditions were optimized based on the reference³⁶. The quantitation of the bile acids was performed by MultiQuant™ 2.1 (AB SCIEX, Concord, Canada).

Metabolomics data processing

The quantitation of the bile acids was performed by MultiQuant™ 2.1 (SCIEX, Concord, Canada) with a mass width of ± 0.01 Da and Rt width of ± 0.15 min. The untargeted lipidomic data were processed with LipidSearch software (Thermo Fisher Scientific, San Jose, CA, USA) and lipid identification. The quantitation of the lipid IS was performed by Xcalibur (Thermo Fisher Scientific, San Jose, CA, USA). All lipidomic data were normalized by the corresponding lipid IS. Raw metabolomics data from four acquisition modes (HILIC-NEG, HILIC-POS, T3-NEG, T3-POS) were processed using Compound Discoverer (version 3.3, Thermo Fisher Scientific, Waltham, MA, USA) or LipidSearch (version 4.2, Thermo Fisher Scientific) for peak detection, metabolite annotation, and QC-based normalization. In each mode, raw LC–MS data were imported into the respective software, where peaks were detected using automatic peak picking algorithms, and isotopic peaks were removed. Features were aligned across all samples based on accurate mass (m/z) and retention time (RT) tolerances, and missing values were filled using software-integrated gap-filling algorithms. Metabolite annotation was performed by matching accurate mass, isotope pattern, and, where available, MS/MS fragmentation spectra against built-in compound or lipid databases. For Compound Discoverer, identification was assisted by mzCloud, ChemSpider, and in-house databases. For LipidSearch, lipid identification was based on precursor ion m/z, product ion m/z, and characteristic fragment rules for each lipid class. Annotation confidence followed the software’s scoring system, and only features meeting predefined thresholds (e.g., identification score ≥ 5, mass error ≤ 10 ppm) were retained. For metabolites identified as significantly different in statistical analysis, the corresponding MS/MS spectra were manually reviewed by experienced analysts to confirm structural assignments, and only the most reliable identifications were reported. Principal Component Analysis (PCA) was used for unsupervised visualization of global metabolic variation, while Partial Least Squares Discriminant Analysis (PLS-DA, two components) was applied for supervised group separation. For univariate differential analysis, fold change (FC) was calculated as log2 (mean_group1/mean_group2), with significance assessed by two-sided Student’s t tests (paired for matched data, unpaired otherwise). Volcano plots were generated to display both magnitude (log2FC) and significance (-log10 p-value), where significantly upregulated metabolites (p <  0.05). All statistical analyses and visualizations were performed in Python 3.10 using pandas, numpy, scikit-learn, scipy, matplotlib, and seaborn.

Biochemical measurements

Liver injury in animals was assessed by measuring the levels of ALT and AST on a JCA-BM6010/C Automatic Analyzer in accordance with the manufacturer’s instructions. To determine the intrahepatic triglyceride content, tissue samples were homogenized with 0.5 mL of phosphate-buffered saline (PBS), and total triglycerides were extracted with 1.5 mL of chloroform and methanol (2:1 v/v) overnight. The extracted samples were centrifuged, and the lower liquid phase was collected and air-dried. The triglyceride was then dissolved in 1 mL of ethanol containing 1% Triton X-100, and its levels were quantified using a triglyceride assay kit, followed by normalization to tissue weight. Collagen content was assessed by determining the hydroxyproline concentration using a BioTek ELx800 plate reader, which measures the reaction of oxidized hydroxyproline with 4-(dimethylamine) benzaldehyde and produces a colorimetric (560 nm) product that is proportional to the amount of hydroxyproline present in the sample. ATP contents and MDA activity was measured by the commercially available kit from Nanjing Jiancheng Technology Co., Ltd. (Nanjing, China). Mitochondrial ROS was measured using the MitoSOX™ Red kit (Molecular Probes, Eugene, OR) according to the manufacturer’s guide. Caspase-3 activity was assessed with a Nanjing Jiancheng kit. Liver homogenates were centrifuged, and protein levels quantified by BCA. Samples were reacted with Ac-DEVD-pNA (2 mM) at 37 °C for 2 h, and absorbance read at 405 nm. The M30 CytoDeath™ Apoptosis ELISA Kit was used to detect CK18.

Seahorse assays

Seahorse and glycolysis stress test kit (Agilent Technologies, 103017-100) were used to measure glycolytic flux (ECAR). ECAR was measured according to the manufacturer’s protocol. Briefly, 2 × 10⁵ cells were seeded in XFp miniplates. At 1 h prior to analysis, the medium was replaced with Seahorse XF media (Agilent Technologies, 103681-100), and plates were incubated in a CO²-free incubator at 37 °C. The glycolytic flux assays were performed both basal conditions and after injection of 10 mM glucose, 1 μM oligomycin, and 50 mM 2-DG.

Glucose tolerance test

For glucose tolerance tests, all mice were fasted for 12 h before D-glucose administration. Blood glucose levels were examined from tail vein samples at 15, 30, 60, 90, 120 min post-administration.

Immunofluorescence staining

Mouse livers were embedded in paraffin. Sections (5 μm-thick, mouse) were deparaffinized in serial solutions of xylene and ethanol with decreasing concentration and water, followed by antigen retrieval by steaming in 10 mM citric acid (pH 6.0) for 20 min followed by a 20 min cooling period. The liver sections were washed with PBS buffer and blocked with 5% BSA for 60 min. Next, samples were incubated with primary antibody (F4:80, 1: 200) overnight at 4 °C and then corresponding secondary antibodies(1: 1000) for 60 min. The images were recorded by a fluorescence microscope.

Western blot analysis

Total protein of cells or tissues were denatured in SDS-loading by boiling, then underwent SDS-PAGE. Isolated proteins were transferred to PVDF membranes, which were blocked, then incubated with primary antibody at 4 °C overnight, washed then incubated with corresponding secondary antibody (1;5000) for 1 h at room temperature. After washing, ECL was used to acquire images. The gray density of bands was analyzed by using Image J. Antibodies used in the study were purchased from Proteintech: MAT1A (1:1000, 67408-1-Ig), P-PERK (1:1000, 82534-1-RR), PERK (1:1000, 24390-1-AP), and GAPDH (1:1000, 10494-1-AP). ABclonal: P-eIF2α (1:1000, AP0692), and eIF2α (1:1000, A21221).

Mitochondrial DNA (mtDNA) copy number

Total DNA was extracted from liver tissues using the DNeasy Blood & Tissue Kit (TANGEN) and used for the detection of mtDNA copy number by qPCR with the following reagents: TaqMan Universal PCR master mix and the primers (16S rRNA and 18S rRNA). mtDNA levels were assessed using the mitochondrial genes 16S rRNA; nuclear 18S rRNA served as a loading control.

Quantitative real-time PCR (qRT-PCR)

Total RNA extracted from cells or liver tissues was reverse-transcribed for qRT-PCR analysis with SYBR green I dye in the Mx3000 Quantitative PCR System. The relative expression of target genes was normalized to that of Gapdh and analyzed by the 2-ΔΔCT method. Primers used in the experiments were listed in Supplementary Table 1.

Body composition

Body composition was assessed by using a magnetic resonance whole-body composition analyzer (Minispec LF90 II, Bruker, Karlsruhe, Germany).

Statistical analysis

Data were presented as the mean ± SEM. Comparisons between two or more groups were analyzed using a two-tailed unpaired Student’s t test or two-way ANOVA. Statistical analyses were performed using GraphPad Prism software.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualization: Y.Z., R.H., J. Liu, and J. Li. Funding acquisition: Y.Z., J. Liu, and J. Li. Project administration: B.X.T., Q.W.S., J.Y., X.J.H., N.L.Z, Y.K.S., H.R.G., H.Y.C., H.L.W., and H.M.W. Supervision: Y.Z., R.H., J. Liu, and J. Li. Writing – original draft: B.X.T., J.Y., H.R.G., and Y.Z. Writing – review & editing: Y.Z., J. Liu, R.H., and J. Li. All authors reviewed and approved the final manuscript.

Peer review

Peer review information

Nature Communications thanks Zhaoyue Zhang, Shelly Lu, and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

Lingang Laboratory Program (LGL-1901 and LG-GG-202403-02 to Y.Z., LGL-2612-01 to J. Li), Shanghai Outstanding Academic Leader Program (23XD1450800 to Y.Z.), Shanghai Eastern Talents Excellence Program to Y.Z., Shanghai Super Postdoctoral Fellowship Program to B.X.T., National Natural Science Foundation of China (82373827 to J. Liu, 82521004 and 82130099 to J. Li), the Taishan Scholars Program (tstp0648 to J. Li), Shandong Laboratory Program (SYS202205 to J. Li).

Data availability

The data generated in this study are provided in the Source Data file. The transcriptomics data generated in this study have been deposited in the National Genomics Data Center under accession code CRA040577 (https://ngdc.cncb.ac.cn/gsa/). Source data are provided in this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-75822-y.