Liver endothelial zonation orchestrates hepatic steatosis onset through retinoic acid–regulated FGF1

Abstract

The contribution of liver sinusoidal endothelial cell (LSEC) zonation to metabolic dysfunction–associated steatotic liver disease (MASLD) pathogenesis remains undefined. We identified selective lipid deposition in the pericentral zone during early MASLD. Multiomics analyses confirmed enhanced pericentral lipid metabolism in both hepatocytes and LSECs. Mechanistically, pericentral LSEC marker c-Kit transcriptionally activated FGF1 via nuclear receptor RXRG, which suppressed hepatocellular lipid accumulation through FGFR4 signaling. Retinoic acid (RXRG’s endogenous ligand and active vitamin A metabolite) phenocopied FGF1’s antisteatotic effects. Clinical data revealed an inverse correlation between dietary vitamin A and MASLD severity, suggesting therapeutic potential of vitamin A supplementation for early intervention.

Pericentral LSEC c-Kit activates FGF1 via RXRG to reduce hepatic lipids, and vitamin A boosts FGF1 via RXRG to alleviate MASLD.

INTRODUCTION

Metabolic dysfunction–associated steatotic liver disease (MASLD) has emerged as the most prevalent chronic liver disorder globally, affecting ~30 to 40% adult population (1–3). Despite its escalating burden, MASLD remains a therapeutic enigma due to its heterogenous pathogenesis involving intricate interactions among lipotoxicity, insulin resistance, oxidative stress, immune-inflammatory pathway, and extrahepatic factors (4–8). Illuminating the mechanisms and disrupting lipid deposition at the early stage will be promising in preventing MASLD progression.

Hepatic cells originating from periportal (PP) and pericentral (PC) zones exhibit distinct gene expression and metabolic profiles, termed liver zonation. Key regulators include oxygen gradients, Wnt/β-catenin signaling, R-spondin proteins, and E3 ligases ZNRF3/RNF43 (9). While liver regeneration, toxin exposure, and autoimmune hepatitis manifest zonated pathologies (10–13), the role of zonation in MASLD progression remains elusive. Clinical and preclinical evidence indicates PC-predominant lipid deposition (14, 15) and MASLD-driven zonal metabolic reprogramming (16). Recent work by He et al. (17) uncovered the spatial regulation of lipid metabolism by insulin, indicating the zonal-specific initiation of hepatic steatosis.

Liver sinusoidal endothelial cells (LSECs), which are highly specialized endothelial cells, maintain liver homeostasis through spatial zonation. Halpern et al. (18) pioneered paired-cell sequencing to establish the spatial transcriptomic atlas of liver endothelial cells. Inverso et al. (19) further stratified zonated LSECs based on c-Kit expression gradients spanning central-to-portal axes, yielding critical insights into vascular zonation. Building on these foundations, we demonstrated PC LSECs’ pivotal role in supporting hepatocyte regeneration (20) and later revealed that age-driven metabolic disturbances reshape endothelial zonation (21). These findings implicate zonal LSEC heterogeneity in modulating liver pathophysiology. Given LSEC capillarization and dysfunction during steatosis-inflammation-fibrosis progression (22), this study systematically investigates their zonation-dependent contributions to MASLD initiation.

Briefly, we demonstrated hepatic lipid accumulation in the PC zone during early MASLD progression. c-Kit, a master regulator of liver endothelial zonation, exhibited antisteatotic properties. Mechanistic studies revealed that PC LSECs produce fibroblast growth factor 1 (FGF1) through retinoid X receptor gamma (RXRG)–dependent transcriptional activation. Retinoic acid (RA), the endogenous RXR ligand and active vitamin A metabolite, attenuated MASLD by potentiating LSEC-derived FGF1 signaling. Clinical data further established an inverse association between dietary vitamin A levels and MASLD severity, implying that adequate dietary vitamin A supplementation may potentially prevent MASLD onset.

RESULTS

Zonated depostion of hepatic lipids

In high-fat diet (HFD)–fed murine models, hematoxylin and eosin (H&E) and Oil Red O (ORO) staining demonstrated selective PC lipid deposition compared to PP regions (Fig. 1A). This spatial pattern was conserved in human MASLD, with mild cases exhibiting PC-predominant lipid accumulation (Fig. 1B). Glutamine synthetase (GS)⁺ regions consistently colocalized with fat cavitations in MASLD livers (Fig. 1F), and zonal lipid distribution (61.9% of patients) correlated with lower SAF (Steatosis, Activity, Fibrosis) scores and fat cavitation areas than pan-lobular patterns (Fig. 1, G and H, and fig. S1B), revealing that early lipid deposition in patients with MASLD initiated from PC zone. Because sedentary lifestyle and aging also contribute to hepatic steatosis, we observed that age- and sedentary-induced murine models replicated PC-specific lipid deposition via ORO (Fig. 1C) and BODIPY (Boron-dipyrromethene)/GS costaining (Fig. 1D), while aging exacerbated zonal steatosis (Fig. 1E and fig. S1A). The above results collectively suggested that early lipid deposition in either mouse or human liver originated from PC zone. To further observe the dynamic changes of hepatic lipid accumulation in MASLD models with varying degrees of severity, we established mouse MASLD models induced by HFD feeding at different time points. Longitudinal HFD studies revealed initial PC-specific ORO⁺ foci at 8 weeks progressing to confluent PC cavitations by 12 weeks, with hepatic steatosis expanding with the extension of feeding time (Fig. 1I and fig. S1C). Zonal quantification [central vein (CV)–portal vein (PV) axis stratification, Fig. 1J] confirmed that at HFD 12-week group, layer 1 showed the most fat cavitations and ORO-positive staining, and with the severity of MASLD increases (24-week group), hepatic lipid distribution may vary (Fig. 1, K and L). Notably, the NAFLD activity score (NAS) progressively increased with prolonged HFD feeding, indicating a gradual exacerbation of MASLD severity (fig. S1D). Consistently, serum levels of alanine transaminase (ALT), aspartate transaminase (AST), total bilirubin (TBIL), and direct bilirubin (DBIL) were significantly elevated, while serum albumin (ALB) was decreased after 24 weeks of HFD feeding, suggesting prominent hepatocellular injury (fig. S1E). Concurrently, hepatic levels of tumor necrosis factor–α (TNFα), interleukin-1β (IL-1β), IL-6, C-C motif chemokine ligand 2 (CCL2), and CXCL10 were markedly increased (fig. S2, A and B). Immunostaining for F4/80 and CD11b revealed prominent macrophage accumulation at 24 weeks of HFD feeding, indicative of obvious hepatic inflammation (fig. S2, C and D). Collectively, spatial analyses across species and models established PC zonation as the primary site of lipid initiation in MASLD pathogenesis.

Fig. 1. Zonated depostion of hepatic lipids.

(A) Representative images of liver H&E and ORO staining of MASLD mice fed HFD for 12 weeks. PV and CV were indicated in the images. See also (B), (C), (D), (E), (F), and (I). Scale bars, 100 (left) and 50 (right) μm. n = 6. (B) Representative images of liver H&E staining of MASLD patients. Scale bar, 500 μm. n = 21. (C) H&E and ORO staining in liver sections from 7-month-old wild-type mice. Scale bar, 200 μm. n = 6. (D) BODIPY and GS immunostaining in liver sections from 7-month-old wild-type mice. Scale bar, 200 μm. n = 6. (E) H&E and ORO staining in liver sections from young (8-week-old) and aging (2-year-old) mice. Scale bar, 200 μm. n = 5 to 7. (F) GS immunostaining in liver sections of MASLD patients and healthy control. Scale bar, 200 μm. n = 6. (G) Representative images of H&E staining of different hepatic steatotsis distributing patterns and their ratio in MASLD patients. Scale bar, 200 μm. n = 1 to 13. (H) Liver SAF score of zonated and pan-lobular pattern of MASLD patients. n = 7 to 13. (I) H&E, ORO and GS immunostaining in liver sections from chow-fed and HFD-fed mice for 8, 12, 16, and 24 weeks. Scale bar, 200 μm. n = 6. (J) Schematic diagram of liver lobule zonation. (K) Fat cavitation area proportion of different lobular layers in MASLD mice fed HFD for 12 and 24 weeks. n = 6. (L) ORO-positive area proportion of different lobular layers in MASLD mice fed HFD for 12 and 24 weeks. n = 6. Data are means ± SD. ****P < 0.0001 by two-sample t test. W, weeks.

MASLD reshapes hepatocyte metabolic zonation

To investigate molecular mechanisms underlying zonal lipid handling, we conducted single nucleus RNA sequencing (snRNA-seq) on chow-fed mouse hepatocytes. Stratification using established PC and PP hepatocyte markers revealed enhanced fatty acid metabolism in PC populations (fig. S3, A to C), indicating intrinsic PC predisposition for lipid deposition. Spatial transcriptomics confirmed zonation patterns, with clusters exhibiting vascular proximity-dependent distributions (fig. S3, D and E). The heatmap analysis of Halpern et al.’s (18) zonal markers enabled precise lobule mapping (fig. S3, F and G), demonstrating progressive PC marker up-regulation and PP marker down-regulation during MASLD (fig. S3, H and I), indicative of metabolic zonation remodeling. Cluster analysis delineated PP-enriched clusters (Chow: cluster 1 and 4; HFD: cluster 0 and 3) versus PC-enriched clusters (Chow: cluster 2 and 3; HFD: cluster 1 and 2) (fig. S3, D and E). Gene set enrichment analysis (GSEA) confirmed conserved PC enrichment of fatty acid metabolism across diets (fig. S3, J and K), proving that robust fatty acid metabolism in PC cluster was its inherent characteristic. Fatty acid beta oxidation and long-chain fatty acid biosynthetic process were enriched in PP clusters in chow group, while this trend was disrupted in HFD group, suggesting lipid zonation disruption in MASLD (fig. S3L). Glucose metabolism, tricarboxylic acid cycle, and inflammatory response genes were all enriched in chow-PP cluster, and oxidative phosphorylation, glycolysis/gluconeogenesis, and inflammatory response genes were enriched in HFD-PP cluster, while cytochrome p450 genes were enriched in chow and HFD-PC cluster, implying distinct functions between PC and PP clusters (fig. S4A). These findings established PC hepatocytes as primary targets in metabolic dysregulation.

c-Kit governs liver endothelial zonation

Liver endothelial cells exhibit distinct zonal gene expression patterns alongside hepatocytes (18, 19). Among them, c-Kit emerges as a key regulator of endothelial zonation. Immunofluorescence costaining demonstrated predominant c-Kit expression in PC LSECs with a graded PC-to-PP decrease (Fig. 2A). This spatial distribution was validated by Halpern’s paired-seq data (18) (Fig. 2, B and F) and Inverso’s flow cytometry–based stratification of LSECs into four c-Kit gradient subpopulations (19). Transcriptomic analyses confirmed progressive c-Kit enrichment from PV to CV LSECs (Fig. 2C). Furthermore, c-Kit expression in endothelial cells is mainly localized in murine liver and lung (Fig. 2D). Consistently, spatial c-Kit expression showed similar tendency in human liver samples (Fig. 2E). Spatial transcriptomics from (16) revealed c-Kit⁺/Cyp2e1⁺ colocalization (PC zone), contrasting with Cyp2f2⁺ regions (Fig. 2, G to I). To analyze the changes in gene expression patterns of endothelial cells in different zones before and after MASLD, we isolated c-Kit⁺ (PC) LSECs using c-Kit immunomagnetic beads, and the resting LSECs were defined as c-Kit⁻ (PP) LSECs. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis revealed baseline developmental pathway enrichment in chow-fed mice (Fig. 2J), shifting to metabolic pathways under HFD (Fig. 2K). GSEA analysis found that fatty acid metabolism pathways transitioned from PP-enriched (chow) to PC-enriched (HFD) LSECs (Fig. 2, L and M). The above results revealed that PC and PP LSECs underwent metabolic reprogramming during MASLD, particularly in fatty acid metabolism. Moreover, KEGG analysis suggested that HFD-induced MASLD specifically activated NAFLD pathways in c-Kit⁺ LSECs (Fig. 2, N and O). Together, the findings unraveled c-Kit–dependent metabolic reprogramming of LSEC zonation during MASLD progression.

Fig. 2. c-Kit governs liver endothelial zonation.

(A) c-Kit and GS immunostaining in liver sections from 8-week-old wild-type mice. Scale bar, 200 μm. n = 6. CV, central vein; PV, portal vein. (B) c-Kit expression in LSECs from CV to PV. Data were obtained from the article published by Halpern et al. (18). (C) c-Kit expression in LSECs from CV to PV. Data were obtained from the article published by Inverso et al. (19). (D) Endothelial c-Kit expression in mouse organs from EC atlas database (https://doi.org/10.1016/j.cell.2020.01.015). (E) C-Kit and GS immunostaining in liver sections from normal and MASLD patients. Scale bar, 200 μm. n = 6. (F) PP and PC marker gene expression in LSECs from PV to CV. Data were obtained from the article published by Halpern et al. (18). (G) Reanalysis of c-Kit and Cyp2e1 expression in spatial transcriptome sequencing of normal mouse liver published in GEO database (GSE248077). (H) Reanalysis of c-Kit and Cyp2f2 expression in spatial transcriptome sequencing of normal mouse liver published in GEO database (GSE248077). (I) Ratio of c-Kit positive spots overlapped with Cyp2e1 or Cyp2f2 positive spots. (J) GO and KEGG analysis of c-Kit^+/− LSECs isolated from chow-fed mice. (K) GO and KEGG analysis of c-Kit^+/− LSECs isolated from HFD-fed mice. (L) GSEA analysis of fatty acid metabolism-related pathways in c-Kit^+/− LSECs isolated from chow-fed mice. (M) GSEA analysis of fatty acid metabolism-related pathways in c-Kit^+/− LSECs isolated from HFD-fed mice. (N) GSEA analysis of nonalcoholic fatty liver disease pathway in c-Kit^+/− LSECs isolated from chow-fed mice. (O) GSEA analysis of nonalcoholic fatty liver disease pathway in c-Kit^+/− LSECs isolated from HFD-fed mice. DAPI, 4′,6-diamidino-2-phenylindole; FDR, false discovery rate.

PC endothelium-derived c-Kit protects liver from hepatic steatosis

To assess liver endothelial zonation alterations in MASLD, we analyzed PC marker expression in chow/MASLD LSECs. Most genes, particularly c-Kit, were down-regulated in MASLD LSECs (Fig. 3A). Immunofluorescence (IF) staining demonstrated progressive c-Kit reduction during HFD feeding, reaching significance at 12 weeks (Fig. 3, B and C). Quantitative polymerase chain reaction (qPCR) and immunoblotting confirmed c-Kit down-regulation at mRNA and protein levels (Fig. 3, D to F), suggesting its protective role. To validate this, we generated endothelial-specific c-Kit knockout (KO) mice (CDH5-CreERT c-Kit fl/fl) with HFD-induced MASLD. The ablation efficiency was tested after tamoxifen induction and after HFD feeding (Fig. 4, A to C). Compared to control (CT) mice, endothelial cell (EC) c-Kit KO exacerbated hepatic lipid deposition (Fig. 3, G to I) but did not change its spatial distribution pattern (Fig. 3J). Notably, 8-week HFD c-Kit KO mice developed early PC fat cavitations absent in controls (Fig. 3G). Extended 12-week HFD c-Kit KO mice exhibited higher NAS, ORO⁺ areas, and hepatic triglycerides (Fig. 3, M and N), with elevated F4/80⁺ inflammation and alpha–smooth muscle actin–positive fibrosis (Fig. 3, K and L). Moreover, serum ALT and AST levels were higher in c-Kit KO mice, implying worse liver functions (Fig. 3O). These findings established endothelial c-Kit as a critical suppressor of MASLD progression.

Fig. 3. PC endothelium-derived c-Kit alleviates hepatic steatosis.

(A) Heatmap of PC LSEC marker genes in chow and MASLD mice. (B and C) c-Kit and GS immunostaining in liver sections from MASLD mice fed HFD for 4, 8, and 12 weeks and control mice and quantitative analysis of c-Kit staining areas. Scale bar, 200 μm. n = 3 to 6. (D to F) qPCR, Western blot, and densitometric analysis of c-Kit in LSECs isolated from chow or HFD-fed mice. n = 3 to 4. (G to I) H&E and ORO staining and quantitative analysis of fat cavitation and ORO staining areas in liver sections from c-Kit KO or c-Kit CT mice fed HFD for 8, 10, and 12 weeks. Scale bar, 200 μm. n = 5. (J) Fat cavitation area and ORO staining area proportion of different lobular layers. n = 5. (K to O) Eight-week-old c-Kit KO and c-Kit CT mice were fed HFD for 12 weeks. (K and L) F4/80 and αSMA immunostaining in liver sections and quantitative analysis. Scale bar, 200 μm. n = 5. (M) Quantitative analysis of MASLD severity and ORO staining. n = 5. (N) Liver TC and TG concentration. n = 5. (O) Serum ALT and AST levels. n = 5. (P to R) Mice were fed HFD for 10 weeks and infused c-Kit^+/− LSECs for another 2 weeks. (P) H&E, ORO, F4/80, and αSMA immunostaining in liver sections. Scale bar, 200 μm. n = 5. (Q) MASLD severity, ORO, F4/80, and αSMA immunostaining was quantified by NAS and ORO, F4/80, and αSMA staining areas. n = 5. (R) Hepatic TC and TG. n = 5. Data are means ± SD. *P < 0.05; **P < 0.01, and ***P < 0.001; ****P < 0.0001; ns, not significant by two-sample t test (D, F, L to O, Q, and R) or one-way analysis of variance (ANOVA) (C and I). GAPDH, glyceraldehyde-3-phosphate dehydrogenase.

Fig. 4. c-Kit⁺ LSECs are rich in FGF1 production.

(A) Knockout efficiency of c-Kit after tamoxifen induction by qPCR. n = 5 to 7. (B) Knockout efficiency of c-Kit after tamoxifen induction by Western blot. n = 3. (C) Knockout efficiency of c-Kit after tamoxifen induction and HFD feeding for 12 weeks by Western blot. n = 4. (D) c-Kit⁺ LSECs highly expressed genes were crossed with mouse secretome genes. (E) Heatmap of screened genes in chow c-Kit^+/− LSEC RNA-seq. (F) FGF1 expression in LSECs from CV to PV (18). (G) FGF1 expression in LSECs from CV to PV (19). (H) ScRNA-seq cell clustering analysis of chow-fed mouse liver. (I) FGF1 expression in different cell types identified in (H). (J and K) Western blot and densitometric analysis of FGF1 in normal primary hepatic cell. n = 3. (L) qPCR analysis of FGF1 expressions in normal c-Kit^+/− LSECs. n = 3. (M and N) Western blot and densitometric analysis of FGF1 in normal c-Kit^+/− LSECs. n = 4. (O) FGF1 protein concentration in cultured medium of normal c-Kit^+/− LSECs. n = 4. (P) FPKM comparison of FGF1 in chow c-Kit^+/− LSEC RNA-seq. (Q) scRNA-seq analysis of LSEC FGF1 expression in chow- or HFD-fed mice. (R) qPCR analysis of FGF1 expressions in LSECs isolated from chow- or HFD-fed mice. n = 4. (S and T) Western blot and densitometric analysis of FGF1. n = 3. (U) FPKM comparison of FGF1 in HFD c-Kit^+/− LSEC RNA-seq. (V and W) Western blot and densitometric analysis of FGF1 in c-Kit^+/− LSECs isolated from HFD-fed mice. n = 3. Data are means ± SD. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, by one-way ANOVA (K), two-sample t test (A, C, L, N to P, R, T, U, and W), or Mann-Whitney U test (Q).

To assess the therapeutic potential of endothelial c-Kit in MASLD, we isolated c-Kit^+/− LSECs from 8-week-old wild-type mice and administered them via tail vein injection to MASLD mice. Compared to c-Kit⁻ LSEC recipients, c-Kit⁺ LSEC-transplanted mice demonstrated notably reduced hepatic lipid deposition, inflammation, and fibrosis (Fig. 3, P to R). These findings indicated that HFD-induced MASLD compromised c-Kit–mediated endothelial protection, while exogenous c-Kit⁺ LSEC supplementation effectively ameliorated key MASLD manifestations.

c-Kit⁺ LSECs are rich in FGF1 production

Next, we hypothesized that c-Kit⁺ LSECs may exert protective effects through certain secretory protein(s). Integrating c-Kit^+/− LSEC RNA-seq data (chow diet) with the mouse secretome genes (23) identified 28 candidate genes (Fig. 4, D and E). FGF1 demonstrated spatial coexpression with c-Kit across zonation axes (Fig. 4F), corroborated by independent spatial transcriptomic datasets conducted by Inverso et al. (19) (Fig. 4G). The single-cell RNA-seq (scRNA-seq) revealed FGF1 predominance in LSECs compared to hepatocytes, Kupffer cells, and other hepatic populations (Fig. 4, H and I). Above results were further validated by Western blot analysis (Fig. 4, J and K). c-Kit⁺ LSECs exhibited elevated FGF1 mRNA/protein levels versus c-Kit⁻ counterparts in chow-fed mice, with increased secretory output (Fig. 4, L to O). This phenotype was attenuated in MASLD, where LSECs showed reduced FGF1 expression (Fig. 4, Q to T). Chow c-Kit^+/− LSEC RNA-seq analysis showed higher FGF1 expression in c-Kit⁺ LSECs, while the difference disappeared in HFD c-Kit^+/− LSEC RNA-seq (Fig. 4, P and U), validated by immunoblotting (Fig. 4, V and W). Collectively, c-Kit⁺ LSECs maintained zone-specific FGF1 production that becomes dysregulated in MASLD.

FGF1 treats hepatic steatosis predominantly through FGFR4

To investigate FGF1’s role in lipid metabolism, steatotic alpha mouse liver 12 (AML12) cells were treated with recombinant mouse FGF1 (rFGF1). ORO staining and cell total cholesterol/triglyceride (TC/TG) analysis revealed that rFGF1 could effectively alleviate cellular lipid deposition (Fig. 5, A and B). Consistently, rFGF1 also ameliorated cellular lipid deposition in mouse primary hepatocytes (Fig. 5, C and D). Next, in vivo administration lowered fasting blood glucose and serum ALT and AST levels (Fig. 5, E and F) and ameliorated hepatic steatosis, inflammation, and fibrosis (Fig. 5, G to L). Next, we were committed to identifying the specific molecular targets of FGF1 in hepatocytes. FGFRs are the canonical receptors for FGFs. The FGFR family comprises four subtypes: FGFR1, FGFR2, FGFR3, and FGFR4. The qPCR of primary hepatocytes identified FGFR4 as the dominant FGFR subtype, followed by FGFR2 (Fig. 5M). HFD feeding decreased mRNA levels of FGFR1, FGFR2, and FGFR4 instead of FGFR3 (Fig. 5N). To verify whether FGF1 exerted beneficial effects through FGFRs, the pan-FGFR inhibitor infigratinib was used, which reversed rFGF1’s lipid-lowering effects (Fig. 5, O and P). Subtype-specific inhibition using FGFR1- to FGFR4-specific inhibitors demonstrated that FGFR4 mediated the primary therapeutic response (Fig. 5, Q and R). The above results established FGFR4 as the key receptor for FGF1’s antisteatotic action. To investigate the specific metabolic pathways mediating the lipid-lowering effect of FGF1, we examined the expression of lipogenesis/lipid uptake/fatty acid oxidation/lipid export pathway–related genes in steatotic primary mouse hepatocytes treated with rFGF1 and found that FGF1decreased cellular lipid accumulation mainly by decreasing lipogenesis and lipid uptake (Fig. 5S). The de novo lipogenesis (DNL)–related gene expression was further validated by Western blots (fig. S5, A to I). Notably, p-ACC1/ACC1 ratio increased, which was reported to inhibit the lipogenic ability of ACC1 (fig. S5, A and G) (24).

Fig. 5. FGF1 treats hepatic steatosis predominantly through FGFR4.

(A) ORO staining and quantitative analysis of steatotic AML12 cells treated with phosphate-buffered saline (PBS)/rFGF1. Scale bar, 50 μm. n = 4. (B) Cellular TC and TG concentration. n = 3. (C) ORO staining of steatotic mouse primary hepatocytes treated with PBS/rFGF1. Scale bar, 50 μm. n = 3. (D) Cellular TC and TG concentration. n = 3. (E to L) HFD mice administrated with PBS or rFGF1 (0.5 mg/kg) intraperitoneally every other day for another 3 weeks. n = 5 to 6. (E) Fasting blood glucose levels. (F) Serum ALT and AST levels. (G) H&E, ORO, F4/80, and αSMA immunostaining in liver sections. Scale bar, 200 μm. (H to K) Quantitative analysis of NAS and ORO, F4/80, and αSMA staining areas. (L) Liver TC and TG concentration. (M and N) qPCR analysis of FGFR1 to FGFR4 expressions in primary hepatocytes isolated from chow-fed or HFD-fed mice. n = 4. (O) Cellular TC and TG concentration of AML12 cells treated with vehicle or PO or PO + rFGF or PO + rFGF1 + pan-FGFR inhibitor infigratinib. n = 3. (P) ORO staining and quantitative analysis. Scale bar, 50 μm. n = 6. (Q) Cellular TC and TG concentration of AML12 cells treated with vehicle or PO or PO + rFGF1 or PO + rFGF1 + FGFR1 to FGFR4 inhibitor. n = 3. (R) ORO staining and quantitative analysis. Scale bar, 50 μm. n = 6. (S) qPCR analysis of Srebp1c, FASN, SCD1, ACC1, PPARγ, CD36, PPARα, CPT1α, ApoB, and MTTP gene expression in steatotic primary hepatocytes treated with PBS/rFGF1. n = 3. Data are means ± SD. *P < 0.05; **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, by two-sample t test (E, F, H to L, N, Q, R, and S) or one-way ANOVA (A, B, D, M, O, and P).

Endothelial c-Kit counteracts hepatic steatosis in an FGF-1–dependent manner

To determine whether c-Kit regulates FGF1 expression, we examined FGF1 expression in endothelial c-Kit KO and control LSECs. qPCR and Western blot analyses demonstrated notable reductions in FGF1 mRNA and protein expression in c-Kit KO cells (Fig. 6, A to C), accompanied by diminished FGF1 secretion (Fig. 6D). To evaluate whether c-Kit–mediated protection depends on FGF1, we infused c-Kit^+/− LSECs into MASLD mice and observed elevated hepatic FGF1 protein levels in c-Kit⁺ LSEC recipients (Fig. 6, E and F). Rescue experiments using rFGF1 in endothelial-specific c-Kit KO mice on an HFD (Fig. 6G) restored hepatic FGF1 expression (Fig. 6, H and I) and improved fasting blood glucose (Fig. 6J). Crucially, rFGF1 treatment reversed c-Kit KO–induced hepatic steatosis, inflammation, fibrosis, and liver dysfunction (Fig. 6, K to R). To avoid the systemic effects of rFGF1, we constructed endothelial cell–specific FGF1 KO mice, and the ablation efficiency was examined (Fig. 7, A to C). Hepatic FGF1 concentrations notably decreased in FGF1 KO mice liver (Fig. 7, D and E). Consistently, EC-specific FGF1 KO aggravated liver injury, hepatic steatosis, inflammation, and fibrosis (Fig. 7, F to I). To further verify that c-Kit⁺ LSECs exert their protective effects in an FGF1-dependent manner, we isolated c-Kit⁺ LSECs from endothelial cell–specific FGF1 KO mice and performed adoptive transfer experiments in HFD-fed MASLD mice and found that c-Kit⁺ LSECs with FGF1 deficiency failed to ameliorate MASLD-related phenotypes, indicating that c-Kit⁺ LSECs exerts its effects on MASLD predominantly through FGF1 secretion (Fig. 7, J to M). These findings established PC endothelial c-Kit as a critical regulator of FGF1 secretion, mediating protective effects in MASLD.

Fig. 6. Endothelial c-Kit counteracts hepatic steatosis in an FGF-1-dependent manner.

(A) qPCR analysis of FGF1 expressions in LSECs isolated from 8-week-old chow-fed c-Kit KO or c-Kit CT mice. n = 5 to 7. (B and C) Western blot and densitometric analysis of FGF1 in LSECs isolated from 8-week-old chow-fed c-Kit KO or c-Kit CT mice. n = 3 to 4. (D) FGF1 protein concentration in cultured medium of c-Kit^+/− LSECs isolated from 8-week-old chow-fed mice. n = 3. (E and F) Western blot and densitometric analysis of FGF1 in livers of HFD-fed mice infused with c-Kit^+/− LSECs isolated from 8-week-old chow-fed mice. n = 5. (G to R) 8-week-old c-Kit KO or c-Kit CT mice were fed HFD for 12 weeks. At the beginning of 10-week HFD, rFGF1 (0.5 mg/kg) was administrated intraperitoneally (i.p.) every other day for another 3 weeks. (G) Dosing regimen diagram of rFGF1. (H and I) Western blot and densitometric analysis of FGF1 in liver. n = 4. (J) Fasting blood glucose levels. n = 4 to 8. (K and L) MASLD severity and ORO staining was quantified by NAS and ORO staining areas. n = 4 to 8. (M) H&E and ORO staining in liver sections. Scale bar, 200 μm. n = 4 to 8. (N) F4/80 and αSMA immunostaining in liver sections. Scale bar, 200 μm. n = 4 to 8. (O and P) F4/80 and αSMA immunostaining was quantified by F4/80 and αSMA staining areas. n = 4 to 8. (Q) Liver TC and TG concentration. n = 4 to 8. (R) Serum ALT and AST levels. n = 4 to 8. Data are means ± SD. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, by two-sample t test (A, C to E, H, and J) or one-way ANOVA (K, L, and O to R).

Fig. 7. Endothelial cell-specific FGF1 KO abrogated the therapeutic effects mediated by c-Kit⁺ LSECs.

(A) Total RNA was extracted from primary LSECs isolated from EC-specific FGF1 CT and FGF1 KO mice after tamoxifen induction and qRT-PCR was performed to examine FGF1 mRNA expression. n = 5. (B and C) Western blot and densitometric analysis of FGF1 in livers of LSEC-specific FGF1 KO or FGF1 CT mice after tamoxifen induction. n = 3. (D and E) Western blot and densitometric analysis of FGF1 in livers of LSEC-specific FGF1 KO or FGF1 CT mice. n = 4. (F to I) Eight-week-old LSEC-specific FGF1 KO or FGF1 CT mice were fed HFD for 12 weeks. (F) H&E, ORO, F4/80, and αSMA immunostaining in liver sections. Scale bar, 200 μm. n = 4. (G) MASLD severity, ORO, F4/80, and αSMA immunostaining were quantified by NAS and ORO, F4/80, and αSMA staining areas. n = 4. (H) Hepatic TC and TG. n = 4. (I) Serum ALT and AST levels. n = 4. (J to M) Eight-week-old wild-type mice were fed HFD for 10 weeks and infused c-Kit^+/− LSECs isolated from 8-week-old chow-fed FGF1flox/flox mice or c-Kit⁺ LSECs isolated from 8-week-old chow-fed Cdh5^creERT FGF1flox/flox mice three times per week for another 2 weeks. n = 6. (J) H&E and ORO staining, F4/80, and αSMA immunostaining in liver sections. Scale bar, 200 μm. (K) MASLD severity and ORO staining was quantified by NAS and ORO staining areas. F4/80 and αSMA immunostaining was quantified by F4/80 and αSMA staining areas. (L) Liver TC and TG concentration. (M) Serum ALT and AST levels. Data are means ± SD. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, by two-sample t test (A, C, E, and G to I) or one-way ANOVA (K to M).

c-Kit regulates FGF-1 through RXRG

As c-Kit is a membrane-bound protein, we posited its regulation of FGF1 expression likely occurs through transcriptional mechanisms rather than direct protein interaction. To identify candidate mediators, we intersected c-Kit⁺ LSEC-enriched transcription factors (TFs) with FGF1 promoter-predicted regulators from JASPAR, Cistrome DB, and AnimalTFBS databases, revealing RXRG as the sole overlapping candidate (Fig. 8A). Given RXR family heterogeneity (RXRA/RXRB/RXRG), we systematically assessed subtype involvement. While c-Kit KO universally reduced all three subtypes (Fig. 8B), only RXRG exhibited both c-Kit⁺ LSEC specificity (Fig. 8C) and MASLD-associated down-regulation (Fig. 8D), corroborated by reduced RXRG protein in HFD-treated LSECs (Fig. 8, E and F). These results indicated that c-Kit may promote FGF1 expression through RXRG. Functional validation showed RXR agonist bexarotene up-regulated FGF1 mRNA dose and time dependently (Fig. 8, G and H). To avoid the interference of other RXR subtypes, RXRG overexpression and inhibition experiments were performed, demonstrating that RXRG activate FGF1 transcription independent of other subtypes (fig. S6, A to E). To demonstrate the direct transcriptional regulatory role of RXRG on FGF1, chromatin immunoprecipitation sequencing (ChIP-seq) analysis (GSE79815), ChIP-qPCR, and dual-luciferase reporter gene assay were performed. ChIP-seq analysis confirmed RXR binding at the FGF1 promoter [−884 base pair (bp) from transcription start site (TSS); Fig. 8I], and ChIP-qPCR further confirmed this (fig. S6F). Dual-luciferase assays demonstrated RXRG-driven transcriptional activation of FGF1 (Fig. 8J). To further demonstrate that c-Kit regulates FGF1 in an RXRG-dependent manner, we pretreated Bend.3 cells with the RXR antagonist UVI 3003, followed by c-Kit overexpression. We found that UVI 3003 abolished the up-regulation of FGF1 without affecting the expression of c-Kit or RXRG (fig. S6, G and H). Meanwhile, in primary LSECs isolated from c-Kit KO mice, we administered the RXR agonist bexarotene. c-Kit deficiency reduced the expression of RXRG and FGF1, whereas bexarotene restored FGF1 expression, indicating that RXRG acts downstream of c-Kit and is indispensable for the regulation of FGF1 expression (fig. S6, I and J). These data collectively established RXRG as the dominant mediator of c-Kit–dependent FGF1 regulation.

Fig. 8. c-Kit and RA improved hepatic steatosis via RXRG-FGF1 axis.

(A) Transcription factors prediction using JASPAR, Cistrome DB, and AnimalTFBS database. (B) qPCR analysis of RXRs in c-Kit KO or c-Kit CT LSECs. n = 3 to 4. (C) qPCR analysis of RXRs in primary c-Kit^+/− LSECs. n = 3. (D) qPCR analysis of RXRs in LSECs isolated from chow- or HFD-fed mice. n = 4. (E and F) Western blot and densitometric analysis of RXRG. n = 3. (G and H) qPCR analysis of FGF1 expression in LSECs treated with the RXR agonist bexarotene or DMSO at different time points and doses. n = 3. (I) ChIP-seq analysis of RXR binding sites on FGF1 gene. (J) Dual-luciferase reporter gene assay analyzing the transcriptional regulatory effect of RXRG on FGF1. (K) qPCR analysis of FGF1 in LSECs treated with DMSO or RA at different time points and doses. n = 3. (L to W) HFD-fed mice were administered RA or vehicle and chow-fed mice with vehicle served as controls. (L and M) Serum ALT, AST, TC, and TG levels. n = 5 to 6. (N) Fasting blood glucose levels. n = 5 to 6. (O) Liver TC and TG concentration. n = 5 to 6. (P to T) H&E, ORO, F4/80, and αSMA immunostaining in liver sections and quantitative analysis. Scale bar, 200 μm. n = 5 to 6. (U to W) qPCR analysis, Western blot, and densitometric analysis of FGF1 in LSECs isolated from RA or vehicle-treated mice. n = 3. (X) RXRG expression in LSECs from CV to PV (18). (Y) RXRG expression in LSECs from CV to PV (19). Data are means ± SD. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, by two-sample t test (B to D, F, U, and V) or one-way ANOVA (G, H, J to O, and Q to T).

RA improves hepatic steatosis by increasing FGF1

Given that RA serves as both the endogenous ligand for RXRs and the bioactive metabolite of vitamin A, we postulated its direct regulatory role in FGF1 expression and MASLD amelioration. First, in vitro RA treatment of LSECs induced dose- and time-dependent FGF1 up-regulation (Fig. 8K). Furthermore, in vivo RA administration in MASLD mice attenuated MASLD phenotypes (Fig. 8, L to T), concomitant with elevated FGF1 mRNA/protein levels in LSECs (Fig. 8, U to W), suggesting that RA counteracted MASLD by increasing endothelial FGF1. Notably, endothelial RXRG exhibited zonated expression mirroring c-Kit and FGF1 spatial patterns (Fig. 8, X and Y). To examine whether dietary vitamin A itself exhibits comparable therapeutic efficacy, we fed 8-week-old C57 mice with 60% kcal HFD with vitamin A (40,000 IU/kg) or without vitamin A for 12 weeks. Consistently, vitamin A supplementation effectively decreased hepatic steatosis (fig. S6, K and L). These findings established RA as a direct activator of the LSEC RXRG-FGF1 axis, mechanistically linking vitamin A metabolism to MASLD resolution.

Dietary vitamin A intake is inversely related to MASLD risk

To assess the association between dietary vitamin A and MASLD, we analyzed the continuous National Health and Nutrition Examination Survey (NHANES) data between 2011 and 2018. Study participants were selected as shown in fig. S7, yielding 11,592 eligible individuals. Baseline characteristics of participants stratified by MASLD risk was shown in table S2 (6473 high-risk MASLD). No difference was found in preformed vitamin A intake between groups, but provitamin A and total dietary vitamin A intake were notably lower in the high-risk group (table S2). Correlation analyses demonstrated inverse associations between provitamin A/total dietary vitamin A intake and MASLD risk (assessed by HSI), whereas preformed vitamin A showed no significant relationship (table S3). Restricted cubic spline models further confirmed reduced MASLD risk with increasing provitamin A and total dietary vitamin A intake (fig. S8, A and C). Notably, preformed vitamin A exhibited a nonsignificant trend toward increased MASLD risk (potentially linked to concomitant red meat/fat consumption (fig. S8B). Subgroup analyses (age, gender, race, and education) preserved the inverse association between total vitamin A intake and MASLD risk (tables S4 to S7). These findings revealed that total dietary vitamin A supplementation may be therapeutic effective in preventing MASLD onset.

DISCUSSION

Given its high incidence, prevalence, and severe clinical outcomes, MASLD imposes a substantial socioeconomic burden. Although the Food and Drug Administration has approved Rezdiffra for the treatment of metabolic dysfunction–associated steatohepatitis (MASH) with moderate to advanced liver fibrosis, no disease-modifying therapies exist for early-stage MASLD. The liver is organized into hexagonal functional units termed hepatic lobules, where portal-to-central blood flow establishes radial gradients of oxygen, nutrients, bacterial products, hormones, morphogens, etc. (25). This spatial adaptation drives metabolic zonation—a partitioning of hepatocyte functions across lobular zones-that optimizes hepatic physiology. However, zonal architecture becomes dysregulated during disease progression, creating pathological feedback loops. For instance, impaired insulin suppression of PP gluconeogenesis in MASLD exacerbates hyperinsulinemia, promoting progression to type 2 diabetes (26). Similarly, WNT/β-catenin signaling is disrupted in many diseases including MASLD (27) and hepatitis B virus and hepatitis C virus infections (28), which further disrupt key metabolic processes. Deciphering zonal reprogramming mechanisms therefore represents a critical frontier for developing targeted therapies.

While hepatocyte zonation has dominated liver spatial biology research, recent advances in single-cell and spatial transcriptomics have revealed zonated heterogeneity in nonparenchymal cells. Gut microbiota-induced Marco⁺ immunosuppressive macrophages exhibit PP enrichment (29), while PC hepatic stellate cells express RSPO3 to regulate lobular patterning and organ size (30, 31). PC LSECs secrete Wnt2/Wnt9b/Rspo3 that maintains metabolic zonation (18). Halpern and Inverso et al. (18, 19) conducted spatial RNA-seq of liver endothelial cells and mapped zonal LSEC gene programs in homeostasis. Our prior work demonstrated that PC c-Kit⁺ LSECs promoted regeneration via Wnt2 while mitigating age-related steatosis and fibrosis (20, 21). However, LSEC zonal adaptations during MASLD remain undefined. Here, we uncovered that MASLD triggered LSEC lipid metabolic reprogramming, with PC c-Kit⁺ LSECs exerting antisteatotic effects through FGF1 signaling—a pathway potentiated by RA.

Consistent with prior reports (13, 32), we observed predominant lipid deposition in the PC zone during early MASLD. Furthermore, aging mice and sedentary mice fed standard chow for 5 months similarly exhibited PC lipid accumulation. All these phenotyping findings were subsequently confirmed by multiomics analyses. In longitudinal studies, we observed progressive exacerbation of liver injury and hepatic inflammation with prolonged HFD feeding, with marked hepatocellular damage and lobular inflammation evident at 24 weeks, consistent with previous reports (33–35). Intriguingly, fat cavitations spread from perivenous (CV) regions to PP areas, while macrophages shifted from a PP-enriched distribution to a widespread pattern across the hepatic lobule, in accordance with the remodeling of lipid-associated macrophages described by Guilliams et al. (36). In support of earlier studies (15, 16), these observations indicate disrupted liver zonation in the late stage of HFD feeding, which further disturbs hepatic homeostasis. However, the sequential changes in both parenchymal and nonparenchymal cell zonation during longitudinal study of HFD feeding remain to be further elucidated. Notably, recent work has shown that hepatocellular carcinoma preferentially develops in the PC zone relying on PC metabolic genes (37), whose expression levels are increased and expanded following MASLD (16). Thus, exploring early changes in liver zonation and their functional contributions to MASLD progression is critical for preventing the transition to MASH and subsequent hepatocarcinogenesis.

Liver zonation is anatomically defined by vascular architecture; endothelial regulation likely underpins metabolic partitioning. Recent studies confirmed LSEC gene expression zonation in murine and human livers (18, 19, 38, 39), complementing our earlier work showing LSEC involvement in hepatic homeostasis (21, 40). However, zonal LSEC functions in MASLD remain unexplored. Building on Inverso et al.’s (19) demonstration of c-Kit expression gradients along the portal-central axis, we verified this pattern through immunofluorescence and bioinformatics. Using c-Kit–based immunomagnetic sorting, we isolated PC (c-Kit⁺) and PP (c-Kit⁻) LSECs. Bulk RNA-seq showed PP LSECs preferentially enriched fatty acid metabolism pathways (binding, biosynthesis, and oxidation) under basal conditions, whereas PC LSECs up-regulated these pathways post-HFD. This zonal metabolic reprogramming suggested LSEC zonation disruption during MASLD, which may disrupt the LSEC zonation to maintain homeostasis. This reminds us of hepatocyte zonation shifts where PC markers expand into PP regions (16). While spatial transcriptomics confirmed hepatocyte-like LSEC zonation alterations, causal relationships between LSEC and hepatocyte metabolic remodeling require further investigation.

Given that LSECs maintain liver homeostasis through paracrine signaling (41, 42), we speculate that PC LSECs may counteract steatosis via zonal protein secretion. The secretome screening of c-Kit⁺ LSECs identified FGF1 as a compartmentalized effector mirroring c-Kit expression. We then verified the antisteatotic effect of FGF1 both in vivo and in vitro and found that rFGF1 could effectively ameliorate lipid deposition mainly through FGFR4, corroborating prior reports (43, 44). Notably, EC-specific FGF1 deficiency abrogated the therapeutic effects mediated by c-Kit⁺ LSECs to a great extent, revealing the indispensable role of FGF1 in c-Kit–mediated beneficial effects in MASLD. Mechanistically, FGF1 alleviated hepatic steatosis primarily by reducing DNL, as shown by decreased mRNA and protein levels of DNL-related genes (Srebp1, FASN, SCD1, PPARγ, and ACC1) and increased phosphorylated ACC1 (p-ACC1)/ACC1 ratio, which might be attributed to increased AMP-activated protein kinase (AMPK) phosphorylation (43). Whether FGF1 decreases hepatic steatosis via other pathways needs multiomics analysis and further validation. As a pleiotropic FGF family member, FGF1 orchestrates systemic energy balance, glucose/lipid metabolism, and insulin sensitivity (45), with additional roles in adipose remodeling (46, 47). While heparan sulfate binding restricts systemic bioavailability, its spatial confinement enables localized metabolic regulation. Notably, nonmitogenic FGF1 variants retain therapeutic efficacy through AMPK activation, circumventing proliferative risks (43, 48). These findings position FGF1 as a multifaceted regulator of metabolic homeostasis with translational potential for MASLD management.

RXRG, a retinoid X receptor (RXR) family member and RA (the active metabolite of vitamin A) nuclear receptor, functions as homodimer/heterodimer complexes to regulate gene transcription (49). RXRs include RXRα, RXRβ, and RXRγ, which are involved in a wide range of biological processes, including embryogenesis and metabolism (50). While all RXR subtypes were down-regulated upon c-Kit deletion, only RXRG expression was higher in c-Kit⁺ LSECs and decreased after MASLD. This spatial-temporal pattern positions RXRG as a pivotal mediator linking c-Kit to FGF1 transcriptional regulation in MASLD pathogenesis. The ability of RXR antagonist UVI 3003 to abolish c-Kit overexpression–induced FGF1 up-regulation and the strengthened expression of FGF1 in c-Kit KO LSECs by RXR agonist bexarotene further confirmed that RXRG was indispensable for c-Kit–mediated FGF1 expression. Notably, synthetic RXR agonists (rexinoids) enhance insulin sensitivity and fatty acid oxidation (51, 52), mirroring our observation that RA up-regulated LSEC-derived FGF1 while alleviating MASLD phenotypes. The RXRG expression in LSECs showed similar spatial expression pattern compared to c-Kit and FGF1. As a membrane-localized protein, how c-Kit regulates the expression of the nuclear TF RXRG remains to be elucidated. Potential mechanisms underlying membrane c-Kit–mediated RXRG expression maintenance in LSECs may involve three aspects. First, c-Kit (a receptor tyrosine kinase) activates downstream signaling cascades [phosphatidylinositol 3-kinase–AKT, mitogen-activated protein kinase (MAPK), Janus kinase–signal transducers and activators of transcription] upon ligand binding [e.g., Stem Cell Factor (SCF)], and these pathways can phosphorylate TFs that bind to the RXRG promoter to regulate its transcription (53–55). Second, c-Kit may stabilize RXRG mRNA at the posttranscriptional level: RTK-triggered MAPK/ERK signaling can phosphorylate RNA binding proteins [e.g., Held Out Wings (HOW)] that regulate target mRNA levels, a mechanism supported by previous findings (56). Third, c-Kit may exert nuclear effects: Its intracellular domain can be cleaved by γ-secretase and translocated to the nucleus to interact with RXRG (57). In addition, similar to RTK family member FGFR4 [which directly binds nuclear receptor ERRγ to promote phosphorylation (58)], c-Kit may also directly bind RXRG to modulate its expression. Whether the above mechanisms are involved in c-Kit–mediated RXRG regulation, or act in a coordinated manner, needs further validation.

It is noteworthy that RA can up-regulate nitric oxide (NO) synthesis, which serves as a critical mediator for LSECs in maintaining hepatic homeostasis (59, 60), and suppress vascular cell adhesion molecule–1 expression up-regulated in MASLD (61). Clinical corroboration emerged from NHANES data showing inverse correlation between dietary vitamin A intake and MASLD severity, aligning with prior reports that RA ameliorates insulin resistance and hepatic steatosis (62, 63), which is consistent with our in vivo findings. Despite our findings demonstrating that RA-induced FGF1 up-regulation in LSECs contributes to the amelioration of MASLD, it is important to acknowledge the broader biological roles of RA that may limit the exclusivity of this mechanism. RA is a pleiotropic molecule with well-characterized functions in systemic lipid metabolism and immune regulation (64, 65), suggesting that RA treatment could exert beneficial effects on MASLD through alternative pathways, including actions on extrahepatic tissues (e.g., adipose tissue and pancreas) or other intrahepatic cell populations (e.g., hepatocytes, Kupffer cells, and hepatic stellate cells). Thus, the FGF1-mediated effects in LSECs identified in our study likely represent only a subset of the multifaceted therapeutic actions of RA in MASLD, and the relative contribution of LSEC-dependent versus LSEC-independent pathways remains to be quantified. Collectively, these findings establish vitamin A and its derivatives as multimodal therapeutic agents targeting both metabolic dysfunction and endothelial homeostasis in early MASLD.

This study provides the first functional characterization of liver endothelial zonation in MASLD pathogenesis, with a focus on LSEC heterogeneity during disease initiation. Our findings suggest that PC lipid deposition constitutes a critical initiating event for MASLD progression. Furthermore, we identified a protective PC endothelial RA-FGF1 signaling axis, which represents a mechanistically grounded therapeutic target to prevent MASLD development at its earliest stages.

MATERIALS AND METHODS

Materials and antibodies used in this study are listed in Table 1.

Table 1. Materials used in this study.

D-PBS, Dulbecco’s phosphate-buffered saline; HBSS, Hanks’ balanced salt solution; DMEM, Dulbecco’s modified Eagle’s medium; DNase, deoxyribonuclease; MACS, magnetic-activated cell sorting; ECM, endothelial cell medium; GAPDH, glyceraldehyde-3-phosphate dehydrogenase; FBS, fetal bovine serum; DMSO, dimethyl sulfoxide; DAPI, 4′,6-diamidino-2-phenylindole; N/A, not applicable.

Reagent or resource	Source	Identifier
Antibodies
Anti-GS antibody	Proteintech	Catalog no. 11037-2-AP
Anti-c-Kit antibody	R&D Systems	Catalog no. AF1356
Anti-F4/80 antibody	Invitrogen	Catalog no. 14-4801-82
Anti-CD11b antibody	Servicebio	Catalog no. GB11058
Anti-αSMA antibody	Abcam	Catalog no. ab124964
Anti-FGF1 antibody	Proteintech	Catalog no. 17400-1-AP
Anti-RXRG antibody	ABclonal	Catalog no. A1877
Anti-RXRG antibody	Santa Cruz Biotechnology	Catalog no. sc-514134
Anti-Srebp1 antibody	Proteintech	Catalog no. 14088-1-AP
Anti-FASN antibody	Proteintech	Catalog no. 10624-2-AP
Anti-SCD1 antibody	Proteintech	Catalog no. 28678-1-AP
Anti-ACC1 antibody	Proteintech	Catalog no. 21923-1-AP
Anti–phospho-ACC1 (Ser79) antibody	Proteintech	Catalog no. 29119-1-AP
Anti-PPARγ antibody	Proteintech	Catalog no. 16643-1-AP
Anti-vinculin antibody	Proteintech	Catalog no. 26520-1-AP
Anti-GAPDH antibody	Proteintech	Catalog no. 10494-1-AP
Goat anti-rabbit secondary antibody	Abbkine	Catalog no. A21020
Rabbit anti-goat secondary antibody	Abbkine	Catalog no. A21030
Donkey anti-rat secondary antibody	Invitrogen	Catalog no. A48269
Donkey anti-goat secondary antibody	Invitrogen	Catalog no. A32814
Donkey anti-rabbit secondary antibody	Invitrogen	Catalog no. A32754
Anti-CD146 MicroBeads	Miltenyi Biotec	Catalog no. 130-092-007
Anti-CD117 MicroBeads	Miltenyi Biotec	Catalog no. 130-091-224
Chemicals, peptides, and recombinant proteins
Tamoxifen	MilliporeSigma	Catalog no. T5648
Corn Oil	MilliporeSigma	Catalog no. 8001-30-7
D-PBS	Gibco	Catalog no. 14190144
DMSO	MilliporeSigma	Catalog no. D2438
Oil Red O	Servicebio	Catalog no. G1015
BODIPY505/515	MCE	Catalog no. HY-D1237
ECM	ScienCell	Catalog no. 1001
HBSS	Gibco	Catalog no. 24020117
DMEM	Gibco	Catalog no. C11995500BT
Collagenase, type IV	MilliporeSigma	Catalog no. C5138
DNAse I	Roche	Catalog no. 10-104-159-001
MACS buffer	Miltenyi Biotec	Catalog no. 130-091-221
OptiPrep	Axis Shield	Catalog no. 11145422
TRIzol	Invitrogen	Catalog no. 15596026
FBS	Gibco	Catalog no. A5670701
Recombinant mouse FGF1 protein	MCE	Catalog no. HY-P7065
DAPI	Servicebio	Catalog no. G1012
Sodium palmitate+ sodium oleate	Kunchuang	Catalog no. KC006
Bexarotene	MCE	Catalog no. HY-14171
UVI 3003	Selleck	Catalog no. S2833
Retinoic acid	MCE	Catalog no. HY-14649
Infigratinib	MCE	Catalog no. HY-13311
PD-166866	MCE	Catalog no. HY-101296
FGFR2-IN-1	MCE	Catalog no. HY-145230
VSPPLTLGQLLS	MCE	Catalog no. HY-P3695
FGFR4-IN-1	MCE	Catalog no. HY-100631
Lipofectamine 2000	Invitrogen	Catalog no. 11668019
Critical commercial assays
Reverse transcription reagent	Accurate Biology	Catalog no. AG11706
SYBR Green PCR Master Mix	Accurate Biology	Catalog no. AG11701
Rabbit IHC kit	ZSGB-Bio	Catalog no. PV-9001
Triglyceride Assay Kit	ApplyGen	Catalog no. E1025
Total Cholesterol Assay Kit	ApplyGen	Catalog no. E1026
ALT Assay Kit	Nanjing Jiancheng	Catalog no. C009-3-2
AST Assay Kit	Nanjing Jiancheng	Catalog no. C010-3-2
ALB Assay Kit	Nanjing Jiancheng	Catalog no. A028-2-1
TBIL Assay Kit	Nanjing Jiancheng	Catalog no. C019-1-1
DBIL Assay Kit	Nanjing Jiancheng	Catalog no. C019-2-1
Total Cholesterol Assay Kit	Nanjing Jiancheng	Catalog no. A111-1-1
Triglyceride Assay Kit	Nanjing Jiancheng	Catalog no. A110-1-1
Mouse FGF1 ELISA Kit	JONLNBIO	Catalog no. JL51114
Mouse TNFα ELISA Kit	JONLNBIO	Catalog no. JL10484
Mouse IL-1β ELISA Kit	JONLNBIO	Catalog no. JL18442
Mouse IL-6 ELISA Kit	JONLNBIO	Catalog no. JL20268
BCA Protein Assay Kit	Thermo Fisher Scientific	Catalog no. 23227
LIPO8000 Transfection Reagent	Beyotime	Catalog no. C0533
SimpleChIP Enzymatic Chromatin IP Kit	CST	Catalog no. 9003
Dual Luciferase Reporter Gene Assay Kit	Beyotime	Catalog no. RG027
Experimental models: Organisms/strains
Cdh5CreERT	Generated in our laboratory	N/A
c-Kit-flox	Generated in our laboratory	N/A
FGF1-flox	Shanghai Model Organisms Center	N/A
Oligonucleotides
Tnfα qPCR: Forward: 5′-CAGGCGGTGCCTATGTCTC-3′; reverse: 5′-CGATCACCCCGAAGTTCAGTAG-3′	This paper	N/A
Il-1β qPCR: Forward: 5′- CCGTGGACCTTCCAGGATGA-3′; reverse: 5′- GGGAACGTCACACACCAGCA-3′	This paper	N/A
Il-6 qPCR: Forward: 5′- TAGTCCTTCCTACCCCAATTTCC-3′; reverse: 5′- TTGGTCCTTAGCCACTCCTTC-3′	This paper	N/A
Ccl2 qPCR: Forward: 5′- TACAAGAGGATCACCAGCAGC-3′; reverse: 5′- ACCTTAGGGCAGATGCAGTT-3′	This paper	N/A
Cxcl10 qPCR: Forward: 5′- ATGACGGGCCAGTGAGAATG-3′; reverse: 5′- ATGATCTCAACACGTGGGCA-3′	This paper	N/A
c-Kit qPCR: Forward: 5′- GCCTGACGTGCATTGATCC-3′; reverse: 5′-AGTGGCCTCGGCTTTTTCC-3′	This paper	N/A
Fgf1 qPCR: Forward: 5′- CCCTGACCGAGAGGTTCAAC-3′; reverse: 5′- GTCCCTTGTCCCATCCACG-3′	This paper	N/A
Rxra qPCR: Forward: 5′- ATGGACACCAAACATTTCCTGC-3′; reverse: 5′- CCAGTGGAGAGCCGATTCC-3′	This paper	N/A
Rxrb qPCR: Forward: 5′- CCACCTCTTACCCCTTCAGC-3′; reverse: 5′- TGGAAGAACTGATGACTGGGA-3′	This paper	N/A
Rxrg qPCR: Forward: 5′- AGGGAAGCACTACGGTGTGTA-3′; reverse: 5′- TCCCGACAGGTGTAGATGAGA-3′	This paper	N/A
Fgfr1 qPCR: Forward: 5′- ACTCTGCGCTGGTTGAAAAAT-3′; reverse: 5′- GGTGGCATAGCGAACCTTGTA-3′	This paper	N/A
Fgfr2 qPCR: Forward: 5′- GCCTCTCGAACAGTATTCTCCT-3′; reverse: 5′- ACAGGGTTCATAAGGCATGGG-3′	This paper	N/A
Fgfr3 qPCR: Forward: 5′- GCCTGCGTGCTAGTGTTCT-3′; reverse: 5′- TACCATCCTTAGCCCAGACCG-3′	This paper	N/A
Fgfr4 qPCR: Forward: 5′- TCCATGACCGTCGTACACAAT-3′; reverse: 5′- ATTTGACAGTATTCCCGGCAG-3′	This paper	N/A
Srebp1c qPCR: Forward: 5′- GGAGCCATGGATTGCACATT-3′; reverse: 5′- GGCCCGGGAAGTCACTGT-3′	This paper	N/A
Fasn qPCR: Forward: 5′- CTGCGGAAACTTCAGGAAATG-3′; reverse: 5′- GGTTCGGAATGCTATCCAGG-3′	This paper	N/A
Scd1 qPCR: Forward: 5′- TCTTCCTTATCATTGCCAACACCA-3′; reverse: 5′- GCGTTGAGCACCAGAGTGTATCG-3′	This paper	N/A
Acc1 qPCR: Forward: 5′- GGCCAGTGCTATGCTGAGAT-3′; reverse: 5′- AGGGTCAAGTGCTGCTCCA-3′	This paper	N/A
Pparγ qPCR: Forward: 5′- ATTCTGGCCCACCAACTTCGG-3′; reverse: 5′- TGGAAGCCTGATGCTTTATCCCCA-3′	This paper	N/A
Cd36 qPCR: Forward: 5′- GACTGGGACCATTGGTGATGA-3′; reverse: 5′- AAGGCCATCTCTACCATGCC-3′	This paper	N/A
Pparα qPCR: Forward: 5′- TATTCGGCTGAAGCTGGTGTAC-3′; reverse: 5′- CTGGCATTTGTTCCGGTTCT-3′	This paper	N/A
Cpt1α qPCR: Forward: 5′- AGGACCCTGAGGCATCTATT-3′; reverse: 5′- ATGACCTCCTGGCATTCTCC-3′	This paper	N/A
Cpt1α qPCR: Forward: 5′- AGGACCCTGAGGCATCTATT-3′; reverse: 5′- ATGACCTCCTGGCATTCTCC-3′	This paper	N/A
ApoB qPCR: Forward: 5′- AAGCACCTCCGAAAGTACGTG-3′; reverse: 5′- CTCCAGCTCTACCTTACAGTTGA-3′	This paper	N/A
Mttp qPCR: Forward: 5′- ATACAAGCTCACGTACTCCACT-3′; reverse: 5′- TCCACAGTAACACAACGTCCA-3′	This paper	N/A
Software and algorithms
GraphPad Prism 10	Purchased by personal	www.graphpad.com/
Image-Pro Plus 6.0	Purchased by Xijing Hospital	https://mediacy.com/
Image Lab	Purchased by Xijing Hospital	www.bio-rad.com/
Other
Chow diet	Medicience	Catalog no. MD17121
High-fat diet	Research Diets	Catalog no. D12492

Clinical samples

Liver tissues were obtained from patients with biopsy-proven MASLD who underwent a liver biopsy during bariatric surgery at Xijing Hospital and non-MASLD (control) patients who were diagnosed with liver hemangioma and accepted liver surgeries. All procedures were approved by the Human Ethics Committees of Xijing Hospital (KY20253406-1). All participants provided informed written consent. Detailed donor information was listed in table S1.

Animal studies

Male C57BL/6J mice (8 weeks old) were obtained from Beijing Weitong Lihua Experimental Animal Technology Co. Ltd. Cdh5^CreERT c-Kit^flox/flox (c-Kit KO) mice were generated as previously described (21). Cdh5^CreERT FGF1^flox/flox (FGF1 KO) mice were generated by crossing Cdh5^CreERT mice with FGF1^flox/− mice purchased from Shanghai Model Organisms Center Inc. Offspring genotypes were confirmed by PCR and agarose gel electrophoresis using tail-derived DNA. To induce Cre recombinase activity, 6-week-old male mice received intraperitoneal tamoxifen (100 mg/kg) for five consecutive days. Tamoxifen-injected littermates served as vehicle controls to exclude off-target effects. Mice were analyzed at least 1 week after tamoxifen injections.

Aging male mice were established using 24-month-old mice (purchased from Beijing Weitong Lihua Experimental Animal Technology Co.), and 8-week-old male mice were used as young control. For MASLD modeling, 8-week-old male mice were fed an HFD for 8 to 24 weeks to establish progressive disease stages.

For rFGF1 in vivo experiments, 8-week-old wide-type C57BL/6J mice were fed HFD for 9 weeks, followed by intraperitoneal administration of rFGF1 (0.5 mg/kg) every other day for 3 weeks. The mice were euthanized 24 hours after the last treatment. c-Kit KO mice receiving rFGF1 followed identical protocols. Six-week-old FGF1 KO mice were administrated tamoxifen induction and were fed HFD for 12 weeks to examine MASLD phenotypes. For RA treatment, 8-week-old wide-type C57BL/6J mice were fed HFD for 10 weeks and then were administrated RA (10 mg/kg/day) intraperitoneally for 2 weeks. The mice were euthanized 24 hours after the last treatment. For vitamin A intervention experiments, 8-week-old wide-type C57BL/6J mice were fed 60% kcal HFD with vitamin A (40,000 IU/kg) or without vitamin A for 12 weeks.

All experiments were replicated biologically at least three times. All animals were housed in specific pathogen–free facilities with ad libitum access to food/water under 12-hour light/12-hour dark cycles. All animal experiments were performed in accordance with the guide for the Care and Use of Laboratory Animals prepared by the National Academy of Science and were approved by the Experiment Administration Committee of Airforce Medical University (no. 251437) (Xi’an, China).

Primary hepatocytes/LSECs isolation

Mouse primary hepatocytes/LSECs/macrophages (Macs)/hepatic satellite cells (HSCs) were isolated as previously described (21, 40). Specifically, mice were anesthetized with 1% sodium pentobarbital. Then, a laparotomy was performed to expose the liver, inferior vena cava, and PV. Preprepared perfusion solutions A and B [containing type IV collagenase (0.5 mg/ml); Sigma-Aldrich) were sequentially perfused through the PV for liver perfusion. The liver was then excised and placed in a digestion solution containing collagenase and deoxyribonuclease I (100 μg/ml; Roche) for 20 min. The resulting liver suspension was filtered through a 100-μm mesh filter and centrifuged at 4°C and 50g for 3 min to obtain hepatocytes. The supernatant was subsequently centrifuged at 350g for 6 min to isolate nonparenchymal cells. Using a 60% OptiPrep buffer (Axis Shield), density gradient solutions of 17.6 and 11% were prepared. Centrifugation was performed at 4°C and 1400g for 17 min, yielding HSCs in the upper layer, while the middle layer primarily contained LSECs and macrophages. The middle layer cells were collected and sequentially incubated with CD146 and F4/80 magnetic beads (Miltenyi Biotec) to isolate LSECs and Macs following the manufacturer’s instructions, respectively.

Cell culture and in vitro experiments

The AML12, Bend.3, and human embryonic kidney (HEK) 293 T cell lines were purchased from the Cell Bank of Shanghai Academy of Chinese Sciences. AML12 cells were maintained in AML12-specific medium (purchased from the same source), while Bend.3 and HEK 293T cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. All experiments were replicated technically at least three times.

For rFGF1 treatment, AML12 cells were seeded in 6-/12-well plates. Upon reaching 50% confluence, cells were exposed to PAOA (0.15 mM PA and 0.3 mM OA) for 12 hours to induce lipotoxicity, followed by rFGF1 (500 ng/ml) treatment for an additional 24 hours before harvest. For FGFR inhibition studies, AML12 cells pretreated with PAOA for 11 hours were incubated with FGFR inhibitors (infigratinib/PD-166866/FGFR2-IN-1/VSPPLTLGQLLS/FGFR4-IN-1) for 1 hour before rFGF1 administration (500 ng/ml, 24 hours).

For RXR agonist experiments, LSECs seeded in six-well plates were subjected to medium replacement after 24 hours. Time-dependent effects of bexarotene (25 μM) were assessed at 12, 24, and 48 hours, while dose-dependent responses were tested using 10 and 50 μM bexarotene for 24 hours. Parallel experiments with RA used identical temporal (25 μM at 12/24/48 hours) and dose-dependent (10/50 μM for 24 hours) regimens. Primary LSECs were isolated from c-Kit CT and c-Kit KO mice and seeded in six-well plates. The cells were treated with dimethyl sulfoxide (DMSO) or 25 μM bexarotene for 48 hours and harvested for further analysis.

For RXR antagonist experiments, Bend.3 cells seeded in six-well plates were pretreated with RXR antagonist UVI 3003 (10 μM) for 2 hours and then transfected with 2.5 μg of c-Kit/control plasmids constructed by PAIVIBIO (Wuhan, China) using LIPO8000 Transfection Reagent (Beyotime, Shanghai, China) according to the manufacturer’s instructions for 48 hours. Then, these cells were harvested for further analysis.

Rxrg plasmid transfection

For Rxrg plasmid transfection experiment, pCDH-Rxrg and pCDH-neo plasmids were constructed by PAIVIBIO (Wuhan, China). Bend.3 cells were seeded in six-well plates, and when they reached 70% confluence, the culture medium was replaced with fresh medium and plasmids and LIPO8000 Transfection Reagent (Beyotime, Shanghai, China) were added into medium according to the manufacturer’s instructions for 48 hours. Then, cells were collected for further analysis.

Rxrg siRNA transfection

For Rxrg siRNA transfection experiment, Rxrg siRNA (sense: ggc ucu ccg uau aga guc a/dt//dt/; antisense: uga cuc uau acg gag agc c/dt//dt/) and siNC were constructed by Sangon Biotech (Shanghai, China). Bend.3 cells were seeded in six-well plates and when they reached 70% confluence, the culture medium was replaced with fresh DMEM medium and siRNA and LIPO8000 Transfection Reagent (Beyotime, Shanghai, China) were added into medium according to the manufacturer’s instructions for 6 hours. Then, the culture medium was replaced with fresh complete medium for another 48 hours. Then, the cells were collected for further analysis.

Histology

Paraffin and frozen sections were prepared following previously described protocols. H&E, ORO, immunohistochemistry, and IF staining were performed as previously described (40). BODIPY staining was performed following the manufacturer’s instructions.

RT-qPCR

Cell/tissue total RNA was extracted using TRIzol reagent and reverse transcribed into cDNA using the reverse transcription reagent. RT-qPCR was performed using SYBR Green PCR Master Mix. Expression data were normalized to mRNA expression of the β-actin protein. The primer sequences are listed in Table 1.

Western blots

Total protein was extracted using radio immunoprecipitation assay lysis buffer containing protease inhibitors and quantified using the BCA Protein Assay Kit. Equal amounts of protein were separated by SDS–polyacrylamide gel electrophoresis, transferred onto polyvinylidene fluoride membranes, blocked with 5% skimmed milk powder, and then incubated with diluted primary antibodies and secondary antibodies. The protein signals were detected using the ChemiDoc MP Imaging System (Bio-Rad).

Biochemical analysis

Serum levels of ALT, AST, ALB, TBIL, DBIL, TC, and TG were examined using commercial kits following the manufacturer’s instructions on an automatic biochemistry analyzer (Chemray 240, Rayto).

Cell/tissue total cholesterol/triglyceride assay

Cells/tissue total cholesterol/triglyceride contents were examined following the manufacturer’s instructions. Concentrations were normalized to protein contents.

ChIP analysis

ChIP was performed using the SimpleChIP Enzymatic Chromatin IP Kit (magnetic beads) (Cell Signaling Technology) following the manufacturer’s instructions. Briefly, primary LSECs were isolated from 8-week-old wild-type C57BL/6J mice, and cells were then cross-linked by adding 1% (v/v) formaldehyde at room temperature for 10 min. Following the termination of the reaction with glycine (5 min at room temperature), the cells were subjected to two washes using ice-cold phosphate-buffered saline (PBS). Following the digestion of cross-linked chromatin with micrococcal nuclease (20 min, 37°C), 500 μl of aliquots was sonicated on ice. DNA fragments were incubated with anti-RXRG antibody (Santa Cruz Biotechnology), mouse anti-histone H3 (positive control), and normal mouse immunoglobulin G at 4°C overnight, specifically. Then, protein G magnetic beads were added, following incubation for another 2 hours at 4°C. Washed sequentially with low- and high-salt ChIP buffer, protein G magnetic beads were reverse cross-linked, and DNA was eluted. Then, the precipitated DNA fragments were extracted and analyzed by qPCR. The primer sequences used for ChIP-PCR were designed according to the predicted binding sites: Site-1(forward: TGTTTGTTTTTGGAGAACCTGGA; reverse: CTTGTCCAACACTGCCCTCT), site-2 (forward: TGACTGCTACCTCTGGCTCA; reverse: GCAAACACTAGAGAAAGCTGGC), and Site-3 (forward: AGGACAAGGGACACCAGAGA; reverse: TCAACCTTCAGTTTCATTCCCA).

Luciferase reporter assay

The pCDH-Rxrg expression vector was constructed by PAIVIBIO (Wuhan, China). The Fgf1 promoter-containing firefly luciferase reporter vector (GV534-Fgf1) was constructed by GeneChem (Shanghai, China). HEK 293T cells were seeded in 96-well plates, cultured overnight, and then cotransfected with pCDH-neo or different concentrations of the pCDH-Rxrg and the CV308 (control) or GV534-Fgf1 plasmid using Lipofectamine 2000 according to the manufacturer’s protocol. A Renilla luciferase plasmid was used as a normalization control. Reporter gene (luciferase) activity was measured using the Dual Luciferase Reporter Gene Assay Kit (Beyotime).

Fasting blood glucose assay

After fasting for 8 hours, fasting blood glucose was tested by One Touch Ultra glucometer and test strips (LifeScan) using tail vein blood.

ELISA

FGF1 protein concentrations in cell culturing medium were examined using mouse FGF1 ELISA Kit following the manufacturer’s instructions. Hepatic TNFα, IL-1β, and IL-6 levels were examined using commercial ELISA Kit following the manufacturer’s instructions.

c-Kit⁺ LSEC infusion

c-Kit⁺ cells were isolated from purified LSECs using anti-CD117 magnetic beads. The resting LSECs were defined as c-Kit⁻ LSECs. c-Kit^+/− LSECs were infused into MASLD mice fed with HFD for 10 weeks through tail vein three times a week for 2 weeks. The mice were euthanized 24 hours after the last treatment.

snRNA sequencing

Hepatic nuclei were isolated from flash-frozen liver chunks from healthy mice. Approximately 500 mg of tissue was cut and weighed on dry ice using sterile disposable scalpels. The liver tissue was homogenized in ice-cold homogenization buffer (0.25 M sucrose, 5 mM CaCl₂, 3 mM MgAc₂, 10 mM tris-HCl (pH 8.0), 0.1 mM EDTA, 1x protease inhibitor, and RiboLock ribonuclease (RNase) inhibitors (1 U/μl) with a glass-on-glass dounce homogenizer, 10 strokes with the A pestle, followed by 10 strokes of the B pestle. Homogenates were passed through a 70-μm cell strainer to collect the nuclear fraction. The nuclear fraction was mixed with an equal volume of 50% iodixanol and added on top of a 30 to 33% iodixanol solution for 20 min at 10,000g at 4°C. After the removal of the myelin layer from the top of the gradient, the nuclei were collected from the 30 to 33% iodixanol interface. The nuclei were resuspended in nuclei wash and resuspension buffer [0.04% bovine serum albumin, RiboLock RNase inhibitors (0.2 U/μl), 500 mM mannitol, and 0.1 mM phenylmethylsulfonyl fluoride protease inhibitor in PBS) and pelleted for 5 min at 500g and 4°C. The nuclei were passed through a 40-μm cell strainer to remove cell debris and large clumps. Nuclei concentration was manually determined using trypan blue counter staining and hemocytometer. Nuclei concentration was adjusted to 700 to 1200 nuclei/μl and followed immediately by the 10X Genomics Single Cell Protocol. The libraries were constructed and sequenced on the Illumina sequencing platform by Genedenovo Biotechnology Co. Ltd. (Guangzhou, China).

scRNA sequencing

Liver tissues were obtained from 8-week-old male mice fed with chow/HFD for 12 weeks. The scRNA-seq analysis was carried out by the IlluminaNovaSeq 6000 (performed by Biomarker Technologies Corporation, Beijing, China), and bioinformatics analysis was performed by Biomarker Technologies Corporation (Beijing, China).

Spatial transcriptomics sequencing

The spatial transcriptomics sequencing method was performed as previously described (66) with minor adjustments. Briefly, liver tissues resected from 8-week-old male mice fed with chow/HFD for 12 weeks were immediately dissected, washed with 1× PBS, and snap-frozen in liquid nitrogen-chilled isopentane. The tissues were embedded in the Optimal Cutting Temperature (OCT) compound placed on the dry ice and stored at −80°C. Sections were cut at 10-mm thickness and placed onto the chilled Tissue Optimization Slides and Gene Expression Slides (BMKMANU S1000). The Gene Expression Slide has one to eight identical capture areas (6.8 mm by 6.8 mm), each with 2,000,000 spots containing barcoded primers. The primers are attached to the slide by the 5′ end and contain a cleavage site, a T7 promoter region, a partial read 1 Illumina handle, a spot-unique spatial barcode, a unique molecular identifier, and poly(dT)VN. The spots have a diameter of 2.5 mm and are arranged in a centered regular hexagonal grid so that each spot has six surrounding spots with a center-to-center distance of 4.8 mm. Tissue optimization, fixation, H&E staining, and imaging were conducted according to the manufacturer’s protocol. Nuclei staining was also performed and imaged for subsequent cell segmentation. After reverse transcription and spatial library construction, the library was sent for quality control and sequenced on the Illumina NovaSeq 6000 platform to generate 150 bp paired-end reads.

Spatial transcriptomics analysis

Raw sequencing files were processed with BSTMatrix (v2.3j) to perform cell segmentation and obtain spatial gene expression matrices. The spatial expression matrix was processed using Seurat v4.0.5. Cells with more than 200 spatial features and genes expressed in at least five cells were retained. Cell clusters and their differentially expressed genes were obtained using BSTViewer software (v4.9). GSEA analysis was performed among these clusters.

Bulk RNA sequencing

c-Kit^+/− LSECs were isolated from livers of male mice fed with chow/HFD for 12 weeks, and total RNA were extracted using TRIzol reagent. The extracted mRNA is enriched using mRNA Capture Beads. After purification with beads, the mRNA is fragmented using high temperatures. The fragmented mRNA is then used as a template to synthesize the first strand of cDNA in a reverse transcription enzyme mixture system. While synthesizing the second strand of cDNA, end repair and A-tailing are completed. Next, adapters are ligated, and Hieff NGS DNA Selection Beads are used for purification to select target fragments. PCR library amplification is then performed, and last, detection is carried out using the Illumina NovaSeq 6000 System.

Bioinformatics analysis

Bioinformatic analysis of bulk RNA sequencing was conducted on Omicsmart platform (www.omicsmart.com/). The raw data of Figs. 2 (B and F), 4F, and 8X were obtained from Halpern et al. (18). The raw data of Figs. 2C, 4G, and 8Y were obtained from Inverso et al. (19). The raw data of spatial transcriptomics analysis in Fig. 2 (G and H) was deposited in GSE248077. The raw data of ChIP-seq in Fig. 8I was deposited in GSE79815 (GSM2104118 and GSM2104119). The analysis of scRNA-seq and snRNA-seq data generated by our laboratory was performed using R software (R4.4.1; www.r-project.org/). The R packages including Seurat (5.3.0), dplyr (1.1.4), future (1.33.2), ggplot2 (3.5.2), and DoubletFinder (2.0.4) were obtained from the Comprehensive R Archive Network (https://cran.r-project.org/) and Bioconductor (https://bioconductor.org/), respectively.

NHANES database analysis

Study design

This study used data from the National Health and Nutrition Examination Survey (NHANES), a continuous cross-sectional survey designed to assess health and nutritional status in the US population since 1999. The survey gathers demographic, socioeconomic, dietary, and health-related information using a complex, multistage probability sampling approach from the noninstitutionalized civilian population of the US. The examinations consisted of medical, dental, and physiological measurements, as well as laboratory tests conducted by highly trained medical personnel. The study protocol received approval from the National Center for Health Statistics Research Ethics Review Board, with written informed consent obtained from all participants at enrollment. We analyzed data from adults aged ≥20 years across four NHANES cycles (2011–2018). From 39,156 participants, we excluded 16,539 individuals aged <20 years. Subsequent exclusions comprised participants with positive hepatitis B/C markers (surface antigen, antibody, or RNA) or excessive alcohol consumption [≥10 g/day (women); ≥20 g/day (men)] and those lacking dietary vitamin A data, yielding 11,592 subjects for final analysis (fig. S7).

MASLD was defined using the HSI, which was previously validated and excluded those with other causes of chronic liver disease and exposure to steatogenic medication. HSI was calculated on the basis of ALT/AST ratio, body mass index (BMI), gender, and type 2 diabetes mellitus (DM) as follows

$$\mathit{HSI}=8\times \mathit{ALT}/\mathit{AST}+\mathit{BMI} (+2,\text{if} \text{DM};+2,\text{if female})$$

Dietary vitamin A intake

Daily dietary intake was assessed through 24-hour recalls and 30-day supplement records. Information about type, consumption frequency, duration, and amount taken over the past 30 days was collected for each reported dietary supplement. Total vitamin A intake (retinol activity equivalents, μg) was calculated by combining dietary and supplemental sources. Different sources of dietary vitamin A, such as preformed vitamin A (milk and milk products, meat, poultry, fish, eggs) and provitamin A carotenoids (legumes, nuts and seeds, grain products, fruits, and vegetables) were identified using the food codes. The 24-hour recall data were averaged, energy-adjusted (μg/1000 kcal/day), and then stratified into quartiles. As for preformed vitamin A, 1 μg of retinol activity equivalents (RAEs) were equal to 1 μg of all-trans retinol from animal foods. Provitamin A carotenoids were estimated by using the diet equation: 1 RAEs (μg) = 1/12 β-carotene (μg) + 1/24 other provitamin A (μg).

Covariates

Potential confounding factors were adjusted in multivariate models, including age, gender (male and female), race (non-Hispanic white and other races), education level (college degree or other), smoking status (smoking at least 100 cigarettes in life or not), moderate activity (yes or no), and marital status (married or other).

Statistical analysis

All analyses in this study incorporate sample weights, strata, and primary sampling units. The weight was calculated based on the dietary day 1 sample weight divided by the number of cycles in the main analysis. Categorical variables were expressed as unweighted frequencies (weighted percentages), while continuous variables were reported as medians with interquartile ranges (IQRs). To compare the difference between groups, the Chi-square test was used for categorical variables, while the Student’s t test or Mann-Whitney U test was used for continuous variables with normal and non-normal distribution. Logistic regression analyses assessed associations between dietary retinol intake (quartilized with Q1 as reference) and MASLD risk through three sequential models: model 1 (unadjusted), model 2 (age, sex, and race-adjusted), and model 3 (additional adjustment for smoking status, educational level, and material statues).

Logistic regression analyses evaluated associations between total dietary vitamin A intake and MASLD risk, generating odds ratios with 95% confidence intervals. Furthermore, dose-response relationships between total dietary vitamin A intake and MASLD risk were assessed using restricted cubic spline regressions and three knots were located at the 5th, 50th, and 95th percentiles. The nonlinearity P value was computed by examining the values of the quadratic zero spline coefficient. After stratification by age (<65 and ≥65 years), gender (male and female), race (non-Hispanic white and other races), and education level (college degree or other), logistic regression analyses were conducted to analyze the relationship between dietary vitamin A intake and MASLD risk. All analyses were implemented in R 4.4.2, with statistical significance defined as two-tailed P < 0.05.

Statistics

Statistical analysis of data was performed using GraphPad Prism v10 and the results were expressed as the means ± SD. Data were tested for normality (Kolmogorov-Smirnov test) and homogeneity (Brown-Forsythe test) of variance. Difference between the two groups was analyzed by Student’s t test with normal distribution or by Mann-Whitney U test with skewed distribution. Difference among multiple groups was analyzed by one-way analysis of variance (ANOVA) with normal distribution followed by Tukey’s post hoc test or by Kruskal-Wallis test with skewed distribution. P values less than 0.05 were considered statistically significant. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001; ns, not significant.

Supplementary Materials

This PDF file includes:

Figs. S1 to S8

Tables S1 to S7