Angptl4 integrates dietary and microbial signals to disrupt gut barrier function in MASH Authors: Damien Chua, Zun Siong Low, Joseph Han Sol Kim, Yin Hao Lee, Rattanaporn Kiatbumrung, Pornjira Somnark, Min Xu, Yue Shi, Gourav Kaushal, Marcus Ivan Gerard Vos, Aparna Mahadevan, Natalie Hooi, Mathan Raj, Ekaterina Sviriaeva, Beiming Cui, Shaun Tan, Kazuyuki Kasahara, Chun Loong Ho, Walter Wahli, Kuo Chao Yew, Sunny H. Wong, Christine Cheung, Mintu Pal, Ru Zhang, Natthaya Chuaypen, Pisit Tangkijvanich, Hong Sheng Cheng, Liang Li, Nguan Soon Tan Published: 2026 ABSTRACT Metabolic dysfunction-associated steatotic liver disease (MASLD) is a major contributor to liver morbidity, yet mechanisms linking gut barrier dysfunction to early progression remains poorly defined. We identify intestinal angiopoietin-like 4 (Angptl4) as a central integrator of dietary and microbial signals that governs barrier integrity and hepatic oxidative stress, key early MASLD features. Using intestinal-specific Angptl4 knockout mice, mechanistic in vitro systems, humanized microbiota models, and multi-cohort human studies, we show that intestinal Angptl4 expression is regulated by dietary fatty acids via PPARα signaling and microbiota-derived pattern-recognition pathways, including flagellin-activated-TLR5-EGR1 activation, alongside diet-associated shifts in TLR signaling. These signals destabilize epithelial barriers, amplifying gut-to-liver metabolic and microbial flux. In human cohorts, fecal Angptl4 increases with dysbiosis and metabolic dysfunction, capturing a gut barrier-related dimension distinct from endotoxemia or acute injury. Thus, intestinal Angptl4 emerges as a mechanistic hub linking diet, microbiota, and gut-liver dysfunction, supporting precision barrier-targeted strategies in MASLD. Introduction The intestinal barrier is a crucial defense system that protects the body from harmful external agents. It functions through the coordinated efforts of immune, epithelial, and endothelial cells, working alongside the gut microbiota. While this barrier is essential for maintaining intestinal health, its dysfunction has been linked to various liver diseases, such as metabolic dysfunction-associated steatotic liver disease (MASLD) and cirrhosis, as well as metabolic disorders, autoimmune diseases, and allergies1–3. Despite strong clinical and experimental associations, the initiating mechanisms by which intestinal barrier dysfunction contributes to MASLD pathogenesis remain incompletely defined. The gut-liver axis is a bidirectional communication network that plays a central role in the progression of MASLD4,5. MASLD is a spectrum of liver diseases, ranging from simple fat accumulation (steatosis) to more severe conditions like metabolic dysfunction-associated steatohepatitis (MASH), which can progress to liver fibrosis or cirrhosis. Globally, ~ 25% of people are affected by MASLD, with 10–30% of those cases advancing to MASH6,7. This increase parallels the rise in metabolic syndromes such as obesity and diabetes7,8. The translocation of gut-derived microbial components and metabolites influences both intestinal and hepatic physiology, contributing to systemic inflammation and endotoxemia9. Clinical and experimental studies demonstrate that MASLD risk factors are strongly associated with increased intestinal permeability, allowing microbial elements to access the liver and exacerbate disease inflammation5,10,11. However, whether gut barrier dysfunction represents a downstream consequence of hepatic inflammation or an early, initiating event in MASLD remains unresolved. Emerging evidence indicates that early destabilization of the intestinal epithelial and endothelial barrier may precede overt hepatic inflammation. Increased permeability can enhance intestinal exposure to dietary lipids and microbial-associated molecular patterns, activating pattern-recognition signaling within the gut and promoting hepatic oxidative stress before the onset of immune infiltration or fibrosis. This proposed sequence highlights a critical knowledge gap: the molecular regulators that connect dietary and microbial cues to early gut barrier failure and subsequent liver injury in MASLD are poorly understood. Although inflammatory mediators such as TNFα have been implicated in gut barrier disruption, intestinal permeability and liver injury can occur independently of canonical inflammatory pathways12. Additional factors, including bioactive metabolites such as serotonin and histamine, have been shown to modulate barrier function, yet how diet and microbiota orchestrate this early permeability change in MASLD remains unclear5. Defining the epithelial mechanisms that integrate dietary and microbial signals at early disease stages is therefore essential for resolving causality within the gut-liver axis. Angiopoietin-like 4 (Angptl4) is a secreted glycoprotein with established roles in lipid metabolism and inflammation13. Studies on Angptl4-knockout mice demonstrate that diets enriched in saturated fatty acids provoke severe intestinal inflammation, characterized by peritonitis and fibrosis, whereas trans-fat–rich diets do not elicit similar effects14,15. This excessive inflammation causes gut damage in conditions like colitis and inflammatory bowel disease. In models of colitis and inflammatory bowel disease, Angptl4 deficiency results in sustained leukocyte infiltration, impaired resolution of inflammation, and defective tissue repair, partly attributed to altered immune cell differentiation and prolonged chemokine mRNA stability16,17. These findings establish Angptl4 as an important regulator of intestinal inflammatory responses, yet its role in coordinating gut barrier functions under metabolic stress and its relevance to MASLD remain insufficiently explored. Early studies further implicate intestinal Angptl4 in diet-microbiota interactions relevant to metabolic disease18–20. Germ-free (GF) mice lacking Angptl4 fail to exhibit protection from high-fat diet-induced obesity, unlike wild-type counterparts18,19. Colonization with gut microbiota suppresses intestinal Angptl4 expression, suggesting microbial regulation of host lipid handling through this pathway. The relationship between diet, gut microbiota, and Angptl4 expression is complex, with different microbial species influencing Angptl4 levels in various ways. For instance, colonization with Clostridium tyrobutyricum increases Angptl4 expression in GF mice21. Elevated Angptl4 has also been associated with intestinal barrier dysfunction in models of peanut allergy22. These observations suggest that Angptl4 may occupy a critical intersection among dietary cues, microbial signals, and intestinal barrier regulation, yet its mechanistic contribution to early MASLD pathogenesis has not been defined. Here, we aimed to establish intestinal ANGPTL4 as a mechanistic mediator linking diet- and microbiota-driven gut barrier dysfunction to early MASLD/MASH pathogenesis. Focusing on early disease stages, we tested the hypothesis that intestinal Angptl4 integrates dietary and microbial signals to destabilize epithelial and endothelial barriers, thereby promoting hepatic oxidative stress that precedes inflammation and fibrosis. By resolving and clarifying this early causal axis, our study aims to explain how gut barrier failure initiates MASLD progression and to provide a mechanistic framework for understanding gut-liver communication under metabolic stress. Results Early gut barrier disruption triggers hepatic oxidative damage and inflammation In this study, we investigated early changes in gut-liver communication during MASLD by assessing intestinal permeability in mice fed either a Liver Disease Progression Aggravation Diet (LIDPAD) or a control diet23. Using FITC-dextran to measure gut permeability, we found that after one week, both groups exhibited similar permeability levels. However, permeability progressively increased in LIDPAD-fed mice, becoming statistically significant by the second week and preceding the onset of overt hepatic steatosis, consistent with an early barrier defect (Fig. 1A). All analyses were performed in age- and sex-matched Angptl4fl/fl mice fed diets in parallel, ensuring comparability across time points.Fig. 1Gut permeability is an early event during MASLD progression.A Plasma concentration of FITC-dextran in Angptl4fl/fl mice fed with Control and LIDPAD diets for 2 weeks after FITC-dextran gavage. P-value = 0.142 (week 1), 0.330 (control, week 1 vs week 2), 0.00223 (week 2). B Relative mRNA expression levels of tight junction biomarkers, ZO-1, OCLN, JAM-A, across different regions of the small intestine of Angptl4fl/fl mice after 2 weeks of control and LIDPAD feeding. 18S rRNA levels are used as endogenous reference. P-values (duodenum) = 0.029 (ZO-1), 0.048 (OCLN), 0.0037 (JAM-A); P-values (jejunum) = 0.048 (ZO-1), 0.0186 (OCLN), 0.023 (JAM-A); P-values (ileum) = 0.037. (ZO-1), 0.0186 (OCLN), 0.0001 (JAM-A). Total area of VeCad/total area of CD31 was quantified, representing intestinal vascular integrity. P-values = 0.0001 (duodenum), 0.0003 (jejunum), 0.000099 (ileum). C Hepatic portal blood LPS levels measured in Angptl4fl/fl mice after 2 weeks of feeding with control or LIDPAD diets. P-value = 0.000099. D Quantification of 4HNE positive area (%), a marker of oxidative stress, in livers of Angptl4fl/fl mice after 2 weeks of control diet and LIDPAD feeding. P-value = 0.0037. E Representative histogram of hepatocyte count against CellROX, a dye for ROS, comparing LIDPAD- and control-fed mice through flow cytometry analysis. F Flow cytometric analysis of total immune cell populations from liver and intestine of control- and LIDPAD-fed Angptl4fl/fl mice. The graph represents the proportion of CD45 + immune cells. The graph represents the percentage of CD45 + immune cells in the respective organs. P-values = 0.063 (liver), 0.18 (intestine). G Volcano plot showing the differential gene expression (DEGs) in the intestinal mucosa from Angptl4fl/fl mice after 2 weeks of control diet and LIDPAD feeding. DEGs with significant fold changes (adjusted p-value < 0.05) are highlighted in red. H Heatmap showing DEGs of intestinal mucosa from control- and LIDPAD-fed Angptl4fl/fl mice. Corresponding Gene Ontology (GO) biological processes are annotated. Color intensity reflects z-scores of expression levels. I Gene Set Enrichment Analysis (GSEA) results showing enriched pathways altered by LIDPAD. Positive enrichment scores for pathways like “negative regulation of cell-cell adhesion” highlight their activation, while negative scores for “fatty acid homeostasis” and “triglyceride lipase activity” pathways suggest their downregulation. J Network analysis and Circo plot reveal interconnected genes involved in fatty acid homeostasis, cell-cell adhesion regulation, extracellular matrix, inflammatory and stress response and cellular responses to bacteria. Key hub genes such as Angptl4 are highlighted in red in the Circo plot, indicating their pivotal role in mediating these processes. Two innermost tracks of the Circo plot demonstrate the centrality scores, Density of Maximum Neighborhood Component (DMNC) and bottleneck scores, while the outer tracks indicate the functions of each gene. For (A) n = 3 mice per group; (B–F) n = 5 mice per group. For (A, B, D), data expressed as mean ± S.D. For (C), boxplots show the median (center line), the 25th and 75th percentiles (bounds of the box), and whiskers extending to maxima and minima. ***p < 0.001, **p < 0.01, *p < 0.05 (two-sided unpaired t test). n.s. denotes not significant. Gut permeability is an early event during MASLD progression. A Plasma concentration of FITC-dextran in Angptl4fl/fl mice fed with Control and LIDPAD diets for 2 weeks after FITC-dextran gavage. P-value = 0.142 (week 1), 0.330 (control, week 1 vs week 2), 0.00223 (week 2). B Relative mRNA expression levels of tight junction biomarkers, ZO-1, OCLN, JAM-A, across different regions of the small intestine of Angptl4fl/fl mice after 2 weeks of control and LIDPAD feeding. 18S rRNA levels are used as endogenous reference. P-values (duodenum) = 0.029 (ZO-1), 0.048 (OCLN), 0.0037 (JAM-A); P-values (jejunum) = 0.048 (ZO-1), 0.0186 (OCLN), 0.023 (JAM-A); P-values (ileum) = 0.037. (ZO-1), 0.0186 (OCLN), 0.0001 (JAM-A). Total area of VeCad/total area of CD31 was quantified, representing intestinal vascular integrity. P-values = 0.0001 (duodenum), 0.0003 (jejunum), 0.000099 (ileum). C Hepatic portal blood LPS levels measured in Angptl4fl/fl mice after 2 weeks of feeding with control or LIDPAD diets. P-value = 0.000099. D Quantification of 4HNE positive area (%), a marker of oxidative stress, in livers of Angptl4fl/fl mice after 2 weeks of control diet and LIDPAD feeding. P-value = 0.0037. E Representative histogram of hepatocyte count against CellROX, a dye for ROS, comparing LIDPAD- and control-fed mice through flow cytometry analysis. F Flow cytometric analysis of total immune cell populations from liver and intestine of control- and LIDPAD-fed Angptl4fl/fl mice. The graph represents the proportion of CD45 + immune cells. The graph represents the percentage of CD45 + immune cells in the respective organs. P-values = 0.063 (liver), 0.18 (intestine). G Volcano plot showing the differential gene expression (DEGs) in the intestinal mucosa from Angptl4fl/fl mice after 2 weeks of control diet and LIDPAD feeding. DEGs with significant fold changes (adjusted p-value < 0.05) are highlighted in red. H Heatmap showing DEGs of intestinal mucosa from control- and LIDPAD-fed Angptl4fl/fl mice. Corresponding Gene Ontology (GO) biological processes are annotated. Color intensity reflects z-scores of expression levels. I Gene Set Enrichment Analysis (GSEA) results showing enriched pathways altered by LIDPAD. Positive enrichment scores for pathways like “negative regulation of cell-cell adhesion” highlight their activation, while negative scores for “fatty acid homeostasis” and “triglyceride lipase activity” pathways suggest their downregulation. J Network analysis and Circo plot reveal interconnected genes involved in fatty acid homeostasis, cell-cell adhesion regulation, extracellular matrix, inflammatory and stress response and cellular responses to bacteria. Key hub genes such as Angptl4 are highlighted in red in the Circo plot, indicating their pivotal role in mediating these processes. Two innermost tracks of the Circo plot demonstrate the centrality scores, Density of Maximum Neighborhood Component (DMNC) and bottleneck scores, while the outer tracks indicate the functions of each gene. For (A) n = 3 mice per group; (B–F) n = 5 mice per group. For (A, B, D), data expressed as mean ± S.D. For (C), boxplots show the median (center line), the 25th and 75th percentiles (bounds of the box), and whiskers extending to maxima and minima. ***p < 0.001, **p < 0.01, *p < 0.05 (two-sided unpaired t test). n.s. denotes not significant. Histological analysis revealed lipid droplets in the intestinal microvilli of LIDPAD-fed mice, which were absent in the control-fed group (Supplementary Fig. S1A). This increase in epithelial permeability was accompanied by a significant reduction in mRNA levels of key tight junction proteins (ZO-1, OCLN, JAM-A) in the intestinal mucosa, indicating a compromised gut barrier (Fig. 1B). Intestinal vascular integrity was evaluated using complementary structural approaches. Reduced VE-Cadherin coverage relative to CD31-positive endothelium revealed destabilization of endothelial junctions within the intestinal and colonic microvasculature (Fig. 1B and Supplementary Figs. S1A, S2B), consistent with increased vascular permeability. In the liver, mild steatosis was observed in LIDPAD-fed mice after two weeks, while control-fed mice showed minimal lipid accumulation (Supplementary Fig. S1C). Increased intestinal permeability in LIDPAD-fed mice was associated with elevated levels of lipopolysaccharide (LPS) in the hepatic portal blood, indicating translocation of gut-derived microbial products to the liver (Fig. 1C). This was accompanied by a pronounced increase in hepatic oxidative stress, as evidenced by the accumulation of lipid peroxidation adduct 4-hydroxy-2-nonenal (4-HNE)24 (Fig. 1D and Supplementary Fig. S1D). Flow cytometric analysis using CellROXTM further confirmed increased reactive oxygen species (ROS) production in hepatocytes, highlighting the role of oxidative stress in liver inflammation, a hallmark of MASLD25 (Fig. 1E). Notably, these oxidative stress signatures appeared in the absence of significant immune cell infiltration into the liver of LIDPAD-fed mice, whereas a modest, non-significant increase in immune cells was detected in the intestinal compartment (Fig. 1F and Supplementary Fig. S1E, F). These data support a temporal sequence in which gut barrier disruption and hepatic oxidative stress precede overt hepatic inflammation. To further delineate early gut-liver crosstalk, we performed RNA sequencing of intestinal mucosa from LIDPAD- and control-fed mice. A volcano plot revealed an even distribution of differentially expressed genes (DEGs), with both upregulated and downregulated genes between the two dietary groups (Fig. 1G). In LIDPAD-fed mice, genes associated with cell adhesion, bacterial sensing, and Toll-like receptor (TLR) signaling were upregulated, contributing to barrier dysfunction. Conversely, lipid metabolism and antibacterial defense genes were downregulated, potentially contributing to cell membrane destabilization and gut barrier deterioration (Fig. 1H). Gene-set enrichment analysis (GSEA) revealed a significant loss of cell-cell adhesion and downregulation of fatty acid homeostasis in LIDPAD-fed mice (Fig. 1I). Network analysis demonstrated that these processes were interconnected through stress-responsive and bacteria-responsive gene modules (Fig. 1J). Within this network, Angptl4 emerged as a central hub based on Density of Maximum Neighborhood Component (DMNC) and bottleneck centrality scores, nominating it as a candidate integrator of diet and microbiota-driven barrier dysfunction. Intestinal-specific Angptl4 deficiency reduces diet-induced gut permeability and MASH Consistent with its role as a central regulatory node, LIDPAD feeding robustly induced Angptl4 expression in the intestinal mucosa of Angptl4fl/fl mice (Fig. 2A). To directly test the role of intestinal Angptl4 in gut-liver communication, we generated mice with an intestinal epithelial-specific deletion of Angptl4 (Angptl4Vil1-/-) and subjected them, alongside littermate Angptl4fl/fl controls, to LIDPAD or control diet for two or eight weeks. After two weeks of LIDPAD feeding, Angptl4Vil1-/- mice showed minimal changes in body weight, liver weight (Supplementary Fig. S2A–C), and adiposity (Supplementary Fig. S2D–F) compared to control-fed mice. Histopathological analysis revealed minimal hepatic steatosis in Angptl4Vil1-/- mice (Supplementary Fig. S2D), in contrast to LIDPAD-fed Angptl4fl/fl mice, which developed extensive microvesicular steatosis and adipocyte hypertrophy (Supplementary Figs. S1C, S2E,F). This protective phenotype persisted at eight weeks, with Angptl4Vil1-/- mice remaining resistant to diet-induced hepatic steatosis and maintaining epididymal white adipose tissue comparable to control-fed Angptl4Vil1-/- mice (Supplementary Fig. S2D, S2G).Fig. 2Intestinal-specific Angptl4 knockout protects gut permeability and mitigates MASLD progression.A Relative expression level of cAngptl4 mRNA and protein in the intestinal mucosa of Angptl4fl/fl mice on control or LIDPAD diets for 2 weeks. Representative immunoblots of cAngptl4 and β-actin are shown. Loading controls were from the same samples. n = 3 independent experiments. B Plasma concentration of FITC-dextran, a marker of intestinal permeability, in Angptl4fl/fl and Angptl4Vil1-/- gavaged with FITC-dextran after 2 weeks on control or LIDPAD diets n = 8 mice per group. P-value = 0.0022 (week 2, LIDPAD, Angptl4fl/fl vs Angptl4Vil1-/-). C Quantification of 4HNE positive area (%), a marker of oxidative stress, in livers of Angptl4fl/fl and Angptl4Vil1-/- after 2 weeks of control diet and LIDPAD feeding. n = 5 mice per group. P-value = 0.0875. D Relative mRNA levels of tight junction biomarkers, ZO-1, OCLN, JAM-A, across different regions of the small intestine of Angptl4fl/fl and Angptl4Vil1-/- mice fed control or LIDPAD diets for 2 weeks. The ratio of VE-Cadherin to CD31 was used to assess vascular integrity. n = 5 mice per group. P-values, Angptl4Vil1-/-+LIDPAD vs Angptl4Vil1-/-+ Control (duodenum) = 0.029 (ZO-1), 0.048 (OCLN), 0.0037 (JAM-A); P-values (jejunum) = 0.048 (ZO-1), 0.0186 (OCLN), 0.023 (JAM-A); P-values (ileum) = 0.037. (ZO-1), 0.0186 (OCLN), 0.0001 (JAM-A). Total area of VeCad/total area of CD31 was quantified, representing intestinal vascular integrity. P-values = 0.0001 (duodenum), 0.0003 (jejunum), 0.000099 (ileum). E Relative protein expression and immunoblotting of tight junction proteins ZO-1 and OCLN (normalized to β-tubulin) in intestinal epithelia of Angptl4fl/fl and Angptl4Vil1-/- after 2 weeks of control and LIDPAD diet feeding. Representative Western blots showing ZO-1 (210 kDa), OCLN (59 kDa), and β-tubulin (55 kDa) expression. n = 3 mice per group. For ZO-1, p-values = 0.0206 (Angptl4fl/fl Ctrl VS Angptl4fl/fl LIDPAD), 0.037 (Angptl4Vil1-/- Ctrl vs Angptl4fl/fl LIDPAD), 0.0069 (Angptl4Vil1-/- LIDPAD vs Angptl4fl/fl LIDPAD). For OCLN, p-values = 0.00099 (Angptl4fl/fl Ctrl VS Angptl4fl/fl LIDPAD), 0.0023 (Angptl4Vil1-/- Ctrl vs Angptl4fl/fl LIDPAD), 0.0044 (Angptl4Vil1-/- LIDPAD vs Angptl4fl/fl LIDPAD). F Volcano plot showing differentially expressed genes (DEGs) in the intestinal mucosa from Angptl4Vil-/- mice on LIDPAD and control diets for 2 weeks. G Gene Ontology (GO) biological process terms associated with the 151 upregulated genes in the intestinal mucosa of Angptl4Vil1-/- mice on the LIDPAD diet, depicted as a bubble plot where the size represents gene count and color intensity represents the false discovery rate (FDR). For (A–D). Gray outlines indicate replicated data from Fig. 1. Data are expressed as the mean ± SD. *p < 0.05; ***p < 0.001, n.s. denotes not significant. (Kruskal‒Wallis test with Dunn test for multiple comparisons). Intestinal-specific Angptl4 knockout protects gut permeability and mitigates MASLD progression. A Relative expression level of cAngptl4 mRNA and protein in the intestinal mucosa of Angptl4fl/fl mice on control or LIDPAD diets for 2 weeks. Representative immunoblots of cAngptl4 and β-actin are shown. Loading controls were from the same samples. n = 3 independent experiments. B Plasma concentration of FITC-dextran, a marker of intestinal permeability, in Angptl4fl/fl and Angptl4Vil1-/- gavaged with FITC-dextran after 2 weeks on control or LIDPAD diets n = 8 mice per group. P-value = 0.0022 (week 2, LIDPAD, Angptl4fl/fl vs Angptl4Vil1-/-). C Quantification of 4HNE positive area (%), a marker of oxidative stress, in livers of Angptl4fl/fl and Angptl4Vil1-/- after 2 weeks of control diet and LIDPAD feeding. n = 5 mice per group. P-value = 0.0875. D Relative mRNA levels of tight junction biomarkers, ZO-1, OCLN, JAM-A, across different regions of the small intestine of Angptl4fl/fl and Angptl4Vil1-/- mice fed control or LIDPAD diets for 2 weeks. The ratio of VE-Cadherin to CD31 was used to assess vascular integrity. n = 5 mice per group. P-values, Angptl4Vil1-/-+LIDPAD vs Angptl4Vil1-/-+ Control (duodenum) = 0.029 (ZO-1), 0.048 (OCLN), 0.0037 (JAM-A); P-values (jejunum) = 0.048 (ZO-1), 0.0186 (OCLN), 0.023 (JAM-A); P-values (ileum) = 0.037. (ZO-1), 0.0186 (OCLN), 0.0001 (JAM-A). Total area of VeCad/total area of CD31 was quantified, representing intestinal vascular integrity. P-values = 0.0001 (duodenum), 0.0003 (jejunum), 0.000099 (ileum). E Relative protein expression and immunoblotting of tight junction proteins ZO-1 and OCLN (normalized to β-tubulin) in intestinal epithelia of Angptl4fl/fl and Angptl4Vil1-/- after 2 weeks of control and LIDPAD diet feeding. Representative Western blots showing ZO-1 (210 kDa), OCLN (59 kDa), and β-tubulin (55 kDa) expression. n = 3 mice per group. For ZO-1, p-values = 0.0206 (Angptl4fl/fl Ctrl VS Angptl4fl/fl LIDPAD), 0.037 (Angptl4Vil1-/- Ctrl vs Angptl4fl/fl LIDPAD), 0.0069 (Angptl4Vil1-/- LIDPAD vs Angptl4fl/fl LIDPAD). For OCLN, p-values = 0.00099 (Angptl4fl/fl Ctrl VS Angptl4fl/fl LIDPAD), 0.0023 (Angptl4Vil1-/- Ctrl vs Angptl4fl/fl LIDPAD), 0.0044 (Angptl4Vil1-/- LIDPAD vs Angptl4fl/fl LIDPAD). F Volcano plot showing differentially expressed genes (DEGs) in the intestinal mucosa from Angptl4Vil-/- mice on LIDPAD and control diets for 2 weeks. G Gene Ontology (GO) biological process terms associated with the 151 upregulated genes in the intestinal mucosa of Angptl4Vil1-/- mice on the LIDPAD diet, depicted as a bubble plot where the size represents gene count and color intensity represents the false discovery rate (FDR). For (A–D). Gray outlines indicate replicated data from Fig. 1. Data are expressed as the mean ± SD. *p < 0.05; ***p < 0.001, n.s. denotes not significant. (Kruskal‒Wallis test with Dunn test for multiple comparisons). Hepatic transcriptomic profiling revealed that Angptl4 deficiency attenuated LIDPAD-induced activation of hepatic inflammatory and fibrogenic processes, including integrin signaling, lipid oxidation, IL1α and IL8 production and collagen metabolism. Conversely, pathways associated with negative regulation of oxidative stress and maintenance of cell adhesion were preserved in Angptl4Vil1-/- mice under LIDPAD feeding (Supplementary Fig. S2H). Consistent with these hepatic effects, Angptl4Vil1-/- mice were protected from diet-induced gut and vascular permeability defects. FITC-dextran assays and immunofluorescence analyses demonstrated preserved epithelial and endothelial barrier integrity in Angptl4Vil1-/- mice relative to Angptl4fl/fl controls (Fig. 2B, Supplementary Fig. S2I). This protection was linked to significantly decreased hepatic oxidative damage, with 4-HNE and ROS levels similar to those seen in control-fed mice (Fig. 2C and Supplementary Fig. S2J). Molecular analyses of intestinal tissue showed stable or enhanced expression of tight junction and endothelial markers in Angptl4Vil1-/- mice, indicating that intestinal Angptl4 deficiency effectively preserves gut barrier integrity. In the duodenum, occludin levels increased, and VeCad/CD31 ratio decreased, predominantly reflecting changes in the submucosal vasculature (Fig. 2D). In the jejunum and ileum, expressions of ZO-1, occludin, JAM-A and VeCad, remained stable despite LIDPAD feeding, indicating that intestinal Angptl4 deficiency prevented diet-induced increases in gut permeability (Fig. 2D). Immunoblot analysis confirmed that LIDPAD feeding significantly reduced ZO-1 and OCLN protein levels in Angptl4fl/fl mice, whereas intestinal-specific deletion of Angptl4 (Angptl4Vil1-/-) prevented this loss, maintaining protein abundance comparable to control-fed mice (Fig. 2E). These concordant mRNA and protein data establish intestinal Angptl4 as a necessary determinant of diet-induced tight junction disruption. Finally, transcriptomic analysis of intestinal mucosa from Angptl4Vil1-/- mice revealed a unique gene expression profile, with 96% of DEGs upregulated and minimal overlap with DEGs in Angptl4fl/fl mice (Fig. 2F). Gene Ontology analysis revealed upregulation of genes associated with cell-cell adhesion and morphogenesis in Angptl4Vil1-/- mice (Fig. 2G), underscoring a role for Angplt4 in reducing gut permeability under dietary stress. Notably, genes involved in intestinal sulfation, a key process for colonic protection26, were significantly upregulated Angptl4Vil1-/- mice27. Additionally, the dysregulated hydrolase activity observed in these mice has implications for barrier function (Fig. 2G). Together, these findings identify intestinal Angptl4 as a critical molecular checkpoint linking dietary stress to gut barrier dysfunction and downstream liver pathology. Modulation of intestinal Angptl4 by dietary fats and MAMPs To directly assess whether Angptl4 modulates intestinal barrier integrity, normal human intestinal epithelial cells (hIECs, NCM460 cell line) were treated with recombinant cAngptl4 protein (rcAngptl4), corresponding to the C-terminal fibrinogen-like domain of Angptl4. rcAngptl4 caused a dose-dependent decrease in trans-epithelial electrical resistance (TEER), indicating weakened barrier integrity (Fig. 3A). This effect was neutralized by Angptl4-specific blocking antibodies (mAb11F6C4), while an IgG isotype control had no effect, confirming the specificity of Angptl4-mediated barrier disruption (Fig. 3A).Fig. 3Molecular crosstalk between dietary fats, microbial signals, and intestinal Angptl4 expression in MASLD.A Transepithelial electrical resistance (TEER) measurement of human intestinal epithelial cells (hIECs) treated with recombinant cAngptl4 (rcAngptl4) at varying concentrations and in combination with an anti-cAngptl4 monoclonal antibody (mAb), mAb11F6C4. The graph shows the impact of cAngptl4 on cell barrier function over time. n = 3 independent experiments. B Relative expression of Angptl4 mRNA (top panel) and protein (bottom panel) in hIECs treated with different fatty acids or vehicle control. Representative immunoblots of cAngptl4 (35 kDa) and β-actin (42 kDa) are shown. n = 3 independent experiments. C Chromatin immunoprecipitation (ChIP) assays to assess PPARα binding to the PPRE element in hIECs treated with C18:2 fatty acid or vehicle. Data are normalized to IgG control and presented as fold change. D Relative Angptl4 mRNA expression in hIECs treated with C18:2 and 10 μM fenofibrate as PPARα agonizts. These are co-treated with either si-scrambled or siPPARα cells. n = 4 independent experiments. E Relative mRNA levels of TLR2, 4, 5, 6 across different regions of the small intestine of mice fed with LIDPAD or control diet for two weeks. n = 5 mice per group. For duodenum, p-values = 0.93 (TLR2), 0.047 (TLR4), 0.0384 (TLR5), 0.33 (TLR6); for jejunum, p-values = 0.0154 (TLR2), 0.0019 (TLR4), 0.0015 (TLR5), 0.477 (TLR6); for ileum, p-values = 0.0430 (TLR2), 0.044 (TLR4), 0.0005 (TLR5), 0.476(TLR6); F, G Relative expression of Angptl4 in ex vivo intestinal explants (F) and in vitro hIECs (G) treated with various TLR agonizts and vehicle control. For (F, G), n = 5 independent experiments. For (F) p-values (compared with vehicle group) = 0.0038 (LPS), 0.0007 (Pam3CSK4), 0.069 (HKLM), 0.8046 (FSL-1), 0.026 (FLA). For G, p-values (compared with vehicle group) = 0.476 (LPS), 0.0159 (Pam3CSK4), 0.78576 (HKLM), 0.2738 (FSL-1), 0.000099 (FLA). H Relative Angptl4 mRNA expression in cells transfected with control siRNA (si-scrambled, white bars) or TLR4-targeting siRNA (siTLR4, black bars) and stimulated with LPS (+). n = 3 independent experiments. I Relative expression of Angptl4 in in vitro human intestinal epithelial cells treated with Salmonella typhi flagellin (FLA) and various inhibitors against downstream kinases of TLR5 pathways. n = 4 independent experiments. p-values = 0.476 (FLA vs vehicle group), 0.0011 (FLA + GSKi vs FLA), 0.000099 (FLA + JNKi vs FLA). J ChIP assays to assess EGR1 binding to its cognate binding sites in human intestinal epithelial cells treated with flagellin or vehicle. Data are normalized to IgG control and presented as fold enrichment. n = 3 independent experiments. K Relative Angptl4 mRNA expression (normalized to 18S) in hIECs transfected with scrambled siRNA (si-scrambled, white bars), siEGR1 (black bars), and stimulated with Salmonella typhi flagellin (FLA). n = 3 independent experiments. L, M Box plot of the ratio of Gram-negative to Gram-positive bacteria (L) and stimulator-to-silent flagellin (M) on control and LIDPAD diet for 2 weeks. All box plots need to be defined in the legends in terms of minima, maxima, center, bounds of box and whiskers and percentile. n = 3 biological replicates. (N) Relative expression of Angptl4 in patient-derived intestinal organoids incubated with recombinant flagellins from Roseburia hominis, Romboutsia hominis and Salmonella typhi. n = 3 independent experiments. p-values (compared with vehicle control) = 0.459 (Roseburia hominis), 0.0346 (Romboutsia hominis), 0.0379 (Salmonella typhi). O TEER measurements of patient-derived intestinal organoids incubated with recombinant flagellins from Salmonella typhi, Roseburia hominis, Romboutsia hominis, Salmonella typhi with anti-cAngptl4 mAb, Romboutsia hominis with anti-cAngptl4 mAb (mAb11F6C4). The graph shows the impact of stimulator and silent flagellins as well as stimulator flagellins with cAngptl4 mAb (mAb11F6C4) treatment on cell barrier function over time. n = 3 independent experiments. P FITC-Dextran levels (Δ ng/μL) represent the change in gut permeability before and after bacterial transplantation in germ-free mice colonized with Roseburia hominis or Romboutsia hominis. Mice colonized with Romboutsia exhibited a significantly greater increase in FITC-Dextran levels, indicating heightened gut permeability compared to Roseburia-colonized mice. Data are presented as box-and-whisker plots, with individual data points shown, n = 8 independent biological replicates. Statistical significance was determined using an unpaired two-sided t test (p < 0.001). Q Quantification of 4HNE positive area (%) in livers of mice colonized with Roseburia hominis and Romboutsia hominis after 2 weeks of control diet feeding. n = 3 biological replicates, p-value = 0.027. R Schematic illustrates how dietary fats and MAMPs influence gut permeability and hepatic oxidative stress in MASLD. Dietary fats stimulate PPARα, which directly enhances Angptl4 expression. Meanwhile, gut dysbiosis alters microbial composition in two key ways: (1) a decrease in the Gram-negative:Gram-positive ratio, which suggests a reduction in overall available LPS levels, and (2) an increase in the stimulator-to-silent flagellin ratio, indicating a shift toward more stimulator flagellins. As a result, dysbiotic bacteria have higher stimulator flagellin, which activates TLR5, leading to downstream signaling through GSK3 and JNK, ultimately inducing EGR1, which transcriptionally upregulates Angptl4 expression. In parallel, the reduced LPS exposure from Gram-negative bacteria leads to a diminished TLR4-mediated repression of Angptl4. The net effect is an increase in gut permeability and hepatic oxidative stress. The gray line denotes diminished regulatory effects. Figures referenced in parentheses correspond to experimental data supporting each molecular interaction. Created in BioRender. Li, L. (2026) https://BioRender.com/jo5i54i. For (A–K, N, O, Q), data are expressed as mean ± SD. For (L, M, P), boxplots show the median (center line), the 25th and 75th percentiles (bounds of the box), and whiskers extending to maxima and minima. ***p < 0.001, **p < 0.01, *p < 0.05 (two-sided unpaired t test). n.s. denotes not significant. Molecular crosstalk between dietary fats, microbial signals, and intestinal Angptl4 expression in MASLD. A Transepithelial electrical resistance (TEER) measurement of human intestinal epithelial cells (hIECs) treated with recombinant cAngptl4 (rcAngptl4) at varying concentrations and in combination with an anti-cAngptl4 monoclonal antibody (mAb), mAb11F6C4. The graph shows the impact of cAngptl4 on cell barrier function over time. n = 3 independent experiments. B Relative expression of Angptl4 mRNA (top panel) and protein (bottom panel) in hIECs treated with different fatty acids or vehicle control. Representative immunoblots of cAngptl4 (35 kDa) and β-actin (42 kDa) are shown. n = 3 independent experiments. C Chromatin immunoprecipitation (ChIP) assays to assess PPARα binding to the PPRE element in hIECs treated with C18:2 fatty acid or vehicle. Data are normalized to IgG control and presented as fold change. D Relative Angptl4 mRNA expression in hIECs treated with C18:2 and 10 μM fenofibrate as PPARα agonizts. These are co-treated with either si-scrambled or siPPARα cells. n = 4 independent experiments. E Relative mRNA levels of TLR2, 4, 5, 6 across different regions of the small intestine of mice fed with LIDPAD or control diet for two weeks. n = 5 mice per group. For duodenum, p-values = 0.93 (TLR2), 0.047 (TLR4), 0.0384 (TLR5), 0.33 (TLR6); for jejunum, p-values = 0.0154 (TLR2), 0.0019 (TLR4), 0.0015 (TLR5), 0.477 (TLR6); for ileum, p-values = 0.0430 (TLR2), 0.044 (TLR4), 0.0005 (TLR5), 0.476(TLR6); F, G Relative expression of Angptl4 in ex vivo intestinal explants (F) and in vitro hIECs (G) treated with various TLR agonizts and vehicle control. For (F, G), n = 5 independent experiments. For (F) p-values (compared with vehicle group) = 0.0038 (LPS), 0.0007 (Pam3CSK4), 0.069 (HKLM), 0.8046 (FSL-1), 0.026 (FLA). For G, p-values (compared with vehicle group) = 0.476 (LPS), 0.0159 (Pam3CSK4), 0.78576 (HKLM), 0.2738 (FSL-1), 0.000099 (FLA). H Relative Angptl4 mRNA expression in cells transfected with control siRNA (si-scrambled, white bars) or TLR4-targeting siRNA (siTLR4, black bars) and stimulated with LPS (+). n = 3 independent experiments. I Relative expression of Angptl4 in in vitro human intestinal epithelial cells treated with Salmonella typhi flagellin (FLA) and various inhibitors against downstream kinases of TLR5 pathways. n = 4 independent experiments. p-values = 0.476 (FLA vs vehicle group), 0.0011 (FLA + GSKi vs FLA), 0.000099 (FLA + JNKi vs FLA). J ChIP assays to assess EGR1 binding to its cognate binding sites in human intestinal epithelial cells treated with flagellin or vehicle. Data are normalized to IgG control and presented as fold enrichment. n = 3 independent experiments. K Relative Angptl4 mRNA expression (normalized to 18S) in hIECs transfected with scrambled siRNA (si-scrambled, white bars), siEGR1 (black bars), and stimulated with Salmonella typhi flagellin (FLA). n = 3 independent experiments. L, M Box plot of the ratio of Gram-negative to Gram-positive bacteria (L) and stimulator-to-silent flagellin (M) on control and LIDPAD diet for 2 weeks. All box plots need to be defined in the legends in terms of minima, maxima, center, bounds of box and whiskers and percentile. n = 3 biological replicates. (N) Relative expression of Angptl4 in patient-derived intestinal organoids incubated with recombinant flagellins from Roseburia hominis, Romboutsia hominis and Salmonella typhi. n = 3 independent experiments. p-values (compared with vehicle control) = 0.459 (Roseburia hominis), 0.0346 (Romboutsia hominis), 0.0379 (Salmonella typhi). O TEER measurements of patient-derived intestinal organoids incubated with recombinant flagellins from Salmonella typhi, Roseburia hominis, Romboutsia hominis, Salmonella typhi with anti-cAngptl4 mAb, Romboutsia hominis with anti-cAngptl4 mAb (mAb11F6C4). The graph shows the impact of stimulator and silent flagellins as well as stimulator flagellins with cAngptl4 mAb (mAb11F6C4) treatment on cell barrier function over time. n = 3 independent experiments. P FITC-Dextran levels (Δ ng/μL) represent the change in gut permeability before and after bacterial transplantation in germ-free mice colonized with Roseburia hominis or Romboutsia hominis. Mice colonized with Romboutsia exhibited a significantly greater increase in FITC-Dextran levels, indicating heightened gut permeability compared to Roseburia-colonized mice. Data are presented as box-and-whisker plots, with individual data points shown, n = 8 independent biological replicates. Statistical significance was determined using an unpaired two-sided t test (p < 0.001). Q Quantification of 4HNE positive area (%) in livers of mice colonized with Roseburia hominis and Romboutsia hominis after 2 weeks of control diet feeding. n = 3 biological replicates, p-value = 0.027. R Schematic illustrates how dietary fats and MAMPs influence gut permeability and hepatic oxidative stress in MASLD. Dietary fats stimulate PPARα, which directly enhances Angptl4 expression. Meanwhile, gut dysbiosis alters microbial composition in two key ways: (1) a decrease in the Gram-negative:Gram-positive ratio, which suggests a reduction in overall available LPS levels, and (2) an increase in the stimulator-to-silent flagellin ratio, indicating a shift toward more stimulator flagellins. As a result, dysbiotic bacteria have higher stimulator flagellin, which activates TLR5, leading to downstream signaling through GSK3 and JNK, ultimately inducing EGR1, which transcriptionally upregulates Angptl4 expression. In parallel, the reduced LPS exposure from Gram-negative bacteria leads to a diminished TLR4-mediated repression of Angptl4. The net effect is an increase in gut permeability and hepatic oxidative stress. The gray line denotes diminished regulatory effects. Figures referenced in parentheses correspond to experimental data supporting each molecular interaction. Created in BioRender. Li, L. (2026) https://BioRender.com/jo5i54i. For (A–K, N, O, Q), data are expressed as mean ± SD. For (L, M, P), boxplots show the median (center line), the 25th and 75th percentiles (bounds of the box), and whiskers extending to maxima and minima. ***p < 0.001, **p < 0.01, *p < 0.05 (two-sided unpaired t test). n.s. denotes not significant. We next examined how dietary fatty acids associated with MASLD regulate intestinal Angptl4 expression. Fatty acids such as C16:0, C18:1, and C18:2 significantly increased Angptl4 mRNA and protein levels in NCM460 cells (Fig. 3B). Chromatin immunoprecipitation (ChIP) analysis revealed that these dietary lipids activated Angptl4 expression through the PPARα pathway (Fig. 3C). To functionally validate PPARα dependency, siRNA-mediated knockdown of PPARα markedly attenuated Angptl4 induction by both C18:2 and PPARα agonist fenofibrate (10 μM) (Fig. 3D and Supplementary Fig. S3A), establishing that dietary fatty acids directly induce Angptl4 expression through a PPARα-dependent transcriptional mechanism. Further investigation into the impact of microbial-associated molecular patterns (MAMPs) on Angptl4 revealed changes in TLR signaling pathways in LIDPAD-fed mice12,28. We focused on TLRs 2, 4, 5, and 6, known for their roles in barrier function and diet-induced metabolic disease29–31. In LIDPAD-fed mice, we observed a reduction in TLR4 expression, alongside an upregulation of TLR2 and TLR5 throughout the intestine, suggesting that diet-induced alterations in TLR expression may influence the regulation of Angptl4 (Fig. 3E). The activation of TLRs by their prototypical agonists upregulated the expression of canonical target genes, such as IL-6 and TNF-α (Supplementary Fig. S3B,C). In vitro and ex vivo studies showed that TLR2 and TLR4 agonists (Pam3CSK4 and LPS) suppressed Angptl4 expression in intestinal epithelial cells, whereas TLR5 activation by S. typhi flagellin robustly induced Angptl4 expression (Fig. 3F, G). To establish direct causality for TLR4-mediated Angptl4 repression, siRNA-mediated knockdown of TLR4 was performed during LPS stimulation. (Supplementary Fig. S3D). LPS induced marked increase in IL1B expression and intracellular oxidative stress in IECs, effects that were significantly attenuated upon TLR4 knockdown (Supplementary Fig. S4D). Notably, LPS-mediated suppression of Angptl4 was completely abolished in siTLR4-treated cells, demonstrating that TLR4 is required for LPS-dependent repression of Angptl4 (Fig. 3H). Flagellin stimulation caused a robust increase in Angptl4 expression that scaled with TLR5 abundance, indicating sensitivity to host TLR5 expression level (Fig. 3F, G). To define downstream signaling mechanisms, kinase inhibitor screening of canonical and non-canonical TLR5 pathways (Supplementary Fig. S3E) identified GSK3 and JNK as essential mediators of flagellin-induced Angptl4 expression (Fig. 3I). Inhibition of p38 MAPK, MEK1/2 and IKK/NF-κB signaling did not alter flagellin-induced Angptl4 expression, confirming pathway specificity (Supplementary Fig. S3F). In silico promoter analysis identified EGR1 as a candidate transcriptional regulator of Angptl432–34, which was validated by qChIP showing increased EGR1 occupancy at the Angptl4 promoter following flagellin stimulation (Fig. 3J). siRNA-mediated knockdown of EGR1 significantly reduced flagellin-induced Angptl4 expression, establishing EGR1 as a necessary downstream effector of TLR5 signaling (Fig. 3K, Supplementary Fig. S3G). These findings uncover a novel mechanism of Angptl4 regulation through TLR5-EGR1 signaling. By analyzing the gut microbiome, we further investigated how shifts in microbial MAMPs, especially LPS and flagellins, influence Angptl4-mediated gut permeability. Metagenomic profiling revealed a reduction in microbial α-diversity (Supplementary Fig. S3H, I) in LIDPAD-fed mice compared to those on the control diet. β-diversity analysis also showed significant differences in microbial composition between the two dietary groups (Supplementary Fig. S3J, K). Notably, LIDPAD-fed mice exhibited a significantly reduced ratio of Gram-negative to Gram-positive bacteria (Fig. 3L and Supplementary Fig. S3H, K, L), suggesting a diet-induced change in LPS abundance and reflecting alterations in the microbiome’s functional capacity. Given recent classification of flagellins as “silent,” “stimulator,” or “evader” types based on their interactions with TLR5 receptors35,36, we curated murine flagellin datasets by analyzing paired metagenomic and metatranscriptomic data to identify microbial MAMP changes in our models36 (Supplementary Fig. S4A). Our findings reveal a significant shift in the microbial MAMP profile of LIDPAD-fed mice, with an increased stimulator-to-silent flagellin ratio by week 2 (Fig. 3M). Similar changes in the ratios of Gram-negative to Gram-positive bacteria and stimulator-to-silent flagellin were also observed in high-fat diet (HFD)-induced MASLD models37 (Supplementary Fig. S4B, C), further supporting the idea that such shifts during gut microbiome remodeling in diet-induced MASLD potentially contribute to gut barrier dysfunction and disease progression. To functionally interrogate the impact of flagellin subtypes, we focused on two Gram-positive bacterial species, Romboutsia hominis and Roseburia hominis, selected for their distinct flagellin structures. Romboutsia produces stimulatory flagellin, while Roseburia produces silent flagellins38. Recombinant flagellins from both species were cloned, expressed, and purified, with protein purity confirmed by SDS-PAGE and functional binding to TLR5 independently validated (Supplementary Fig. S4D, E). In patient-derived human differentiated intestinal organoids (Supplementary Fig. S4F–H), flagellins from S. typhi and Romboutsia hominis significantly induced Angptl4 expression, whereas Roseburia hominis flagellin had no effect (Fig. 3N). Stimulator flagellins induced canonical TLR5-mediated expression of TNFα, whereas silent flagellins did not (Supplementary Fig. S4I). Functionally, stimulator flagellins caused a significant drop in TEER within 6-12 h that persisted for 24 h, an effect rescued by blocking antibodies against Angptl4 (mAb11F6C4) (Fig. 3O). These findings establish Angptl4 as the primary mediator of stimulator flagellin-induced epithelial barrier disruption. To expand these findings in vivo, we developed gnotobiotic mouse models colonized with either Roseburia hominis or Romboutsia hominis. Germ-free mice on a control diet were gavaged with the bacteria during the first week to establish colonization (Supplementary Fig. S4J). Mice colonized with Roseburia hominis gained weight more rapidly than those colonized with Romboutsia hominis (Supplementary Fig. S4K), although both groups showed similar glucose tolerance after two weeks (Supplementary Fig. S4K). Romboutsia hominis colonization significantly increased gut permeability, evidenced by elevated FITC-dextran levels in plasma, while Roseburia hominis maintained gut barrier integrity (Fig. 3P). These effects were corroborated by altered expression of tight junction genes (Supplementary Fig. S4L). Consistent with early gut-liver axis activation, Romboutsia hominis colonization resulted in increased hepatic 4-HNE accumulation compared with Roseburia hominis (Fig. 3Q and Supplementary Fig. S4M). These results highlight the distinct roles of silent and stimulator flagellins in regulating gut barrier function and their broader implications for gut-liver communication in metabolic disorders. Together, these data define an integrated regulatory framework in which dietary fatty acids and microbial MAMPs converge on intestinal Angptl4 to control epithelial barrier integrity (Fig. 3R). From a dietary perspective, LIDPAD-derived fatty acids directly induce Angptl4 expression through PPARα-dependent transcriptional activation. Simultaneously, LIDPAD-induced dysbiosis reshapes microbial MAMP availability through two coordinated shifts. First, a reduction in the Gram-negative to Gram-positive bacteria ratio is associated with diminished LPS availability, thereby relieving TLR4-mediated suppression of Angptl4 expression. Second, an increased ratio of stimulator to silent flagellins enhances TLR5 signaling, driving Angptl4 induction via a GSK3-JNK-EGR1 transcriptional axis. Through these combined dietary and microbiome-dependent mechanisms, Angptl4 emerges as a molecular checkpoint that integrates lipid sensing and microbial pattern-recognition signals to destabilize the gut barrier, providing a mechanistic basis for early gut–liver axis dysfunction in MASLD. Dietary Reversion Restores Gut Signaling While Inulin Fails to Reverse Dysbiosis To explore strategies for improving gut permeability in MASLD, we compared two dietary interventions: reversion to a control diet (MASL_R) following LIDPAD-induced MASLD, and inulin supplementation for MASH (LIDPAD+inulin) (Fig. 4A). Inulin has shown variable efficacy in MASLD in previous studies38–40. Following LIDPAD induction, we assessed the impact on intestinal Angptl4 expression, gut barrier integrity, TLR expression and MAMP profiles.Fig. 4Dietary reversion (MASL_R) and inulin supplementation (LIDPAD+Inulin) differentially impact gut microbiome composition, TLR signaling, and Angptl4 expression.A Schematic of dietary interventions corresponding to panels B-G. Two distinct experimental timelines are shown. In the dietary reversion model (top), mice were fed LIDPAD (red line) for 4 weeks to induce MASLD, followed by 4 weeks of control diet (MASL_R), yielding a total duration of 8 weeks. Control mice were maintained on a control diet throughout. In the inulin supplementation model (bottom), mice were maintained on LIDPAD for a total of 12 weeks (red line), with or without 10% (w/w) inulin supplementation (LIDPAD + inulin) (gray). B–G Data are presented according to weeks on diet/treatment within each respective experimental arm, not as a shared timeline across all groups. B Relative expression of Angptl4 in the intestinal mucosa across control, dietary reversion, and inulin supplementation groups. C FITC-dextran permeability assay indicating plasma levels of FITC-dextran across control, dietary reversion, and inulin supplementation groups, after FITC-dextran gavage. D Relative mRNA level of TLR2, 4, 5 and 6 across different regions of the small intestine of mice on the indicated feeding regimes. E Box plot of the ratio of Gram-negative to Gram-positive bacteria across 2 weeks, dietary reversion, and inulin supplementation groups. F, G Box plot showing the abundances of total flagellins (F) and the ratio of stimulator-to-silent flagellin (G) across control, dietary reversion, and inulin supplementation groups. H Schematic representation of gut microbiome and host signaling changes in MASL_R (left) and LIDPAD+inulin (right) interventions following LIDPAD-induced MASLD. Both interventions exhibit a persistently low Gram-negative to Gram-positive bacterial ratio, suggesting microbiome hysteresis and a stabilized dysbiotic equilibrium, despite dietary changes. TLR4 expression is downregulated in both groups, leading to diminished LPS-mediated repression of Angptl4. However, TLR5 expression diverges, with low TLR5 in MASL_R limiting the effect of stimulator flagellins, resulting in Angptl4 levels similar to control, while high TLR5 in LIDPAD+inulin enables stimulator flagellins to sustain elevated Angptl4 expression via GSK3-JNK-EGR1 signaling. MASL_R restores gut barrier integrity and reduces hepatic oxidative stress, whereas LIDPAD + inulin fails to mitigate LIDPAD-induced dysbiosis, sustaining gut permeability and oxidative stress. These findings highlight the complex interplay between diet, microbial MAMPs, and host signaling, where dietary reversion partially restores gut homeostasis, but inulin supplementation fails to reverse dysbiosis-induced effects. Created in BioRender. Li, L. (2026). For (B–F): n = 5 mice per group. Data expressed as mean ± S.D. For (E, F), boxplots show the median (center line), the 25th and 75th percentiles (bounds of the box), and whiskers extending to maxima and minima. ***p < 0.001, **p < 0.01, *p < 0.05 (one-way ANOVA followed by Tukey’s HSD test). n.s. denotes not significant. Dietary reversion (MASL_R) and inulin supplementation (LIDPAD+Inulin) differentially impact gut microbiome composition, TLR signaling, and Angptl4 expression. A Schematic of dietary interventions corresponding to panels B-G. Two distinct experimental timelines are shown. In the dietary reversion model (top), mice were fed LIDPAD (red line) for 4 weeks to induce MASLD, followed by 4 weeks of control diet (MASL_R), yielding a total duration of 8 weeks. Control mice were maintained on a control diet throughout. In the inulin supplementation model (bottom), mice were maintained on LIDPAD for a total of 12 weeks (red line), with or without 10% (w/w) inulin supplementation (LIDPAD + inulin) (gray). B–G Data are presented according to weeks on diet/treatment within each respective experimental arm, not as a shared timeline across all groups. B Relative expression of Angptl4 in the intestinal mucosa across control, dietary reversion, and inulin supplementation groups. C FITC-dextran permeability assay indicating plasma levels of FITC-dextran across control, dietary reversion, and inulin supplementation groups, after FITC-dextran gavage. D Relative mRNA level of TLR2, 4, 5 and 6 across different regions of the small intestine of mice on the indicated feeding regimes. E Box plot of the ratio of Gram-negative to Gram-positive bacteria across 2 weeks, dietary reversion, and inulin supplementation groups. F, G Box plot showing the abundances of total flagellins (F) and the ratio of stimulator-to-silent flagellin (G) across control, dietary reversion, and inulin supplementation groups. H Schematic representation of gut microbiome and host signaling changes in MASL_R (left) and LIDPAD+inulin (right) interventions following LIDPAD-induced MASLD. Both interventions exhibit a persistently low Gram-negative to Gram-positive bacterial ratio, suggesting microbiome hysteresis and a stabilized dysbiotic equilibrium, despite dietary changes. TLR4 expression is downregulated in both groups, leading to diminished LPS-mediated repression of Angptl4. However, TLR5 expression diverges, with low TLR5 in MASL_R limiting the effect of stimulator flagellins, resulting in Angptl4 levels similar to control, while high TLR5 in LIDPAD+inulin enables stimulator flagellins to sustain elevated Angptl4 expression via GSK3-JNK-EGR1 signaling. MASL_R restores gut barrier integrity and reduces hepatic oxidative stress, whereas LIDPAD + inulin fails to mitigate LIDPAD-induced dysbiosis, sustaining gut permeability and oxidative stress. These findings highlight the complex interplay between diet, microbial MAMPs, and host signaling, where dietary reversion partially restores gut homeostasis, but inulin supplementation fails to reverse dysbiosis-induced effects. Created in BioRender. Li, L. (2026). For (B–F): n = 5 mice per group. Data expressed as mean ± S.D. For (E, F), boxplots show the median (center line), the 25th and 75th percentiles (bounds of the box), and whiskers extending to maxima and minima. ***p < 0.001, **p < 0.01, *p < 0.05 (one-way ANOVA followed by Tukey’s HSD test). n.s. denotes not significant. LIDPAD feeding induced a rapid increase in intestinal Angptl4 expression within four weeks. Reversion to the control diet (MASL_R) significantly reduced Angptl4 expression, whereas inulin supplementation (LIDPAD+inulin) failed to attenuate the elevated Angptl4 levels (Fig. 4B). MASL_R mice exhibited marked improvement in gut barrier integrity, while LIDPAD+inulin group showed persistent gut permeability similar to LIDPAD-fed animals (Fig. 4C). Consistent with barrier restoration, MASL_R mice displayed normalization of intestinal TLR2, TLR5, and TLR6 expression to levels observed in control-fed mice (Fig. 4D). In contrast, TLR expression in the LIDPAD+inulin group remained similar to that of LIDPAD-fed mice, indicating limited impact of inulin on host TLR regulation (Fig. 4D). Notably, TLR4 expression remained low in both intervention groups, pointing to a diminished suppressive effect of LPS on Angptl4 expression. Next, we examined how these interventions affected microbial composition and MAMP diversity. Neither MASL_R nor LIDPAD+inulin restored the Gram-negative to Gram-positive bacterial ratio, which remained similar to that of LIDPAD-fed mice. Although α-diversity in both Gram-negative and Gram-positive bacteria increased, the Gram-positive bacteria returned to levels similar to control-fed mice (Fig. 4E, Supplementary Fig. S5A). Total flagellin levels rose significantly by week 4 of LIDPAD feeding and stabilized at a lower level by week 12, remaining elevated relative to control-fed mice. Neither intervention reduced total flagellin levels (Fig. 4F). Both dietary strategies restored the stimulator-to-silent flagellin ratio to control or lower levels (Fig. 4G). Integration of host and microbial data revealed that the balance between LPS and flagellin signaling is a key determinant of Angptl4 expression (Fig. 4H). In both the MASL_R and LIDPAD+inulin groups, a persistently low Gram-negative to Gram-positive bacterial ratio and reduced TLR4 expression diminished LPS-mediated repression of Angptl4. In contrast, the impact of stimulator flagellins diverged between the two interventions due to differences in host TLR5 expression. In the LIDPAD+inulin group, elevated TLR5 expression enabled stimulator flagellins to sustain high levels of Angptl4, whereas in MASL_R mice, TLR5 expression was low, returning to control levels, rendering stimulator flagellins less effective in maintaining Angptl4 levels (Fig. 4F, G). To determine whether microbial metabolite recovery contributed to these outcomes, we assessed gut microbial short-chain fatty acid (SCFA) biosynthetic capacity using functional metagenomic profiling. LIDPAD feeding markedly reduced predicted SCFA pathway abundance, including butyrate biosynthesis, relative to control-fed mice (Supplementary Fig. S5B). Inulin supplementation on the LIDPAD background failed to significantly restore these SCFA-associated pathways, which remained substantially suppressed compared with controls (Supplementary Fig. S5B). Importantly, the limited recovery of SCFA biosynthetic capacity did not correlate with improvements in Angptl4 expression, gut barrier integrity, or hepatic outcomes, indicating that SCFA availability is not the primary determinant of Angptl4-mediated barrier dysfunction in this model. MASL_R mice achieved ~ 12% body weight reduction and exhibited improved glucose tolerance (Supplementary Fig. S5C), whereas inulin supplementation had no significant effect on body weight or glucose homeostasis (Supplementary Fig. S5D). MASL_R group also showed substantial reductions in hepatic lipid accumulation, while LIDPAD + inulin mice exhibited no improvement in liver histology (Supplementary Fig. S5E). These results suggest that adding prebiotic fiber alone cannot offset LIDPAD-induced damage. Hepatic transcriptomic profiling confirmed these phenotypic outcomes: MASL_R reversed the MASLD-associated gene expression signature, restoring pathways related to metabolism, inflammation, and fibrosis (Supplementary Fig. S5F), indicating a recovery of liver health and gut-liver communication. In contrast, LIDPAD+inulin worsened chronic inflammation and fibrosis compared to LIDPAD alone, with minimal changes in metabolic processes (Supplementary Fig. S5G). Collectively, this comparison highlights the tightly coupled relationship between dietary interventions, intestinal Angptl4 regulation, gut barrier integrity, and downstream liver pathology during MASLD progression (Fig. 4H). Restoration of metabolic and barrier homeostasis required not only changes in microbial composition, but coordinated recalibration of host pattern-recognition signaling and microbial functional outputs. Dietary reversion effectively reduced intestinal Angptl4 expression by normalizing host TLR signaling, limiting the impact of stimulator flagellins, and restoring epithelial barrier integrity, thereby alleviating hepatic oxidative stress and injury. In contrast, inulin supplementation failed to disrupt the dysbiosis-induced signaling state, as persistent alterations in TLR expression and MAMP availability sustained Angptl4 induction despite partial changes in microbial diversity. These findings demonstrate that successful intervention in MASLD depends on resetting the integrated host-microbe signaling network that governs Angptl4-dependent barrier function, rather than simply augmenting individual microbial metabolites or selectively modifying microbial taxa. The limited efficacy of inulin, therefore, underscores the challenge of correcting complex diet-induced dysbiosis using single-nutrient strategies. Altered gut microbial ecosystem in humanized microbiota-modified mice with MASLD To enhance the translational relevance and capture the complexity of human gut microbial ecosystems, we used fecal microbiota transplantation (FMT) from healthy human donors to generate human microbiota-modified (HMM) mice (Fig. 5A). This model permits evaluation of diet–microbiome interactions within a diverse microbial background that more closely reflects human gut ecology than conventional or gnotobiotic systems. HMM mice fed a LIDPAD diet exhibited significant weight gain and impaired glucose tolerance over 15 weeks compared with control-fed HMM mice, as reflected by increased IGTT area under the curve values (Fig. 5B, C). Histological examination of liver tissues revealed pronounced hepatic steatosis, ballooning, inflammation, and fibrosis in LIDPAD-fed HMM mice (Fig. 5D, E), confirming that LIDPAD induces MASLD features in a humanized microbiota context.Fig. 5Human microbiota-modified (HMM) mice exhibit MASLD progression on LIDPAD.A Schematic diagram of the experimental design. Fecal microbiota transplantation (FMT) from healthy human donors to antibiotic-induced microbiome-depleted mice (AIMD), followed by LIDPAD or control diet feeding. B Body weight progression in humanized-microbiota model (HMM) mice on LIDPAD vs control diet over 16 weeks. C Mean IGTT AUC values in HMM mice on LIDPAD vs control diet at different time points. D Representative liver histology (H&E and Picrosirius Red staining) of HMM mice on LIDPAD and control diets at baseline, 4, 8, and 12 weeks. n = 5 independent experiments. Scale bars: 100 μm. E Stacked bar graph showing the progression of steatosis, ballooning, inflammation, and fibrosis scores in HMM mice on LIDPAD vs control diet over time. F Experimental schematic of probiotic experiments on HMM with control and MASLD. Mice were given FMT gavage for 4 weeks to generate HMM. Subsequently, mice were fed on a control or LIDPAD diet for 4 weeks, while receiving probiotic gavages of Roseburia hominis, Romboutsia hominis, or vehicle, respectively (n = 3 per group). FITC-dextran permeability assay and body weight changes are assessed on week 0 (baseline) and week 4. G Body weight progression in HMM mice on LIDPAD + vehicle, LIDPAD + Roseburia hominis, LIDPAD + Romboutsia hominis, control + vehicle, control diet + Roseburia hominis, control diet + Romboutsia hominis after 4 weeks. H FITC-Dextran permeability assay of HMM mice on LIDPAD + vehicle, LIDPAD + Roseburia hominis, LIDPAD + Romboutsia hominis, control + vehicle, control diet + Roseburia hominis, control diet + Romboutsia hominis after 4 weeks. I Representative liver histology (H&E and Picrosirius Red staining) of HMM mice on LIDPAD + vehicle, LIDPAD + Roseburia hominis, control + vehicle, control diet + Romboutsia hominis vs control diet after 4 weeks. Stacked bar graph showing the mean of steatosis, ballooning, inflammation, and fibrosis scores in respective groups. Scale bars: 200 μm. For (A–I), n = 5 mice per group. For (B), data are presented as mean ± SEM. C, G and H, data are presented as box plots, showing the median (center line), the 25th and 75th percentiles (bounds of the box), and whiskers extending to maxima and minima. with individual data points. *p < 0.05, **p < 0.01, ***p < 0.001 (one-way ANOVA followed by Tukey’s HSD test), n.s. denotes not significant. Human microbiota-modified (HMM) mice exhibit MASLD progression on LIDPAD. A Schematic diagram of the experimental design. Fecal microbiota transplantation (FMT) from healthy human donors to antibiotic-induced microbiome-depleted mice (AIMD), followed by LIDPAD or control diet feeding. B Body weight progression in humanized-microbiota model (HMM) mice on LIDPAD vs control diet over 16 weeks. C Mean IGTT AUC values in HMM mice on LIDPAD vs control diet at different time points. D Representative liver histology (H&E and Picrosirius Red staining) of HMM mice on LIDPAD and control diets at baseline, 4, 8, and 12 weeks. n = 5 independent experiments. Scale bars: 100 μm. E Stacked bar graph showing the progression of steatosis, ballooning, inflammation, and fibrosis scores in HMM mice on LIDPAD vs control diet over time. F Experimental schematic of probiotic experiments on HMM with control and MASLD. Mice were given FMT gavage for 4 weeks to generate HMM. Subsequently, mice were fed on a control or LIDPAD diet for 4 weeks, while receiving probiotic gavages of Roseburia hominis, Romboutsia hominis, or vehicle, respectively (n = 3 per group). FITC-dextran permeability assay and body weight changes are assessed on week 0 (baseline) and week 4. G Body weight progression in HMM mice on LIDPAD + vehicle, LIDPAD + Roseburia hominis, LIDPAD + Romboutsia hominis, control + vehicle, control diet + Roseburia hominis, control diet + Romboutsia hominis after 4 weeks. H FITC-Dextran permeability assay of HMM mice on LIDPAD + vehicle, LIDPAD + Roseburia hominis, LIDPAD + Romboutsia hominis, control + vehicle, control diet + Roseburia hominis, control diet + Romboutsia hominis after 4 weeks. I Representative liver histology (H&E and Picrosirius Red staining) of HMM mice on LIDPAD + vehicle, LIDPAD + Roseburia hominis, control + vehicle, control diet + Romboutsia hominis vs control diet after 4 weeks. Stacked bar graph showing the mean of steatosis, ballooning, inflammation, and fibrosis scores in respective groups. Scale bars: 200 μm. For (A–I), n = 5 mice per group. For (B), data are presented as mean ± SEM. C, G and H, data are presented as box plots, showing the median (center line), the 25th and 75th percentiles (bounds of the box), and whiskers extending to maxima and minima. with individual data points. *p < 0.05, **p < 0.01, ***p < 0.001 (one-way ANOVA followed by Tukey’s HSD test), n.s. denotes not significant. Metagenomic analysis revealed a reduction in the Gram-negative to Gram-positive bacterial ratio in the LIDPAD-fed HMM mice, consistent with previous findings (Supplementary Fig. S6A). Analysis of microbial co-occurrence and exclusion patterns revealed distinct community structures between the control-HMM and LIDPAD-HMM groups. Control-HMM group maintained a balanced and stable microbial ecosystem, characterized by a dynamic but non-competitive relationship between the flagellated genera Roseburia and Romboutsia (Supplementary Fig. S6B). In contrast, LIDPAD feeding induced a marked ecological shift, characterized by a strong co-exclusion relationship between Roseburia and Romboutsia, indicating disruption of microbial community structure. Despite comparable overall flagellin abundance, these Gram-positive taxa exerted opposing functional effects on intestinal Angptl4 expression and barrier integrity, that microbial function rather than taxonomic presence alone governs host responses. To directly test whether these strain-specific effects persist within a humanized context, HMM mice were colonized with Roseburia hominis or Romboutsia hominis after establishing a human microbiota, under the control or LIDPAD diet. Mice first received FMT from healthy human donors for 4 weeks to establish the HMM, and were subsequently fed control or LIDPAD diet for an additional 4 weeks while receiving weekly gavages of Roseburia hominis (10⁸ CFU), Romboutsia hominis (10⁸ CFU), or vehicle control (Fig. 5F). This design enabled direct assessment of bacterial strain effects across dietary contexts while maintaining the complexity of host-microbe and microbe-microbe interactions. Control-fed HMM mice maintained stable body weight throughout the 4-week intervention period, whereas LIDPAD-fed mice receiving vehicle gained substantial weight (Fig. 5G). Roseburia hominis supplementation significantly attenuated LIDPAD-induced weight gain, while exerting minimal effects on body weight under the control diet. In contrast, Romboutsia hominis supplementation exacerbated weight gain in LIDPAD-fed mice yet paradoxically caused significant weight reduction in control-fed mice (Fig. 5G). These divergent metabolic responses indicate that the physiological impact of individual bacterial strains is strongly modulated by dietary context and host metabolic state. Gut permeability, measured by serum FITC-dextran at baseline (week 0) and endpoint (week 4), increased progressively in LIDPAD-fed HMM mice receiving vehicle control. Notably, regardless of dietary background, Romboutsia hominis gavage consistently induced significant gut barrier dysfunction, whereas Roseburia hominis transplantation markedly reduced diet-induced increases in intestinal permeability (Fig. 5H). Histological analysis and SAF scoring revealed that LIDPAD+vehicle HMM mice developed moderate-to-severe hepatic steatosis and early fibrosis, which were attenuated by Roseburia hominis transplantation (Fig. 5l). In contrast, both LIDPAD-fed and control-fed HMM mice receiving Romboutsia hominis exhibited pronounced hepatic lipid accumulation and significant fibrosis (Fig. 5I), consistent with their elevated gut permeability and altered metabolic trajectories. Control-fed mice receiving Roseburia hominis or vehicle control displayed minimal steatosis and no detectable fibrosis, as expected. Taken together, these findings demonstrate that the impact of gut bacteria on the gut-liver axis is strain-specific and highly context-dependent within a humanized microbiota. Roseburia hominis confers protection against diet-induced barrier dysfunction and hepatic injury, whereas Romboutsia hominis promotes epithelial permeability and liver pathology even in the absence of dietary stress. These results underscore that microbial contributions to MASLD progression are governed by functional effects on barrier integrity and host signaling, rather than by microbial abundance or taxonomy alone. Dysbiotic gut microbiome in patients with MASLD MASLD poses a significant public health challenge in Asia, with an estimated prevalence of 34%41. To address the lack of non-invasive measures that reflect gut barrier-associated disease activity, we evaluated Angptl4 mRNA derived from exfoliated intestinal epithelial cells in stool samples. A training cohort of 288 patients with MASLD from Thailand was recruited. These patients exhibited significantly elevated dysbiosis scores, characterized by reduced α-diversity and disrupted microbial homeostasis (Supplementary Fig. S7A, B). Both Gram-negative and Gram-positive bacterial populations were reduced, resulting in a lower Gram-negative to Gram-positive bacteria ratio than in no MASLD controls (Fig. 6A, B). Taxonomic analysis revealed increased Romboutsia and decreased Roseburia abundance in patients with MASLD (Fig. 6C). Patients with MASLD also exhibited distinct β-diversity patterns, indicating divergence in microbial community structure (Supplementary Fig. S7C). These microbial features are consistent with alterations observed in our murine models and support a link between dysbiosis, gut barrier perturbation, and intestinal Angptl4 regulation rather than simple taxonomic shifts.Fig. 6Fecal Angptl4 expression fecal samples as a diagnostic for gut permeabilty.A, B Violin plots of α-diversity (Chao1) of Gram-negative and Gram-positive bacteria, adjusted p-values: 0.001 (***), 0.002 (**) (A) and the ratio of Gram-negative to Gram-positive bacteria abundance, adjusted p-value = 0.033 (*) (B) in healthy and patients with MASLD as determined by ultrascan-based transient elastography. C Violin plot showing the relative abundance of Romboutsia and Roseburia genera in patients with MASLD and healthy volunteers. D Principal component analysis (PCA) depicting the relationship of explanatory clinical variables and MASLD progression stages. Individual points indicate each patient and are colored by their MASLD staging. Clinical variables are represented by vectors whose length (magnitude) and directionality depict their relationship with other clinical parameters. E Violin plots of normalized fecal Angptl4 levels (-dCT) in healthy subjects, patients with MASLD, and patients with HCC. Data are mean ± SD, adjusted p-values for all comparisons: All groups: 0.011 (*), HCC v No MASLD: 0.0015 (**), HCC v No MASLD: 0.002 (**), MASLD v No MASLD: 0.0110 (*). F Violin plots of normalized fecal Angptl4 levels (-dCT) in non-diabetic, prediabetic, and diabetic subjects. Data are mean ± SD, adjusted p-values for all comparisons: All groups: 0.021 (*), Simple steatosis v No MASLD: 0.02 (*), MASH v No MASLD: 0.03 (*), MASH with cirrhosis v No MASLD: 0.01 (**), MASH with cirrhosis v MASLD: 0.003 (**). The rest of the comparisons are n.s. G Scatter plot showing the relationship between intestinal Angptl4 mRNA expression ( − dCT) and FIB-4 score in the human cohort. Each dot represents an individual subject. Two-sided Pearson correlation test, adjusted for age and gender. R = 0.16, p = 0.031. H, I Violin plots of normalized fecal Angptl4 levels (-dCT) in healthy subjects and patients with MASLD from Shenzhen, China MASLD cohort, p-value = 0.001 (H) and India MASLD cohort, p-value = 0.0069 (I). For Fig. 6A–E, H, I, statistical testing is performed using two-sided multivariate analysis of variance adjusted for age and gender. Data are mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001. Fecal Angptl4 expression fecal samples as a diagnostic for gut permeabilty. A, B Violin plots of α-diversity (Chao1) of Gram-negative and Gram-positive bacteria, adjusted p-values: 0.001 (***), 0.002 (**) (A) and the ratio of Gram-negative to Gram-positive bacteria abundance, adjusted p-value = 0.033 (*) (B) in healthy and patients with MASLD as determined by ultrascan-based transient elastography. C Violin plot showing the relative abundance of Romboutsia and Roseburia genera in patients with MASLD and healthy volunteers. D Principal component analysis (PCA) depicting the relationship of explanatory clinical variables and MASLD progression stages. Individual points indicate each patient and are colored by their MASLD staging. Clinical variables are represented by vectors whose length (magnitude) and directionality depict their relationship with other clinical parameters. E Violin plots of normalized fecal Angptl4 levels (-dCT) in healthy subjects, patients with MASLD, and patients with HCC. Data are mean ± SD, adjusted p-values for all comparisons: All groups: 0.011 (*), HCC v No MASLD: 0.0015 (**), HCC v No MASLD: 0.002 (**), MASLD v No MASLD: 0.0110 (*). F Violin plots of normalized fecal Angptl4 levels (-dCT) in non-diabetic, prediabetic, and diabetic subjects. Data are mean ± SD, adjusted p-values for all comparisons: All groups: 0.021 (*), Simple steatosis v No MASLD: 0.02 (*), MASH v No MASLD: 0.03 (*), MASH with cirrhosis v No MASLD: 0.01 (**), MASH with cirrhosis v MASLD: 0.003 (**). The rest of the comparisons are n.s. G Scatter plot showing the relationship between intestinal Angptl4 mRNA expression ( − dCT) and FIB-4 score in the human cohort. Each dot represents an individual subject. Two-sided Pearson correlation test, adjusted for age and gender. R = 0.16, p = 0.031. H, I Violin plots of normalized fecal Angptl4 levels (-dCT) in healthy subjects and patients with MASLD from Shenzhen, China MASLD cohort, p-value = 0.001 (H) and India MASLD cohort, p-value = 0.0069 (I). For Fig. 6A–E, H, I, statistical testing is performed using two-sided multivariate analysis of variance adjusted for age and gender. Data are mean ± SD. *p < 0.05, **p < 0.01, ***p < 0.001. Principal Component Analysis (PCA) showed clear separation between no MASLD controls and patients across the MASLD spectrum, including MASLD, MASH, and MASH with cirrhosis (Fig. 6D). Metabolic parameters such as BMI, triglycerides (TG), and diabetes clustered with early disease states, emphasizing their connection to disease development. Liver function markers, including the AST/ALT ratio and fibrosis markers like FIB-4 and transient elastography (TE), correlated with advanced MASLD stages, reinforcing their relevance in assessing disease severity. Fecal Angptl4 positioned at the intersection of metabolic and fibrotic vectors within the PCA space, indicating that it captures a biological dimension related to gut epithelial stress that spans metabolic and hepatic injury domains. In this cohort, ~ 90% of patients with MASLD were either diabetic (47-49%) or pre-diabetic (40–44%), highlighting the significant role of impaired glucose homeostasis as a risk factor for MASLD development (Supplementary Fig. S7D). Fecal Angptl4 levels were significantly elevated in patients with MASLD compared with no MASLD individuals (Fig. 6E). Instead of showing strict stage discrimination, fecal Angptl4 levels were most strongly associated with metabolic risk status, increasing in parallel with diabetic and pre-diabetic states (Supplementary Fig. S7E). When patients were grouped by fibrosis category, fecal Angptl4 levels were higher on average in groups with greater fibrosis burden, but substantial overlap was observed across fibrosis stages (Fig. 6F and Supplementary Fig. S7F). Consistent with this pattern, fecal Angptl4 showed a weak but statistically significant correlation with the FIB-4 index (Fig. 6G), indicating partial association with fibrosis-related features rather than strong fibrosis staging capability. Importantly, fecal Angptl4 levels did not correlate with the overall dysbiosis score, highlighting its specificity as a marker of gut barrier integrity, rather than of broader microbiome changes (Supplementary Fig. S7G). Likewise, fecal Angptl4 showed no significant correlation with circulating markers of endotoxemia or acute epithelial injury, including LPS-binding protein (LBP) and intestinal fatty acid–binding protein (I-FABP), which also did not correlate with each other (Supplementary Fig. S7H, I). Furthermore, among these gut barrier markers, only I-FABP correlate significantly with liver disease severity FIB4 score (Supplementary Fig. S7J). This dissociation indicates that fecal Angptl4 reflects a chronic intestinal epithelial transcriptional response associated with barrier stress, distinct from acute epithelial damage or systemic endotoxin exposure. To assess generalizability across different populations, we expanded our analysis to include independent Asian cohorts. In a Chinese cohort of 154 individuals, including 52 no-MASLD controls and 102 early patients with MASLD diagnosed by hepatic ultrasound, fecal Angptl4 levels were significantly elevated in patients with MASLD (Fig. 6H). Among controls, 27% were prediabetic, and 73% were normoglycemic, whereas patients with MASLD showed a higher prevalence of metabolic dysfunction, with 22% diagnosed with diabetes, 24% with prediabetes, and 55% without diabetes (Supplementary Fig. S7K). Consistent with the Thailand cohort, fecal Angptl4 enrichment tracked with metabolic status and diabetic risk rather than discrete disease stage (Supplementary Fig. S7L). In an independent Indian cohort of 69 patients with advanced MASH diagnosed by transient elastography, fecal Angptl4 levels were also significantly elevated compared with controls (Fig. 6I). Taken together, these multi-center data demonstrate that fecal Angptl4 functions as an exploratory, non-invasive risk-stratification marker reflecting gut barrier-associated metabolic stress across the MASLD spectrum. Its enrichment in early MASLD, strong association with metabolic dysfunction, weak but significant overlap with fibrosis indices, and dissociation from systemic endotoxemia markers support its utility for identifying patients with heightened gut-liver axis perturbation, rather than as a stand-alone diagnostic or fibrosis-staging biomarker. Discussion The gut-liver axis plays a critical role in the development of MASLD, with intestinal barrier integrity serving as a key checkpoint that limits the translocation of microbial and dietary signals that can trigger hepatic injury and inflammation42,43. While gut barrier dysfunction has been repeatedly linked to MASLD, the molecular mechanisms connecting dietary and microbial cues to barrier destabilization have remained incompletely defined. In this study, we identify intestinal Angptl4 as a central mediator that integrates dietary lipid sensing and microbiota-derived pattern-recognition signals to regulate epithelial and endothelial barrier integrity, thereby promoting early hepatic oxidative stress and downstream liver pathology. Using intestinal-specific Angptl4 knockout (Angptl4Vil1-/-) mice, we demonstrate that loss of Angptl4 confers robust protection against diet-induced gut permeability, vascular leakiness, and hepatic oxidative damage. Preservation of tight junction and endothelial markers, together with enrichment of pathways related to cell-cell adhesion and morphogenesis, supports a primary role for Angptl4 in barrier destabilization rather than inflammatory initiation. Mechanistically, we and others have previously shown that Angptl4 activates key signaling pathways, including PI3K/AKT, JNK, and MEK/ERK, culminating in activation of transcription factors such as AP1, CREB, and NF-κB16,44. Notably, NF-κB signaling has been reported to downregulate specific tight junction components, a change associated with increased intestinal permeability and diarrhoeal phenotypes in human disease45. In the present study, however, NF-κB activity did not regulate Angptl4 expression, and claudin-1 protein abundance was not altered in LIDPAD-fed mice, indicating that Angptl4-mediated barrier dysfunction occurs independently of claudin-1 transcriptional modulation in this context. Instead, prior studies have shown that Angptl4 can weaken vascular integrity through interactions with integrin α5β1 and VE-cadherin, promoting endothelial leakiness via integrin α5β1-dependent Rac1/PAK signaling46. Beyond the gut, Angptl4 has also been shown to induce epithelial permeability in the lung, contributing to pulmonary edema, reinforcing its role as a regulator of epithelial and endothelial barrier function47–49. Our findings extend these established barrier-regulatory mechanisms to the gut-liver axis, showing that intestinal Angptl4 links microbial dysbiosis and dietary factors to hepatic injury in MASLD, thereby broadening its functional role beyond metabolic regulation to include pathological disruption of epithelial and endothelial barriers. At the mechanistic level, we delineate three convergent regulatory inputs controlling intestinal Angptl4 expression. First, dietary fatty acids activate PPARα, directly increasing Angptl4 transcription. Second, LPS-TLR4 signaling functions as a negative regulatory axis that constrains Angptl4 expression. In MASLD-associated dysbiosis, reduced LPS availability and diminished TLR4 signaling attenuate this repression, leading to de-repression of Angptl4 and contributing to its elevated expression. Third, dysbiosis-associated stimulator flagellins engage TLR5 signaling to induce Angptl4 via a GSK3-JNK-EGR1 axis. Together, these pathways converge to destabilize gut barrier integrity, amplifying microbial and dietary flux to the liver. Notably, PPARα has also been shown to regulate FABP1, thereby facilitating fatty acid uptake and contributing to lipid accumulation in MASLD50. These intersecting pathways position Angptl4 at the interface of barrier regulation and metabolic flux, providing a mechanistic explanation for how dietary and microbial cues cooperatively drive early MASLD progression. Our functional dissection of microbial flagellin subtypes refines current understanding of microbiota-host interactions in metabolic disease. While earlier studies have linked increased flagellin levels in metabolic inflammatory issues51, we demonstrate that functional differences between flagellins are critical determinants of barrier integrity. Flagellins from Roseburia hominis exert minimal effects on Angptl4 expression and preserve barrier integrity, whereas flagellins from Romboutsia hominis significantly induce Angptl4 and disrupt epithelial integrity. By integrating in vitro systems, gnotobiotic models, and humanized microbiota-modified mice, we establish causality between specific microbial drivers and Angptl4-dependent barrier dysfunction. Additionally, we developed a robust flagellin database and analysis pipeline integrating paired metagenomic and metatranscriptomic datasets, enabling the identification of actionable microbial MAMPs that disrupt gut barrier health. This marks a significant shift, moving beyond associative microbiome analyses toward functionally actionable microbial signals. Dietary intervention studies further highlight the resilience of diet-induced dysbiosis. The persistent low Gram-negative to Gram-positive bacterial ratio observed in both MASL_R and LIDPAD+inulin groups indicates that LIDPAD feeding creates a stable dysbiotic state resistant to short-term intervention. Although dietary reversion effectively restored host signaling, reduced intestinal Angptl4 expression, and improved barrier and hepatic outcomes, microbial community composition remained only partially reversible, suggesting that prior dietary exposure exerts lasting effects on microbiota structure and function even after diet normalization. In contrast, inulin supplementation failed to produce comparable benefits, highlighting the limitations of single-nutrient approaches52. The limited restoration of SCFA production capacity by inulin, despite sustained barrier dysfunction and MASLD features, indicates that the pathological phenotype in the LIDPAD model is not primarily driven by SCFA deficiency. Instead, barrier failure tracks with MAMP-dependent regulation of Angptl4, obviating the need for butyrate rescue in this context. Such microbial imprinting has been reported in human obesity and NAFLD, where prior prolonged dietary patterns reshape microbial ecosystems in ways that are not readily reversed by lifestyle modification alone53. Host-microbiome crosstalk may further reinforce this altered equilibrium, as persistent dysbiosis has been observed in patients with MASLD, despite dietary or lifestyle intervention, sustaining endotoxemia and hepatic stress. Consistent with this notion, downregulation of TLR4 in both intervention groups suggests attenuated LPS-mediated signaling, which may alter epithelial immune responses and antimicrobial peptide production, thereby stabilizing the dysbiotic state. Together, these findings underscore that diet-induced microbiome changes are not easily reversible and emphasize the need for comprehensive dietary and microbiome-targeted strategies, beyond dietary reversion or single-nutrient supplementation alone, to restore gut-liver homeostasis effectively. To enhance translational relevance, we employed human microbiota-modified (HMM) mice, which recapitulate key features of human microbial ecology. Dysbiosis induced by a MASLD-promoting diet disrupted microbial community dynamics, particularly the balance between Roseburia and Romboutsia. This imbalance elevated Angptl4 expression and gut permeability, thereby worsening liver pathology. Together, these findings reinforce the principle that microbial functional outputs, rather than taxonomic composition alone, govern gut-liver axis dysfunction, and underscore the value of humanized models for evaluating microbiome-targeted interventions. When interpreting these findings, it is important to consider the biological context in which Angptl4 operates. Angptl4 is a well-established inhibitor of lipoprotein lipase and related lipases; however, its canonical lipase-inhibitory activity occurs predominantly within the vascular compartment, where it regulates peripheral lipid uptake rather than luminal lipid digestion. Supporting this spatial restriction, preservation of gut barrier integrity in intestinal-specific Angptl4-deficient mice despite identical dietary lipid exposure argues against altered lipid handling as the primary mechanism underlying increased permeability in this model. Rather, our data support a temporal framework in which Angptl4-dependent barrier destabilization and hepatic oxidative stress represent early events in MASLD, preceding extensive immune infiltration or fibrotic remodeling, with additional pathways likely contributing at later disease stages. Similarly, fecal Angptl4 should not be interpreted as a diagnostic or fibrosis-staging biomarker. While elevated across MASLD cohorts and weakly associated with fibrosis indices, fecal Angptl4 exhibits substantial overlap across disease stages and instead appears to reflect gut barrier-associated metabolic stress, capturing a biological dimension distinct from systemic endotoxemia or acute epithelial injury, as supported by its dissociation from circulating LBP and I-FABP. Finally, we acknowledge the limitations of using the Gram-negative to Gram-positive bacterial ratio as a sole metric for dysbiosis. While this ratio decreases in patients with MASLD and LIDPAD-fed mice, it oversimplifies the complexity of microbial ecosystems. For example, certain Gram-negative species, such as Akkermansia muciniphila, exert protective effects by producing lipooligosaccharide (LOS), which activates anti-inflammatory TLR2 signaling and counteracts the pro-inflammatory effects of TLR4 pathways54. This highlights the need for a more refined approach to understanding microbial-derived molecules, including distinct LPS and LOS subtypes, within broader microbial network interactions. Future studies leveraging advanced techniques such as mass spectrometry will be crucial for characterizing these molecular signatures with greater specificity. The expression of Angptl4 at the intestinal epithelium is subject to opposing regulation by distinct pattern-recognition receptor pathways, adding an important layer of contextual control to its barrier-modulatory function. TLR4 activation by lipopolysaccharide has been reported to suppress Angptl4 expression in macrophages and epithelial cells, whereas TLR5 activation by bacterial flagellin directly induces Angptl4 expression, consistent with our findings that flagellin-mediated TLR5 signaling is a dominant upstream driver of Angptl4-dependent barrier disruption in the LIDPAD model. This opposing regulation suggests that the composition of the luminal microbial signal, specifically the relative burden of flagellated versus LPS-producing bacteria, may determine the net Angptl4 tone at the epithelium, thereby linking microbial community structure to intestinal permeability and downstream metabolic inflammation in MASLD. Furthermore, therapeutic inhibition of Angptl4, via genetic or immunoneutralization, represents a promising strategy to preserve or restore gut barrier integrity. By attenuating Angptl4-driven permeability, such interventions may restrict dysbiosis-induced gut–liver or gut-tissue crosstalk and its metabolic sequelae. Extended 14-week immunoneutralization studies are ongoing and show early evidence of improvement in MASLD-associated pathology, beyond the scope of this study. By elucidating how dietary factors and specific microbial components regulate Angptl4 expression, our findings establish a foundation for developing gut-targeted therapies and precision nutrition strategies for MASLD management. In conclusion, this study advances the understanding of gut-liver communication by confirming intestinal Angptl4 as a central mediator linking dietary factors, microbial dysbiosis, and liver injury. Our findings provide a solid basis for innovative therapeutic approaches integrating diet, microbiota, and molecular mechanisms to combat MASLD and other gut-liver disorders. Methods Animals and diets All experiments were approved and in compliance with the guidelines of Nanyang Technological University Institutional Animal Care and Use Committee (A19032, A20055, A20173). Male wild-type C57BL/6 J mice (Invivos, Singapore) aged 8-9 weeks were fed with LIDPAD (PD23081401, Syse bio, China) and control diets (PD23081404, Syse Bio, China), anthropomorphic measurements and organs were harvested at the indicated time as consistent with previous work23. Animals were housed in standard cages maintained at 30 ± 2 °C (within the thermoneutral zone) and 49.9% relative humidity, with a regulated 12-hour light–dark cycle. The inulin-supplemented diet comprised of LIDPAD with 10% inulin (w/w)55. Inulin compacted into feed pellets. Intestinal epithelial-specific Angptl4-deleted mice (Angptl4Vil1-/-) were generated by crossing Villin(Vil1)-cre transgenic mice (Jackson Laboratory, strain #: 004586)56 with Angptl4fl/fl mutant mice (EUCOMM/KOMP repository). Experiments involving Angptl4fl/fl and Angptl4Vil1-/- mice are age-, sex-, and cage-matched littermates, and performed with mice fed diets in parallel, phenotyped at the same time points, and processed using identical protocols. Patient cohorts All patients consented to the cohort studies, and these studies were approved by the Institutional Review Board of their respective institutions: Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand (Approval No. 049/63, 957/64), Shenzhen People’s Hospital, Shenzhen, China (Approval No. LL-KY-2024170-01) and AIIMS, Bathinda (IEC/AIIMS/BTI/2024/11/02). For our discovery cohort, 203 patients with MASLD, 61 patients with HCC and 30 no MASLD subjects were recruited at King Chulalongkorn Hospital, Bangkok, Thailand, and processed at the Faculty of Medicine, Chulalongkorn University. Inclusion criteria were the following: adult Thai patients diagnosed with hepatic steatosis via a controlled attenuation parameter (CAP, > 248 dB/m) using the M or XL probes as appropriate. Participants who met any of the following conditions were excluded from the study: (i) significant alcohol consumption ( ≥ 30 g for men and ≥20 g for women), (ii) the presence of other liver diseases, such as chronic viral hepatitis and autoimmune hepatitis, (iii) a history of other types of cancers. Participants were advised against taking any antibiotics, prebiotics, or probiotic supplements for at least 3 months before enrollment. For validation cohorts, we leveraged on patient recruitment from Shenzhen People’s Hospital, Shenzhen, China, where patients with MASLD were diagnosed using ultrasound instead of FibroScan. A second validation cohort is recruited at All India Institute of Medical Sciences (AIIMS), Bathinda, India, with the same inclusion criteria as the discovery cohort. All patients were de-identified, and demographic, blood panel and liver function test data were collected (Supplementary Table S1, 2). Fecal samples were collected in DNA/RNA Shield and stored at − 80 oC within 24 h. Histology, immunohistochemistry and immunofluorescence Harvested liver and intestine samples are processed, imaged and scored as previously described57. Immunofluorescence and immunohistochemistry were performed as previously described58. For 4-HNE immunohistochemistry staining, sectioned slides are heat-fixed at 60 oC for 30 min before rehydration in a graded series of organic solvents xylene and ethanol. Subsequently, antigen retrieval is performed with sodium citrate buffer (pH 6) in the automated Antigen Retriever 2100. Slides were washed before blocking with 10% FBS, 1% BSA in TBS for 1 h at room temperature. After this, the slides were incubated in an anti-mouse 4HNE antibody, followed by a DylightTM 594 conjugated goat anti-mouse polyclonal secondary antibody (Thermofisher Scientific, Cat #35510). The primary and secondary antibodies used are listed in Supplementary Table S3. Then, nuclei were stained with Hoescht’s dye. Samples are imaged using Axioscan Z1 (Zeiss, Germany). VE-cadherin coverage was comprehensively quantified relative to CD31+ endothelium using five complementary Imaris parameters: (1) total VE-cadherin area as a fraction of total CD31 area (lower fraction = reduced junctional coverage and increased permeability); (2) VE-cadherin/CD31 area ratio specifically within villi microvasculature; (3) VE-cadherin/CD31 area ratio within submucosal vascular plexus; (4) villus microvascular diameter (CD31 staining); (5) submucosal vessel diameter (CD31 staining); and (6) villus microvascular plexus length normalized to total villus length. This multi-parametric approach objectively assesses both junctional integrity and vascular morphology across LIDPAD conditions and Angptl4 genotypes, with ~ 200 villi analyzed per section from n = 3 mice/group. Multivariate analysis Clinical data is consolidated and aggregated with dysbiosis score and the ratio of gram-negative to gram-positive bacteria. Each clinical parameter is z-scaled before PCA calculation using FactoMineR R package59 and represented in a biplot using factoextra R package60. FITC-dextran intestinal permeability assay Mice were fasted for 4 h and given a FITC-dextran dose (80 mg/mL in PBS) at 600 mg/kg body weight via oral gavage. After 4 h, blood sample was collected via the tail-snip method. Extracted plasma was then diluted (1:5 v/v in PBS), and fluorescence was measured at excitation and emission λ values of 485 nm and 528 nm, respectively. The concentration of FITC-dextran in the plasma sample was determined using a calibration curve from known concentrations of FITC-dextran. Hepatocyte ROS measurement Harvested livers were broken down into single cells after digestion in Accutase (Sigma–Aldrich) using a 21 G needle and syringe. The cell suspension was subjected to RBC lysis (155 mM NH4Cl, 10 mM KHCO3, 0.1 mM EDTA). The cells were then stained using CellROX™ (Thermo Fisher) according to the manufacturer’s instructions. Viable cells (unstained) were identified using propidium iodide (PI) staining. Flow cytometry was then carried out to determine the number of live cells and the intensity of the CellROX™ stain. Cell culture NCM460 (INCELL Corporation LLC, USA) normal human intestinal epithelial cell line is cultured in DMEM supplemented with 10% FBS. Fatty acids (200 µM) were added to cells complexed with BSA (2.5:1). Cells were treated for 12 h unless otherwise indicated. Transepithelial electrical resistance (TEER) was measured using an electrical cell-substrate impedance sensing system (Applied Biophysics) as previous described46. Resistance values for each microelectrode were normalized as the percentage of measured resistance to baseline resistance and were plotted as a function of time. TLR agonist treatments were performed using a human TLR1-9 agonist kit per their recommended protocol (Invivogen, France, #tlrl-kit1hw). For kinase inhibitor experiments, NCM460 cells were pre-treated with kinase inhibitors, SB-21673, CC-401, SB242235, MEK162, obtained from SYN2103 kinase inhibitor (SYNKinase) and TargetMol libraries at IC80 concentrations, and vehicle controls for 24 h. Subsequently, pre-treated cells are challenged with TLR5 agonizts. siRNA experiments are performed with Accell Human TLR4 siRNA (Horizon, UK, #E-008088-00-0005), Accell Human EGR1 siRNA (Horizon, UK, #E-006526-00-0005), and negative control Accell Non-targeting Pool (Horizon, UK, #D-001910-10-05) with the manufacturer’s recommended protocol. Organoid experiments Adjacent normal tissues of colorectal cancer patients were biopsied, cultured and maintained in a 3D organoid system as previously described61. Briefly, the tissue biopsies were minced into small fragments and washed thoroughly with cold PBS containing 2% penicillin/streptomycin, 50 mg/L gentamicin, and 5 mg/L amphotericin B. These tissue fragments were dissociated in digestion buffer to obtain single cells for Matrigel embedding. After Matrigel solidification, the organoid culture medium was added. Organoids cultures were passaged every 5-7 days. Subsequently, at passages 4 to 7, organoids were dissected and seeded on semipermeable transwell inserts (70000 cells/inset) that had been precoated with Matrigel. The upper insert and lower compartment were both supplemented with serum-free organoid culture media. Once the cells reached 100% confluency, organoids were cultured for an additional three days before use (Supplementary Fig. S4F–H). Recombinant flagellins were added at a concentration of 10 nM and treated for 24 h. TEER of organoids was measured using Millicell® ERS-2 epithelial cell volt-ohm meter (Millipore) at 6 h intervals. Resistance values for each microelectrode were normalized as the percentage of measured resistance to baseline resistance and are plotted as a function of time. Biopsies were obtained from consenting patients, and the study is approved by the Institutional Review Board of Shenzhen People’s Hospital, Shenzhen, China (Approval No. LL-KY-2024170-01). Chromatin immunoprecipitation (ChIP) Cells were fixed with 0.5 % formaldehyde at 37 °C for 10 min, then sonicated in SDS lysis buffer (1 % SDS, 10 mM EDTA, 50 mM Tris-HCl, pH 8.1) to shear chromatin to fragments of 200–500 bp. Ten percent of the lysate was set aside as input; the remainder was immunoprecipitated with using antibodies against human PPARα (Santa Cruz, Heidelberg, Germany) on NCM460 treated with either vehicle or fatty acids, and human EGR-1 (#4154, Cell Signaling Technology, USA) for either vehicle or flagellin (Invivogen, France, #tlrl-kit1hw) treatments respectively, and normal IgG was used as negative control. Immune complexes were captured using Protein A/G beads (Santa Cruz) and eluted in elution buffer (1 % SDS, 0.1 M NaHCO₃). Cross-links were reversed by heating at 65 °C for 6 h before detection of respective DNA fragments, this method is as previously described44. Fecal RNA extraction and RTqPCR Host RNA from fecal samples were homogenized at room temperature using glass bead beating. Following centrifugation, the supernatant was collected, and RNA were extracted according to the ZymoBiomics RNA Miniprep kit (for Thai, China cohort) and Make Himedia HiPurA® Multi-Sample RNA Purification Kit (for India cohort). Purified RNA was reverse transcribed into cDNA using Quantabio qSCRIPT cDNA SuperMix® kit (for Thai cohort), TAKARA Toyobo PrimeScriptTM (for China cohort), Make Himedia Hi-cDNA Synthesis Kit (for India cohort). KOD SYBR® qPCR mix (for Thai cohort), ThunderbirdTM Next SYBR qPCR mix (for China cohort), MAKE – SIGMA KAPA SYBR® FAST (for India cohort). Respective forward and reverse primer sequences used are found in Supplementary Table S4. The PCR protocol was as the manufacturer’s recommendations and previously described58. Real-time qPCR instruments used were Applied Biosystems StepOnePlus (for Thai, India cohort), BIOER LineGene 9600 Plus (for China cohort). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting summary Transparent Peer Review file Supplementary Information Reporting summary Transparent Peer Review file Source data Source data Source data