Targeted inhibition of hepatic de novo ceramide synthesis ameliorates MASH

Abstract

Increasing evidence implicates ceramides in the pathogenesis of metabolic dysfunction-associated steatohepatitis (MASH). However, the therapeutic potential of liver-targeted ceramide lowering remains unclear. In this study, we demonstrate that elevated ceramide levels in MASH patients and mouse models are closely associated with the activation of hepatic de novo ceramide synthesis. The analysis of human hepatic single-nucleus RNA sequencing (snRNA-seq) data revealed predominant up-regulation of SPTLC2, which encodes a subunit of the rate-limiting enzyme in the de novo ceramide synthesis pathway, in hepatocytes. By targeted inhibition of SPTLC2 with lipid nanoparticle–mediated siRNA delivery to hepatocytes, we reduced both hepatic and circulating ceramide levels. This intervention suppressed hepatic lipid uptake and lipogenesis, thereby alleviating MASH progression. Therapeutic efficacy was demonstrated in an 8-week methionine-choline–deficient diet-induced MASH model and validated in a 1-year choline-deficient high-fat diet–induced MASH model. Our findings highlight hepatocyte Sptlc2 as a promising therapeutic target for MASH.

Hepatocyte-targeted siRNA silencing of ceramide production ameliorates MASH in young and aged mice.

INTRODUCTION

Metabolic dysfunction-associated steatotic liver disease [MASLD; formerly known as nonalcoholic fatty liver disease (NAFLD)] has become one of the most prevalent chronic liver diseases worldwide, with an estimated global prevalence of 25% and ~40% in Singapore (1, 2). MASLD encompasses a disease spectrum from benign liver steatosis to progressive metabolic dysfunction-associated steatohepatitis (MASH) that is characterized by severe steatosis with lobular inflammation, hepatocellular ballooning, and varying degrees of fibrosis (3). While progression to advanced fibrosis, cirrhosis, or hepatocellular carcinoma may lead to liver failure and even death, early-stage MASH with moderate inflammation remains a reversible condition (4). Despite substantial advances in our understanding of MASH pathogenesis, the therapeutic development has lagged. Most investigational drug candidates in clinical trials have shown limited efficacy or intolerable side effects. In March 2024, the US Food and Drug Administration approved Rezdiffra (resmetirom), the first therapy for noncirrhotic MASH with moderate to advanced liver fibrosis. Although this approval represents a milestone, important clinical challenges remain. Notably, Rezdiffra elicits a therapeutic response in only ~30% of patients (5). Moreover, many other drugs in development, such as glucagon-like peptide-1 receptor agonists, primarily target the systemic metabolic milieu rather than the liver per se, raising concerns about potential off-target effects. Thus, liver-targeted therapies with improved efficacy and safety are urgently needed.

The molecular mechanisms governing the transition from steatosis to MASH remain poorly understood. Notably, the hepatic lipidome is profoundly altered in MASH, and ceramides, a class of bioactive sphingolipids, have emerged as critical mediators of disease progression (6). Elevated plasma ceramide levels strongly correlate with MASH severity (7–13). The excessive deposition of ceramides in the liver can lead to hepatic steatosis (6). Furthermore, ceramides aggravate hepatic inflammation, oxidative stress, apoptosis, and insulin resistance, all of which contribute to various stages of MASLD progression (6, 14). Ceramide biosynthesis occurs primarily in the liver via three main pathways: de novo synthesis, sphingomyelin hydrolysis, and the salvage pathway. Among these, the de novo pathway appears to be the predominant contributor to ceramide accumulation in MASH, both in human patients and animal models (6, 14). Inhibition of this pathway by genetic or pharmacological means has been shown to alleviate MASH progression in preclinical models (7, 15–19), highlighting the therapeutic potential of targeting de novo ceramide synthesis.

Serine palmitoyltransferase (SPT), comprising subunits including SPTLC2, catalyzes the initial and rate-limiting step of de novo ceramide synthesis (20). In this study, we observed up-regulated hepatic Sptlc2 expression along with increased plasma ceramides in patients with MASLD from an Asian cohort and in three MASH animal models. We hypothesized that liver-targeted inhibition of Sptlc2 could suppress ceramide synthesis and thereby alleviating MASH progression.

In this study, we used a clinically validated lipid nanoparticle (LNP) platform, analogous to that used in the Food and Drug Administration–approved small interfering RNA (siRNA) drug Onpattro (21), to deliver Sptlc2-targeting siRNA specifically to hepatocytes with our LNP-SPTLC2 siRNA formulation. This approach effectively silenced Sptlc2 expression both in vitro and in vivo. Intravenous administration of LNP-SPTLC2 siRNA in MASH mice resulted in lower hepatic and circulating ceramide levels, decreased hepatic lipid uptake and de novo lipogenesis, reduced hepatic steatosis and inflammation, and overall attenuation of disease progression in both young and aged mice. Therapeutic efficacy was demonstrated in a short-term methionine-choline–deficient (MCD) diet model and validated in a more clinically relevant long-term choline-deficient high-fat diet (CD-HFD) model. Together, these findings provide compelling evidence that hepatocyte-targeted inhibition of ceramide de novo synthesis via LNP-siRNA represents a promising therapeutic strategy for MASH. This approach holds great translational potential for the development of innovative RNA drugs targeting metabolic liver diseases.

RESULTS

Ceramide levels and hepatic de novo ceramide synthesis correlate with MASLD severity in human patients

Elevated plasma ceramide levels have been consistently observed in patients with MASLD and advanced MASH (7, 8, 10–13, 22). Using a high-throughput lipidomics approach, we assessed plasma ceramide levels in a Singaporean cohort of patients with MASLD. As expected, total plasma ceramide levels were markedly higher in patients with either MASLD (simple steatosis) or advanced MASH (with varying extents of fibrosis) compared to healthy subjects (Fig. 1, A and B). Notably, plasma levels of dihydroceramides were similarly elevated in patients with MASLD and MASH (Fig. 1B and fig. S1).

Fig. 1. Ceramide levels and hepatic de novo ceramide synthesis correlate with MASH severity in human patients.

(A) Plasma total ceramide levels in an Asian human cohort, presented as relative values compared to healthy subjects. (B) Heatmap of human plasma ceramide species, presented as fold change relative to the healthy group. (C) Schematic illustration of major ceramide biosynthesis pathways: I, de novo synthesis; II, sphingomyelinhydrolysis; III, salvage pathway. (D) Heatmap of hepatic expression levels of genes involved in de novo ceramide synthesis, presented as fold change relative to healthy controls. (E) Hepatic SPTLC2 expression levels in the human cohort. (F) Immunohistochemical staining of SPTLC2 in liver paraffin sections from healthy individuals and MASH patients (F0-F1) showing SPTLC2 in brown and nuclei in blue. (G) Quantification of SPTLC2 protein expression based on immunohistochemical staining in healthy individuals and MASH patient liver tissues (n = 3 subjects per group). (H) Violin plot showing SPTLC2 expression across hepatic cell clusters in healthy and MASH human liver samples from single-nucleus RNA sequencing (snRNA-seq). n.s., not significant; *P < 0.05, **P < 0.01, and ***P < 0.001; [(A), (B), (D), (E), and (H)], one-way ANOVA; (G) unpaired Student’s t test. SPT, serine palmitoyl transferase; SPTLC, serine palmitoyl transferase long chain base subunit; Cer, ceramide; 3-KSR, 3-ketodihydrosphingosine reductase; CerS, ceramide synthase; DES, dihydroceramide desaturases; SMase, sphingomyelinase; SMS, sphingomyelin synthase; CDase, ceramidase; S1PP, sphingosine-1-phosphate phosphatase; SphK, sphingosine kinase; GCS, glucosylceramide synthase; GCase, glucosylceramidase.

Ceramides can be generated through three distinct synthesis pathways: de novo synthesis, sphingomyelin hydrolysis pathway, and the sphingolipid salvage pathway, as illustrated in Fig. 1C. To discern the specific synthesis pathway primarily contributing to ceramide elevation in human MASLD, we analyzed hepatic gene expression profiles of ceramide synthesis enzymes using liver biopsies from a European patient cohort encompassing different stages of MASLD. Disease severity was clarified on the basis of the SAF score (steatosis, activity, and fibrosis) ranked by experienced clinicians and pathologists (23). Compared to healthy subjects, patients with simple steatosis or various stages of liver fibrosis showed markedly higher hepatic expression of genes associated with the de novo ceramide synthesis pathway, including Sptlc2, which encodes a subunit of the rate-limiting enzyme, SPT (Fig. 1D). These findings align with our current understanding that enzymes involved in de novo ceramide synthesis—such as SPT, ceramide synthases (CerS2/4/6), and dihydroceramide desaturase 1 (DES1)—play pivotal roles in the pathogenesis of MASLD (6). The CerS5 expression was unexpected down-regulated in MASLD (Fig. 1D). Given that CerS5 and CerS1 are generally considered dispensable in the development of MASLD, the observed down-regulation of CerS5 warrants further investigation. The up-regulation of SPTLC2 across multiple stages of MASLD (Fig. 1E) was confirmed at the protein level by immunohistochemical staining of human liver tissue sections (Fig. 1, F and G), further validating the transcriptional data.

The liver is composed of both hepatocytes and nonparenchymal cells. To identify the primary cell types contributing to the elevated expression of SPTLC2 in MASLD livers, we analyzed publicly available human hepatic single-nucleus RNA sequencing (snRNA-seq) data from the Gene Expression Omnibus (GSE189600) (24). As shown in Fig. 1H, SPTLC2 was predominantly up-regulated in hepatocytes and hepatic stellate cells (HSCs) in patients with MASLD. Given that hepatocytes constitute ~60% while HSCs only 4 to 10% of the total cell population in the human liver (25), our findings suggest that the increased hepatic expression of SPTLC2 and associated ceramide accumulation in MASLD are primarily attributable to hepatocytes. Accordingly, the modest reduction in hepatic SPTLC2 expression observed in stage F4 compared to earlier stages of MASH (Fig. 1E) is likely due to extensive loss of hepatocytes.

Ceramide de novo synthesis is activated in MASH animal models

To ascertain if the elevation of ceramide levels during MASLD progression is recapitulated in animal models, we measured hepatic ceramides in a widely used MASH animal model established by feeding C57BL/6 mice with a MCD diet (26). As expected, MCD diet–fed mice developed steatosis, inflammation (immune cell infiltration), ballooning, and mild-to-moderate perivascular and interstitial fibrosis in the liver (Fig. 2A and fig. S2, A to C), accompanied by elevation of serum levels of alanine aminotransferase (ALT) and aspartate aminotransferase (AST) (Fig. 2, B and C). In addition, hepatic inflammation, as indicated by increased expression of tumor necrosis factor–α (TNFα) and interleukin-1β (IL-1β), developed in these mice (Fig. 2, D and E). All together, these data point to the establishment of MASH features within 8 weeks after starting the MCD diet. Similar to human patients, as shown in Fig. 2F, total hepatic ceramides increased while MASH was progressing. Further analysis revealed that many species of ceramides—including Cer d18:0/16:0, d18:0/20:0, d18:1/16:0, d18:1/18:0, d18:1/24:0, d18:1/24:1, d18:1/26:1, and d18:2/24:1—increased in the liver tissue of MASH mice, particularly after 8 weeks of diet feeding (Fig. 2G). To examine if hepatic ceramide increase could be attributed to de novo ceramide synthesis as observed in human patients, we analyzed all three pathways of ceramide synthesis in the MCD-induced MASH model. As shown in fig. S3 (A and B), hepatic expression levels of acid sphingomyelinase (SMase; encoded by Smpd1) and glucosylceramidase (GCase; encoded by Gba1) was not altered in mice fed the MCD diet. However, hepatic expression of Sptlc2, Degs1, and Cers2—three key genes involved in the de novo ceramide synthesis pathway—was markedly up-regulated (Fig. 2H and fig. S3, C and D). Notably, the hepatic mRNA expression of Sptlc2 increased twofold at 4 weeks and threefold at 8 weeks after starting the MCD diet (Fig. 2H). Accordingly, Western blot analysis demonstrated that SPTLC2 protein levels progressively increased in MASH livers, after 8 weeks of MCD diet, reaching threefold higher than that in healthy controls (Fig. 2I). Although macrophages express Sptlc2 (Fig. 1H), the expression increase in hepatic Sptlc2 was unlikely due to immune infiltration as Sptlc2 was up-regulated by more than two times at 4 weeks when the expression of inflammatory cytokines (TNFα and IL-1β) was not up-regulated yet (Fig. 2, D, E, and H). These data indicate that ceramide de novo synthesis is markedly activated in MCD diet–induced MASH.

Fig. 2. Activation of de novo ceramide synthesis in MASH models.

(A) Representative H&E, Oil Red O (ORO) and Masson Trichrome (MT) staining of mouse liver tissue from the MCD diet–induced MASH model. Scale bar, 100 μm. (B) Serum ALT levels. (C) Serum AST levels. (D) Hepatic Tnf-α mRNA expression. (E) Hepatic Il-1β mRNA expression. (F) Hepatic total ceramide levels in MASH mice. (G) Heatmap of hepatic ceramide species, presented as fold change to the healthy group. Relative concentrations of each detected ceramide species were provided in data S1. (H and I) Hepatic Sptlc2 mRNA and SPTLC2 protein expression across different stages of MCD diet–induced MASH. (J and K) Hepatic Sptlc2 mRNA expression in the CD-HFD (24 weeks on diet) and STAM (8 weeks old) mouse models. N = 3 mice per group. *P < 0.05, **P < 0.01, and ***P < 0.001; [(B) to (E) and (H)] two-way ANOVA; [(F), (G), and (I)] one-way ANOVA; [(J) and (K)] unpaired Student’s t test.

To examine whether ceramide de novo synthesis is universally activated in MASH animal models, we also evaluated the hepatic expression of Sptlc2 in two additional well-established MASH mouse models, i.e., the CD-HFD model (27) and the stelic animal model (STAM), which was established by a single streptozotocin injection at 2 days after birth followed by HFD feeding starting at 4 weeks old (28). When MASH was established (27, 28), the liver tissue was collected from CD-HFD mice at week 24 and STAM mice at week 8, respectively. Similar to the MCD model, the hepatic expression of Sptlc2 was markedly elevated in both the CD-HFD and STAM mouse models (Fig. 2, J and K).

In summary, our data from human patients and three MASH animal models provide compelling evidence that the excessive ceramide production observed in MASH is primarily driven by the hepatic de novo ceramide synthesis pathway. Therefore, we hypothesized that specific knockdown of Sptlc2, the subunit of the rate-limiting enzyme SPT, might be a promising therapeutic strategy of lowering ceramides and thereby possibly alleviating MASH.

Sptlc2 gene knockdown reduces hepatocyte lipid accumulation in vitro

To achieve Sptlc2 knockdown in hepatocytes, we opted to use siRNA for cell transfection. Seven proprietary siRNA sequences (labeled from SPT.S1 to SPT.S7), designed to target both human and murine Sptlc2, were synthesized and screened for knockdown efficiency in murine Hepa 1-6 and human Hep G2 cell lines by transfection using Lipofectamine RNAiMAX. As shown by quantitative polymerase chain reaction (qPCR) analysis of Sptlc2 mRNA expression (Fig. 3, A and B), the siRNA sequence SPT.S1 demonstrated the highest gene silencing efficacy in both cell lines. Accordingly, SPT.S1 was selected for use in subsequent experiments.

Fig. 3. Design and in vitro evaluation of LNP-siRNA to reduce hepatic ceramides.

(A) Screening of seven Sptlc2 siRNA candidates in Hepa 1-6 cells. (B) Screening of seven Sptlc2 siRNA candidates in Hep G2 cells. (C) Schematic outline of the LNP formulation process. (D) Characterization of LNPs: mean particle size, PDI, and zeta potential. (E and F) Stability assessment of LNP stored at 4°C over 28 days, shown as changes in (E) mean particle size and (F) PDI. (G) Dose-dependent knockdown efficiency of LNP-SPTLC2 siRNA in Hepa1-6 cells. (H) Dose-dependent induction of Sptlc2 mRNA expression in Hepa1-6 cells treated with increasing concentrations of OA. (I) ORO staining of Hepa1-6 cells treated with 1.0 mM OA for 24 hours before exposure to LNPs containing scrambled or SPTLC2 siRNA (100 nM). ORO staining was quantified by the staining area and normalized to that in OA-treated cells. Bars represent means ± SEM of three independent experiments. *P < 0.05, **P < 0.01, and ***P < 0.001; one-way ANOVA.

Although Lipofectamine RNAiMAX is cheaper and faster for siRNA screening in vitro, it is unsuitable for intravenous. Administration, we therefore used the LNP technology developed for Onpattro that can deliver siRNA efficiently and specifically to liver hepatocytes following intravenous administration (21). The LNPs were formulated using a T-junction system, as illustrated in Fig. 3C, which ensured a controlled mixing environment and reproducible production of LNPs (29, 30). The encapsulation efficiency of siRNA into LNPs was 95 ± 2%, and the produced LNPs exhibited a quite homogenous size distribution [mean size about 65 nm and polydispersity index (PDI) smaller than 0.1] and neutral surface charge (Fig. 3D). Furthermore, the LNPs demonstrated stability for 21 days during storage at 4°, without notable changes in mean size and PDI (Fig. 3, E and F). To assess the silencing efficiency of siRNA loaded in LNPs, we incubated Hepa1-6 cells with the siRNA-loaded LNPs, i.e., LNP-SPTLC2 siRNA. Dose-dependent silencing of the target gene was observed, with Sptlc2 knockdown reaching a high efficiency of 78.0% (Fig. 3G) comparable to that achieved with Lipofectamine (Fig. 3A).

To evaluate the efficacy of Sptlc2 knockdown in reducing hepatocyte steatosis, we established an in vitro steatosis model by incubating Hepa1-6 cells with oleic acid (OA), as previously described (31, 32). As shown in Fig. 3H, Sptlc2 mRNA expression increased in a dose-dependent manner with OA concentrations of 0.1, 0.5, and 1.0 mM. Based on this response, 1.0 mM OA was selected to establish in vitro hepatocyte steatosis (Fig. 3I). In this model of steatosis, treatment with LNP-SPTLC2 siRNA resulted in fewer lipid droplets in Hepa1-6 cells compared to LNPs containing scrambled siRNA (Fig. 3I). These results indicate that Sptlc2 silencing effectively reduces lipid accumulation in hepatocytes in vitro.

LNP-SPTLC2 siRNA mediates knockdown of Sptlc2 in hepatocytes after intravenous administration

To evaluate targeting of siRNA delivered by LNPs to the liver after intravenous administration, Cy5-labeled siRNA was encapsulated into LNPs. As shown by in vivo imaging system (IVIS) imaging, 1 hour after intravenous administration, the majority of siRNA accumulated in the liver as evident from stronger fluorescence signal compared to other major organs, including the heart, spleen, kidneys, and lungs (Fig. 4, A and B). At the cellular level, flow cytometry analysis showed that nearly 80% of hepatocytes in MASH mice contained siRNA, approximately twice the proportion observed in healthy mice (Fig. 4, C and D). The increased delivery was also observed in case of other liver cell populations, including endothelial cells, HSCs, dendritic cells, and macrophages. However, compared to hepatocytes, the percentage of cells containing siRNA in other cell populations was much lower (Fig. 4D). Since the hepatocytes are the major cell population in the liver, these findings demonstrate that LNPs can effectively deliver siRNA to hepatocytes, with delivery efficiency further enhanced under MASH conditions.

Fig. 4. LNPs deliver siRNA specifically to hepatocytes and induce efficient target gene knockdown.

(A) Whole-body IVIS imaging of mice injected with nonlabeled LNP-siRNA (control group) or LNPs loaded Cy5-labeled siRNA. (B) Representative IVIS image illustrating the distribution of LNPs across various organs. (C) Flow cytometry analysis showing cellular uptake of Cy5-siRNA in hepatocytes. ASGR1 was used as a hepatocyte specific marker. (D) Distribution of Cy5-labeled siRNA among different liver cell populations in healthy mice and MASH mice fed an MCD diet for 8 weeks. Data were presented as the percentage of Cy5-positive cells within each cell type based on FACS analysis. EC, endothelial cells; HSCs, hepatic stellate cells; DCs, dendritic cells. (E to G) Time-course analysis of hepatic Sptlc2 mRNA levels and SPTLC2 protein levels following a single intravenous dose of LNP-siRNA (0.3 mg/kg) in MASH mice fed an MCD diet for 4 weeks. Glyceraldehyde-3-phosphate dehydrogenase was used as a loading control for Western blot analysis, and protein expression levels were normalized to that on day 2 postdosing of LNP-Scrambled siRNA (control). (H) Time-course analysis of hepatic Sptlc2 mRNA levels in healthy mice following siRNA treatment at 0.3 mg/kg per dose, administered either once or twice weekly. The once-weekly group received doses on days 0 and 7, while the twice-weekly group received doses on days 0, 3, 7, and 10. N = 3 mice per group. *P < 0.05, **P < 0.01, and ***P < 0.001; one-way ANOVA.

To ascertain the duration of siRNA-mediated gene silencing in the diseased liver, we monitored hepatic expression of Sptlc2 for 2 weeks following one intravenous dose of LNP-siRNA in MASH mice (4 weeks on MCD diet). As expected, mice injected with LNP-Scrambled siRNA showed no gene silencing effect in the livers. After one dose of LNP-SPTLC2 siRNA, however, a strong silencing effect was observed with the maximal knockdown efficiency of Sptlc2 achieved at 2 days. Then, the mRNA expression of Sptlc2 started to increase and was fully recovered at 14 days after administration (Fig. 4E). In terms of SPTLC2 protein levels, Western blotting analysis showed a modest decrease (20%) at 2 days, a continued reduction to 50% at 7 days, followed by recovery to 80% at 14 days after administration (Fig. 4, F and G). These results suggest the necessity of multiple dose treatment. To determine the administration frequency required for efficient Sptlc2 gene silencing, healthy mice were injected intravenously with LNP-SPTLC2 siRNA once per week (two doses, at day 0 and day 7) or twice per week (four doses, at day 0, day 3, day 7, and day 10). The results demonstrated that a single dose of LNP-SPTLC2 siRNA reduced hepatic Sptlc2 mRNA expression by 70% compared to the control (LNP-Scrambled siRNA) at 2 days postinjection; however, mRNA expression largely recovered by 7 days (Fig. 4H). The second dose at day 7 did not improve overall gene silencing but generated a similar silencing pattern as seen in the first week. In contrast, the intravenous administration of LNP-SPTLC2 siRNA twice per week could achieve and maintain a maximal hepatic silencing of Sptlc2 by ~70% (Fig. 4H).

Given the essential physiological roles of ceramides, we assessed effect of LNP-SPTLC2 siRNA on the viability of hepatocytes both in vitro and in vivo. As determined by PrestoBlue assay, knockdown of Sptlc2 by LNP-SPTLC2 siRNA (up to 100 nM) did not cause obvious death of Hepa1-6 cells (fig. S4A). In vivo, as revealed by TUNEL (terminal deoxynucleotidyl transferase–mediated deoxyuridine triphosphate nick end labeling) staining, no apparent apoptosis of liver cells was observed in healthy mice receiving intravenous administration of LNPs containing either scrambled or SPTLC2 siRNA (fig. S4, B and C). Since some LNPs accumulated in the spleen following intravenous administration (Fig. 4B), we also examined whether LNP-SPTLC2 siRNA induced off-target effects. Compared with LNP-Scrambled siRNA, LNP-SPTLC2 siRNA did not induce notable alteration in the expression of any immune cell markers in the spleen tissue of MASH mice (fig. S4, D and E), indicating no overt disturbance of immune cell function. Notably, Sptlc2 expression in the spleen was not altered in MASH and not reduced by LNP-SPTLC2 siRNA (fig. S4F). In concordance, lipidomic analysis showed that splenic levels of total ceramides remained unaltered in MASH and unaffected by LNP-SPTLC2 siRNA treatment (fig. S4G). These results suggest that ceramide metabolism in the spleen is not affected by MASH stress nor by the “off-target” LNP-SPTLC2 siRNA following intravenous administration in mice. While the exact mechanism is beyond the scope of this study, it warrants further investigation. Together, our findings support the favorable safety profile of LNP-SPTLC2 siRNA both in vitro and in vivo.

Targeted silencing of hepatic Sptlc2 with LNP-SPTLC siRNA reduces liver steatosis, inflammation, and fibrosis in MCD diet–induced MASH

To evaluate the therapeutic efficacy of LNP-SPTLC2 siRNA in our murine MASH model, we administered LNP-SPTLC2 siRNA intravenously twice per week, a dosing frequency established in Fig. 4, for 4 weeks, as illustrated in Fig. 5A. This dosing frequency resulted in sustained silencing of hepatic Sptlc2, achieving a knockdown efficiency substantially higher than that observed with the once-weekly dosing frequency (fig. S5), and led to a pronounced reduction in hepatic total sphingomyelins and ceramides (Fig. 5B). Detailed lipidomic analysis revealed that numerous species of long- and very long-chain ceramides were reduced in liver tissue (Fig. 5C). Moreover, plasma levels of total sphingomyelins and ceramides were markedly decreased by LNP-mediated Sptlc2 gene slicing (fig. S6). As previously reported (33, 34), hepatic levels of glycosphingolipids—including hexosylceramides (HexCer), dihexosylceramide (Hex2Cer) and monosialodihexosylganglioside (GM3)—were elevated in MASH (fig. S7A), whereas their plasma levels remained unchanged in MASH (fig. S7B). Notably, the knockdown of Sptlc2 did not alter glycosphingolipid levels in the liver or plasma, suggesting that the reduction in ceramides by LNP-SPTLC2 siRNA might be partially compensated through the sphingomyelin hydrolysis pathway rather than the salvage pathway. The hepatic biosynthesis of phosphatidylcholine (PC), as expected, was disrupted by choline deficiency in the MCD model (35, 36), resulting in lower PC levels in the liver (fig. S7C). Among other phospholipids, lysophosphatidylethanolamine (LPE), lysophosphatidylcholine (LPC), and phosphatidylethanolamine (PE) remained largely unchanged in the liver. In addition, plasma levels of the phospholipids LPC, LPE, and PC were substantially decreased, whereas PE levels were slightly elevated in MASH (fig. S7D). Similar to glycosphingolipids, LNP-SPTLC2 siRNA treatment did not affect phospholipid levels in either the liver or plasma (fig. S7). At the histological level, the inhibition of de novo ceramide synthesis via SPTLC2 siRNA knockdown restored disrupted liver tissue morphology, as evidenced by hematoxylin and eosin (H&E) staining (Fig. 5D). Strikingly, both hepatic steatosis and fibrosis were remarkably reduced compared to the control treatment with LNP-Scrambled siRNA (Fig. 5, D to F). In agreement, the hepatic expression of fibrosis-related genes—including Tgf-β, Col1a1, Col3a1 and Timp-1—was inhibited (Fig. 5H). Moreover, hepatic inflammation—as indicated by expression of the cytokine genes of Il-1β, Tnf-α and Ccl2—were attenuated by silencing of Sptlc2 (Fig. 5G). Consequently, the overall liver damage, as evaluated by serum AST and ALT levels, was mitigated (Fig. 5I). Since dosing frequency influences the intensity of gene silencing (Fig. 4H), we also evaluated the therapeutic efficacy of a less intensive Sptlc2-silencing regimen. As shown in figs. S8 and S9, a treatment protocol involving four doses of LNP-SPTLC2 siRNA (administered once weekly for 4 weeks) produced notable therapeutic effects in MASH, albeit less pronounced than that achieved with the more frequent dosing regimen shown in Fig. 5. Together, sustained hepatic gene silencing of Sptlc2 shows markedly treatment efficacy in MCD-induced MASH.

Fig. 5. Multiple dosing of LNP-SPTLC2 siRNA achieves effective therapy in the MCD diet–induced MASH.

(A) Schematic illustration of the experimental design. Control treatment, LNP-Scrambled siRNA; siRNA treatment, LNP-SPTLC2 siRNA. Dosing regimen: 0.3 mg siRNA/kg, administered twice weekly for 4 weeks. Created in BioRender. Huang, C. (2025) https://BioRender.com/80olali. (B) Hepatic levels of total sphingomyelins and ceramides in MASH mice after treatment. (C) Species of ceramides that decreased in MASH mouse liver after treatment. The relative concentrations of each detected ceramide species were included in data S1. (D) Liver histology analysis by H&E staining (scale bar, 200 μm) for steatosis and inflammation, ORO staining (scale bar, 500 μm) for lipid content and Masson’s trichrome staining (scale bar, 100 μm) for fibrosis. (E) Quantification of lipid content by ORO staining. (F) Quantification of fibrotic area by MT staining. (G) Liver markers for inflammation (Il-1β, Tnf-α, and Ccl2). (H) Liver markers for fibrosis (Tgf-β, Col1a1, Col3a1, and Timp-1). (I) Serum AST and AST levels. (J) MR imaging for evaluation of treatment efficacy (n = 3). N = 5 mice per group except (J). *P < 0.05, **P < 0.01, and ***P < 0.001; one-way ANOVA.

Clinically, liver biopsy is the gold standard for MASH diagnosis. However, because of its invasive nature and accompanied bleeding risk, noninvasive diagnosis approaches including magnetic resonance (MR) imaging are highly desirable and have been explored in recent years (37, 38). Toward this effort, we have recently developed a nanoprobe, ultrasmall iron oxide-Dopamine-Amphiphilic poly-(ethylene glycol)-5-hydroxytryptamine (UISO-DA-PEG-5HT), which enhances MR imaging in response to myeloperoxidase (MPO), an oxidative enzyme highly expressed in MASH liver (fig. S10) (39, 40). Here, to assess the therapeutic efficacy of LNP-SPTLC2 siRNA by noninvasive diagnosis, we used this MPO-responsive nanoprobe for MR imaging (Fig. 5J) (40). In the healthy liver, the marginal activity of MPO resulted in a lower ratio of the T₂ relaxation rate (R₂), attributable to aggregated nanoprobes, to the T₁ relaxation rate (R₁), associated with dispersed nanoprobes, compared to that observed in MASH (Fig. 5J). Notably, the R₂/R₁ values in the LNP-SPTLC2 siRNA treatment group were markedly lower than those observed in animals treated with LNP-Scrambled siRNA (Fig. 5J). As expected, the liver MPO activity was also substantially reduced in LNP-SPTLC2 siRNA–treated mice compared to MASH mice receiving LNP-Scrambled RNA (fig. S10). Together, these results indicate that targeted knockdown of hepatic Sptlc2 with LNP-SPTLC siRNA attenuates MASH and that the achieved therapeutic effects can be noninvasively monitored by MR imaging.

Evaluation of LNP-SPTLC2 siRNA treatment efficacy in a chronic obese MASH model

Considering that MASH is often associated with obesity and diabetes, while in the MCD mouse model the diet leads to loss of body weight (41), we aimed to investigate whether LNP-SPTLC2 siRNA treatment is also efficacious in an obese mouse model, the CD-HFD model established by feeding mice the CD-HFD for 1 year (Fig. 6A). In case of this model, the mice gain weight (fig. S11A) and MASH features are usually established within 26 weeks after starting the diet (27). In this study, mice were fed with a CD-HFD diet for 54 weeks to achieve a severe MASH stage before initiating the treatment. The mice were then treated with two doses of LNP-SPTLC2 siRNA per week for a total of 4 weeks (eight doses in total) and euthanized 3 days after the last dose (illustrated in Fig. 6A). Even 3 days after the final dose, the hepatic expression of Sptlc2 remained reduced by 30% compared to untreated MASH mice or those received LNP-Scrambled siRNA (fig. S11B). Similar to the findings in the MCD diet–induced MASH model, hepatic Sptlc2 knockdown led to a notable reduction of sphingomyelin and ceramide levels in both plasma (fig. S11, C and D) and liver tissue (Fig. 6, B and C). Liver steatosis was reduced by 20% in treated mice, as demonstrated by liver H&E staining and ORO staining (Fig. 6, D and E). Inflammation was also attenuated, as indicated by decreased hepatic mRNA expression of Il-1β, Tnf-α, and Ccl2 in the LNP-SPTLC2 siRNA treatment group (Fig. 6F). Correspondingly, serum levels of AST and ALT were lower upon treatment (Fig. 6G). Moreover, mice treated with LNP-SPTLC2 siRNA appeared to gain less weight compared to those treated with LNP-Scrambled siRNA (fig. S11E). These findings indicate that hepatic inhibition of de novo ceramide synthesis by LNP-SPTLC2 siRNA is also effective in the obese MASH model induced by the CD-HFD diet.

Fig. 6. Therapeutic efficacy of LNP-SPTLC2 siRNA in CD-HFD diet–induced MASH.

(A) Schematic illustration of the experimental design. Control treatment, LNP-Scrambled siRNA; siRNA treatment, LNP-SPTLC2 siRNA. Dosing regimen: 0.3 mg siRNA/kg, administered twice weekly for 4 weeks. Created in BioRender. C. Huang (2025). https://BioRender.com/favp6kf. (B) Hepatic levels of total sphingomyelins and ceramides in MASH mice after treatment. (C) Species of main ceramides in the MASH mouse liver after treatment, the relative concentrations of each detected ceramide species are included in data S1. (D) Liver histology analysis by H&E staining (scale bar, 200 μm) for steatosis and inflammation and ORO staining (scale bar, 500 μm) for lipid content. (E) Quantification of lipid content by ORO staining. (F) Liver markers for inflammation (Il-1β, Tnf-α, and Ccl2). (G) Serum AST and AST levels. Healthy mice, n = 5; Untreated MASH mice, n = 4; control and siRNA treatment groups, n = 7 mice per group. *P < 0.05, **P < 0.01, and ***P < 0.001; one-way ANOVA.

Lowering ceramides ameliorates liver steatosis by inhibiting hepatic lipid uptake and lipogenesis

Hepatocytes use a variety of receptors and transporters, such as fatty acid transporter proteins 2 and 5 (FATP2 and FATP5) and the fatty acid translocase CD36, to facilitate the uptake of fatty acids (42–44). The dysregulation of de novo lipogenesis, i.e., the synthesis of fatty acid chains from acetyl–coenzyme A subunits produced during glycolysis, contributes to liver steatosis. In patients with MASLD, de novo lipogenesis has been reported to increase by three to five times (45–48). This process is regulated by sterol regulatory element-binding protein-1c (SREBP-1c), a master transcription factor that governs triglyceride and cholesterol biosynthesis. SREBP-1c is up-regulated in the liver tissue from both MASLD patients (49–51) and MASH animal models (52, 53). Notably, ceramides have been shown to induce the expression and activation of SREBP-1c (54, 55). In our study, we demonstrated that reducing hepatic ceramide synthesis with our hepatocyte-targeted LNP-siRNA formulation improved liver steatosis in MASH. Based on these findings, we hypothesize that LNP-SPTLC2 siRNA–mediated targeted inhibition of hepatic ceramide production mitigates liver steatosis by attenuating both hepatic lipid uptake and de novo lipogenesis (Fig. 7A).

Fig. 7. Ceramide reduction ameliorates liver steatosis by suppression of hepatic lipid uptake and lipogenesis.

(A) Schematic illustration of the proposed mechanism via which lowering hepatic ceramide levels alleviates liver steatosis following LNP-SPTLC2 siRNA treatment. Administration of LNP-SPTLC2 siRNA caused degradation of the target mRNA and a subsequent reduction in hepatic SPTLC2 protein levels, thereby inhibiting the de novo ceramide synthesis pathway. The resulting reduction in ceramide levels is proposed to attenuate CD36-mediated lipid uptake and SREBP-1c–driven lipogenesis, ultimately reducing lipid accumulation in the liver. Created in BioRender. C. Huang (2025). https://BioRender.com/hg7w6y5. (B) Sptlc2 knockdown inhibits expression of Cd36, Fatp2, and Fatp5 in Hepa 1-6 cells incubated with 1.0 mM OA. SPTLC2 siRNA was delivered via LNPs, and mRNA was extracted 24 hours posttransfection. Gene expression levels were normalized to Gapdh. (C) Hepatic Cd36, Fatp2, and Fatp5 levels in the MCD diet–induced MASH. (D) Hepatic Srebp-1c, Acc1, and Fas levels in the MCD diet–induced MASH. (E) Western blot analysis of hepatic CD36 and SREBP-1c protein levels in the MCD diet–induced MASH. Control, LNP-Scrambled siRNA; siRNA, LNP-SPTLC2 siRNA. (F) Heatmap of hepatic TG levels in different groups of mice by lipidomics. (G) Quantification of abundant TG species in mouse livers. MASH was established with the MCD diet, and mice were treated according to the protocol illustrated in Fig. 5A. N = 5 mice per group. *P < 0.05, **P < 0.01, and ***P < 0.001; one-way ANOVA.

To test our hypothesis, we first examined the influence of Sptlc2 knockdown on the expression of the key receptors that mediate lipid uptake. As shown in Fig. 7B, in the OA-induced hepatocyte steatosis in vitro model, Sptlc2 knockdown by LNP-SPTLC2 siRNA resulted in down-regulation of Cd36, Fatp2, and Fatp5 genes. In concordance, similar effects were observed in the MCD-induced MASH liver in vivo (Fig. 7C). These results suggest that hepatic ceramide lowering inhibits fatty acid uptake by the liver. We then examined the expression of SREBP-1c, a crucial transcription factor responsible for regulating the expression of genes involved in lipogenesis and glycolysis. As shown in Fig. 7D, hepatic Sptlc2 knockdown down reduced liver expression of Srebp-1c in MASH. Accordingly, gene expression of Acc1 and Fas, two downstream fatty acid synthesis enzymes regulated by SREBP-1c, also decreased after treatment (Fig. 7D). The down-regulation of CD36 and SREBP-1c in mouse liver tissue by LNP-SPTLC2 siRNA was confirmed by Western blotting (Fig. 7E). Similar to the observations in the MCD model, treatment with LNP-SPTLC2 siRNA also reduced hepatic expression of key mediators involved in lipid uptake (Cd36, Fatp2, and Fatp5) and de novo lipogenesis (Srebp-1c, Acc1, and Fas) in the CD-HFD–induced MASH model (fig. S12). As revealed by lipidomics, hepatic levels of several triglyceride species, the main products of de novo lipogenesis, were reduced in mice treated with LNP-SPTLC2 siRNA (Fig. 7, F and G). These findings indicate that inhibition of hepatic de novo ceramide synthesis ameliorates liver steatosis by suppressing both fatty acid uptake and de novo lipogenesis.

DISCUSSION

Current treatment paradigms for MASH primarily rely on dietary restrictions, lifestyle interventions, and the use of antidiabetic and antiobesity medications. However, the limited therapeutic efficacy of these approaches underscore the urgent need to explore alternative therapeutic strategies. Recent findings implicates ceramides as key contributors to every stage of MASLD progression, suggesting that ceramide reduction has potential to alleviate MASLD-related pathology across the disease spectrum. Various approaches have been investigated to inhibit ceramide production in preclinical models, including adeno-associated virus (AAV)–mediated hepatic gene knockdown with short hairpin RNAs (16) and oral administration of ceramide synthesis inhibitors (7, 19), as reviewed in (6). However, AAV-based gene therapies carry potential risk of immunogenicity, hepatotoxicity, neurotoxicity, and oncogenicity. Several adverse effects also have been reported in clinical trials involving high-dose AAV administration (56). Similarly, ceramide synthesis inhibitors such as myriocin and fenretinide have demonstrated clear therapeutic efficacy in MASH animal models; however, their clinical translation has been impeded by the requirement of long-term, high-dose oral administration and associated side effects (6). In this study, we used the clinically validated Onpattro LNP system to deliver a proprietary siRNA-targeting SPT, the rate-limiting enzyme of the de novo ceramide synthesis pathway, specifically in hepatocytes. This targeted and clinically viable approach achieved robust suppression of hepatic ceramide production and demonstrated promising treatment efficacy in two distinct mouse models of MASH.

By profiling plasma lipids, we observed an increase in various species of ceramides in both patients with MALSD/MASH and across three mouse models of MASH, supporting the notion that increased plasma and hepatic ceramide levels are closely associated with the pathogenesis of MASLD and MASH (7, 8, 10–13, 22). Further analysis revealed that plasma and hepatic levels of dihydroceramides (DhCers) were also markedly elevated, consistent with a recent report (57). Because DhCers are predominantly produced via the de novo ceramide synthesis pathway (with some contribution from the salvage pathway) (6, 16), these findings suggest the activation of ceramide de novo synthesis in MASH. The increase of ceramides was strongly associated with the up-regulation of key enzymes involved in the ceramide de novo synthesis pathway, including SPT and CerSs. In particular, Sptlc2, a subunit of the rate-limiting enzyme SPT in ceramide de novo synthesis, was consistently up-regulated in all three MASH models (MCD, CD-HFD, and STAM) tested in this study and in an atherogenic MASH model (35) and the DIAMOND mouse model (58). These converging lines of evidence support the role of SPT in MASH pathophysiology. In contrast, the hepatic expression of key enzymes involved in the salvage pathway (GCase, encoded by Gba1) or the sphingomyelin hydrolysis pathway (acid SMase, encoded by Smpd1) remained unchanged. The expression of two well-characterized neutral SMAses (nSMases), nSMase1 and nSMase2 (encoded by Smpd2 and Smpd3, respectively), was unaltered in the MCD model but modestly elevated in the CD-HFD model (fig. S3E). These results suggest that the sphingomyelin hydrolysis pathway may contribute to ceramide production in MASH, albeit to a lesser extent. Different from our findings, Jiang et al. (57) reported that hepatic expression of Smpd3 was more pronounced than Sptlc2 in MASH. This seemingly discrepancy might stem from the small sample size and the semi-quantitative nature of Western blot analysis used in that study. Despite the discrepancies, both Jiang’s study and our current work underscore the critical role of ceramide synthesis in MASH pathogenesis and highlight the therapeutic potential of targeting ceramide production. Our data indicate that while multiple ceramide-generating pathways may be involved in MASH progression, Sptlc2 and the de novo synthesis pathway are the predominant contributors in our models, highlighting a promising therapeutic target. Although the primary focus of the current study is to develop ceramide-lowering RNA therapeutics for MASH, future studies should include more comprehensive, quantitative, and statistically powered analyses of all enzymes involved in ceramide metabolism to delineate the specific contributions of each pathway across different stages of the disease. Nevertheless, our findings indicate that activation of hepatic ceramide de novo synthesis pathway contributes to excessive ceramide production in MASH. Intriguingly, our snRNA-seq analysis uncovered that SPTLC2 was specifically up-regulated in hepatocytes and HSCs. These data demonstrate that hepatocytes are the primary ceramide-producing cell population and suggest that targeting de novo ceramide synthesis in hepatocytes mitigates MASH.

Patisiran (trade name Onpattro) is an LNP formulation of siRNA for the treatment of polyneuropathies by inhibiting hepatocyte production of transthyretin amyloid (59, 60). In this study, we confirmed in healthy mice that LNPs were able to deliver siRNAs to up 30% of the hepatocytes within 1 hour following intravenous administration, achieving the highest in vivo transfection efficiency among all liver cell populations. Strikingly, the hepatocyte-specific delivery of siRNA was more than doubled in MASH mice, likely attributed to increased access of the hepatocytes to LNP-siRNA due to enhanced hepatic microvascular permeability under inflammatory conditions. While the exact mechanisms underlying this phenomenon warrants further investigation, this observation prompted us to use the LNP formulation here for hepatocyte-specific gene knockdown. We achieved a sustained 70% hepatic knockdown of the target gene, Sptlc2, in vivo with this LNP-siRNA formulation. Accordingly, the intravenous administration of LNP-SPTLC2 siRNA effectively lowered both hepatic and plasma levels of ceramides. In concordance with previous studies (7, 15, 16, 19), ceramide lowering halted MASH progression, as demonstrated by biochemical markers, histological analysis, and noninvasive MR imaging. Mechanistically, our results suggest that ceramide lowering alleviates MASH by reducing lipid uptake and inhibiting lipogenesis in the liver. Although ceramides are considered to be generated in response to an excess of fatty acids, elevated ceramides can create a positive feedback loop, leading to further accumulation of lipids in the liver (61, 62). In agreement, ceramides have been reported to exert lipotoxicity by enhancing hepatic lipid uptake and storage (16, 55, 63). Moreover, ceramides, especially C16:0 ceramide, have been reported to impair hepatic β-oxidation of fatty acids (18, 64). Accordingly, the inhibition of de novo ceramide synthesis with myriocin or through genetic deletion of its key enzymes (CerS2 or CerS6) enhances fatty acid β-oxidation by increasing mitochondrial complex IV activity (18, 64). Given that a broad spectrum of ceramides, including C16:0 ceramide, were markedly reduced by LNP-SPTLC2 siRNA in both the MCD and CD-HFD models, we speculate that Sptlc2 knockdown may enhance fat breakdown by promoting β-oxidation. Together, reducing ceramide levels may help to break the cycle of hepatic lipid accumulation, ultimately improving liver health.

Several animal models of MASH have been established. While the MCD diet–induced MASH model has been instrumental in studying specific pathological features, such as severe steatosis and inflammation, it does not fully capture the complexity of human MASH pathology (26, 41, 65). To address this, we confirmed the pathological involvement of de novo ceramide synthesis by demonstrating up-regulation of sptlc2 in two additional MASH mouse models, the STAM model and the CD-HFD model, as well as in human patients. Furthermore, we validated the treatment efficacy of LNP-SPTLC2 siRNA in the CD-HFD mouse model, which is considered more clinically relevant (27). In this context, “clinically relevant” refers to the ability of the CD-HFD model to more accurately reflect the multifactorial nature of human MASH, such as insulin resistance, obesity, and atherosclerosis (27). Despite ceramide lowering, the therapeutic effects were less pronounced in the CD-HFD model compared to the MCD model, likely due to the advanced disease stage (following 1 year of diet feeding) and the relatively short treatment duration. Nonetheless, we demonstrated the feasibility of alleviating MASH in obese mice. Our choice to validate the therapeutic effects of LNP-SPTLC2 siRNA in the CD-HFD model was particularly motivated by the necessity to assess its efficacy in a setting that better reflects the obesity-associated features commonly seen in human MASH. Moreover, since MASH was already established in this model by 4 to 6 months (27), our findings also suggest that LNP-SPTLC2 siRNA remains effective in treating advanced MASH in aged mice. Collectively, our findings in both the MCD and CD-HFD models demonstrate the broader applicability of LNP-SPTLC2 siRNA therapy across different stages and phenotypes of MASH. Compared to permanent gene knockout strategies, siRNA-based approaches offer the advantages of transient interventions and potentially minimizing the risk of off-target effects. Moreover, hepatocyte-targeted delivery using LNPs enhances therapeutic efficacy without detectable side effects.

Sptlc2 knockdown may reduce serine utilization for sphingolipid biosynthesis and thereby increasing its availability for one-carbon metabolism (66). Given the critical involvement of one-carbon metabolism in MASH pathogenesis (67–69), one question is raised: Could the observed benefit of Sptlc2 knockdown be due to increased one-carbon units compensating for the dietary deficiency of choline and methionine in the MASH models used in this study? However, the MCD diet is deficient in both methionine and choline, two key substrates required for one-carbon metabolism and phospholipid biosynthesis. Methionine is an essential amino acid and a direct donor of methyl groups, while choline is a conditionally essential nutrient and a precursor for PC synthesis. Although serine contributes one-carbon units via the folate cycle, the absence of methionine and choline represents a critical bottleneck for both methylation reactions and PC synthesis and thus cannot be fully compensated by increased serine availability alone. This is supported by our lipidomics data that hepatic PC production was reduced in MASH, and the reduction was not affected by Sptlc2 knockdown (fig. S7). Even in the CD-HFD model where normal methionine is present, the absence of choline still limits PC synthesis. While increased serine availability due to Sptlc2 knockdown may modestly support one-carbon metabolism in this context, it is unlikely to fully overcome the deficit in choline-dependent pathways. Although we did not directly measure hepatic levels of methionine, choline, serine, or glycine, in theory, Sptlc2 knockdown should increase hepatic levels of serine, potentially leading to a minor increase in glycine. However, methionine and choline levels are unlikely altered as they are tightly controlled by dietary availability and independent biosynthetic pathways (66, 70). Therefore, the therapeutic efficacy of LNP-SPTLC2 siRNA treatment in both the MCD and CD-HFD models is unlikely attributed to increased serine availability, but rather to the direct inhibition of de novo ceramide synthesis.

While this study demonstrates the remarkable therapeutic efficacy of LNP-SPTLC2 siRNA in MASH, some limitations remain. First, in this proof of concept, we used siRNAs with minimal modifications, and thus, a regimen of two injections per week is required to achieve sustained knockdown of sptlc2 and therapeutic effects. More intense therapeutic benefits may be obtained by optimizing siRNA stability (e.g., via chemical modifications) and dosing frequency accordingly. Second, although liver steatosis and injury markers were comprehensively assessed in our study, the effect of sptlc2 knockdown on glycaemic control was not examined. Insulin resistance is a hallmark of MASH, particularly in patients with obesity and type 2 diabetes (3). Given that insulin resistance is present in the CD-HFD MASH model with long-term feeding (27) and that high levels of hepatic ceramides induce insulin resistance in mice (16), it is reasonable to hypothesize that hepatic inhibition of SPTLC2 may increase insulin sensitivity in this model. Our 4-week treatment with LNP-SPTLC2 siRNA in advanced MASH, induced by prolonged CD-HFD feeding, led to progressively reduction in mouse body weight (fig S11E). As inhibition of de novo ceramide synthesis by myriocin or CerS6 deletion does not alter food intake (64, 71), our Sptlc2 knockdown approach is unlikely to affect feeding behavior in the current study. Instead, the observed body weight loss may reflect enhanced insulin sensitivity or increased energy expenditure following treatment. Further investigation with a prolonged treatment regimen is warranted to confirm this hypothesis. Last, long-term safety and efficacy studies of LNP-SPTLC2 siRNA in large animal models, such as the Ossabaw miniature swine (72), are essential to advance clinical translation of our findings.

In conclusion, our study reinforces the pivotal role of ceramides in liver lipid metabolism during MASH progression and highlights the therapeutic potential of LNP-siRNA–based strategies targeting hepatocyte de novo ceramide synthesis. Prospectively, combining siRNA-based therapies with other treatment modalities may further improve the treatment outcomes in MASH. As RNA delivery technologies evolve, LNP-RNA–based therapeutics hold promise to transform the treatment landscape for MASH, offering renewed hope for patients. Given the well-established links between ceramide levels with cardiovascular disease, obesity, and diabetes (73–75), our findings may also pave the way for the development of liver-targeted RNA therapies for a broader spectrum of metabolic disorders.

MATERIALS AND METHODS

Clinical cohort

A total of 117 patients and 24 healthy donors from Asia were enrolled in this study under the Ensemble of Multi-disciplinary Systems and Integrated Omics for NAFLD (EMULSION) program. This cohort was approved by the Singapore National Healthcare Group Domain Specific Review Board. Written informed consent was obtained from all participants. Patients were grouped into control, MASLD, and MASH groups based on NAFLD Activity Score ranked by two independent blinded expert pathologists. Plasma samples from these patients were subsequently obtained by centrifugation at 4000g for 10 min and stored at −80°C until further lipidomic analysis. Healthy liver paraffin sections were obtained from liver transplant donors, whereas MASH liver paraffin sections were obtained from patients undergoing liver biopsies as part of their clinical evaluation for MASH. All liver samples were classified as healthy and MASH by experienced clinicians and pathologists. Ethical approval for sample collection and usage was granted by local ethical boards.

Human liver RNA-seq data were sourced from the European NAFLD Registry (NCT04442334) (76). The discovery cohort included 216 snap-frozen biopsy samples obtained from 206 patients diagnosed with MASLD in France, Germany, Italy, or the United Kingdom. In addition, the dataset comprised 10 healthy obese control cases, selected from patients undergoing bariatric surgery in France, who exhibited no biochemical or histological signs of MASLD. Selection criteria included study participation and the availability of adequate high-quality RNA suitable for liver biopsy sequencing. Hepatic snRNA-seq data were retrieved from the publicly available Gene Expression Omnibus dataset GSE189600. Raw FASTQ files were aligned to the Homo sapiens genome assembly GRCh38 (hg38) using CellRanger (3.0.2). For human snRNA-seq, targeted libraries FASTQ files were directly merged with input libraries. Unique molecular identifier count matrices were imported into Seurat (3.2.0) to generate Seurat object for each experiment. Sequencing was performed on the GPL24676 Illumina NextSeq6000 platform. All analyses were conducted using R version 4.3.1 (24).

Cell culture

The Hepa 1-6 [American Type Culture Collection (ATCC) CRL-1830] and Hep G2 (ATCC HB-8065) cell lines were used for in vitro experiments. Hepa 1-6 is a murine hepatoma derived from the BW7756 hepatoma tumor that arose spontaneously in C57L/J mice. Hep G2 is a human hepatocellular carcinoma cell line with epithelial-like morphology, originally isolated from a 15-year-old Caucasian male with liver cancer. Both cell lines were cultured in a culture medium [90% Dulbecco’s modified Eagle’s medium (Gibco, New York, USA) + 10% fetal bovine serum (Gibco, Massachusetts, USA) + 1% penicillin-streptomycin]. For steatosis induction, Hepa 1-6 cells were seeded in 96-well plates and grown to ~70% confluence, followed by incubation with various concentrations (0.1, 0.5, and 1.0 mM) of sodium oleate (Sigma-Aldrich, Missouri, USA) complexed with bovine serum albumin (BSA) for 24 hours.

Animals models

Male C57BL/6 J mice were obtained from InVivos (Singapore). Three MASH models were established as previously described: the methionine- and choline-deficient (MCD) diet model (26), the CD-HFD model (27), and the STAM (28). Briefly, 8-week-old mice were fed an MCD diet (TD90262, Teklad Mills, Madison, WI) for 8 weeks to induce MASH. For the CD-HFD model, 4-week-old mice were fed a CD-HFD diet (Research Diets, D05010402) for 12 months to establish MASH. The STAM model was induced by administering a single subcutaneous injection of 200 μg streptozotocin (STZ) to male mice at 2 days of age, followed by feeding with a 60% kcal HFD (Research Diets, D12492i) from 4 to 8 weeks of age. All animal experiments were approved by the Institutional Animal Care and Use Committee of the National University of Singapore and conducted in accordance with the National Advisory Committee for Laboratory Animal Research (NACLAR, Singapore, 2004) guidelines and the US National Institutes of Health Guide for the Care and Use of Laboratory Animals (8th edition, 2011).

LNP formulation and characterization

LNPs were prepared via microfluidic mixing using a T-connector (1/16″ outer diameter) and appropriate tubing. The flow was driven by two PHD Ultra Syringe Pumps (Harvard Apparatus, Texas, USA). A lipid solution composed of [(6Z,9Z,28Z,31Z)-heptatriaconta-6,9,28,31-tetraen-19-yl] 4-(dimethylamino)butanoate (DLin-MC3-DMA), cholesterol, 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC), and 1,2-dimyristoyl-rac-glycero-3-methoxypolyethylene glycol-2000 (DMG-PEG 2000) (molar ratio of 50:10:38.5:1.5) in 100% ethanol was mixed with a siRNA solution in 20 mM sodium acetate buffer (pH 4) at a total flow rate of 12 ml/min with a 1:3 volumetric ratio of lipid to siRNA solution. For biodistribution studies, Cy5-labeled siRNA was used. Following formulation, LNPs were dialyzed in phosphate-buffered saline (PBS), filtered through a 0.2-μm membrane, and concentrated using Amicon centrifugal spin filters (Thermo Fisher Scientific, catalog no. 87730) with a 10-kDa molecular weight cut-off.

Particle size and PDI were measured using dynamic light scattering (DLS) at a detection angle of 173° on a Zetasizer Ultra (Malvern Instruments, Worcestershire, UK). Zeta potential was determined via electrophoretic light scattering (ELS) using the same instrument and solutions as used for DLS. Each sample was analyzed in triplicate for DLS and five replicates for ELS, with data acquisition performed using ZX Explorer software (version 7.10).

siRNA encapsulation efficiency was quantified using the Quant-iT RiboGreen RNA Assay Kit (Invitrogen, Thermo Fisher Scientific, catalog no. Q33140). Total siRNA content (RNA_tx-100) was measured in the presence of 0.5% (v/v) Triton X-100, while free (unencapsulated) siRNA (RNA_PBS) was assessed in Tris-EDTA (TE) buffer or PBS. Encapsulation efficiency (%) was calculated using the formula: Encapsulation efficiency = [(RNA_tx-100 − RNA_PBS)/RNA_tx-100] × 100.

Cell viability assay

Hepa 1-6 cells were seeded at a density of 50,000 cells per well in 96-well plates and incubated with varying concentrations (1, 10, and 100 nM) of scrambled or SPTLC2 siRNA-loaded LNPs for 24 hours. After incubation, the cells were washed once with PBS, and PrestoBlue Cell Viability Reagent (Thermo Fisher Scientific, catalog no. A13262) was added. The plates were then incubated for 1 hour before fluorescence was measured using a microplate reader (PerkinElmer). Cell viability was normalized to untreated controls and expressed as a percentage.

TUNEL staining

C57BL/6J mice were intravenously injected with either LNP-Scrambled siRNA or LNP-SPTLC2 siRNA (0.3 mg/kg). Livers were harvested 48 hours postinjection, and cryosections (5-μm thick) were prepared for TUNEL staining. Apoptotic cells were detected using the TUNEL assay kit (Cell Signaling Technology, catalog no. 25879), according to the manufacturer’s protocol. Apoptotic cells were fluorescently labeled in green and quantified as a percentage of total cell nuclei.

Biodistribution of LNPs in vivo

The biodistribution of Cy5-labeled siRNA-loaded LNPs was assessed using an IVIS (PerkinElmer, USA) and flow cytometry (BD Biosciences, New Jersey, USA). Mice were euthanized 1 hour after intravenous injection of LNPs, followed by perfusion. Major organs—including the liver, lung, heart, and spleen—were harvested and imaged using the Cy5 fluorescence channel (excitation: 640 nm; emission: 680 nm). Fluorescence images were analyzed using Living Image software (PerkinElmer).

To evaluate cellular level biodistribution within the liver, tissues were enzymatically digested in 500 μl of collagenase I (Gibco, MA, USA) supplemented with 10 μl of deoxyribonuclease and incubated in a 37°C shaking water bath for 40 min in 5-ml tubes. Tissues were then homogenized using 3-ml syringes fitted with 25-gauge needles (Terumo Corporation, Tokyo, Japan) and filtered through 100-μm nylon mesh filters (Greiner Bio-One, Upper Austria, Austria). The resulting single-cell suspensions were centrifuged at 50g for 5 min at room temperature and washed before being resuspended at appropriate concentrations for downstream analysis. Cells were stained with an ASGR1 antibody (Proteintech, catalog no. CL488-11739, Illinois, USA) for 30 min at 4°C in the dark. After three washes with staining buffer, the percentage of Cy5-positive hepatocytes, indicative of LNP-siRNA uptake, was quantified using flow cytometry (BD Biosciences, New Jersey, USA).

H&E staining

Paraffin-embedded tissue sections were deparaffinized in xylene and rehydrated through a graded ethanol series (100, 95, and 70%). The sections were stained with Mayer’s hematoxylin (Sigma-Aldrich, H9627) for 5 min, followed by a 5-min rinse under running tap water to induce bluing. Subsequently, they were counterstained with 0.5% eosin (Merck, E4382) for 1 min and briefly rinsed by dipping 10 times in distilled water. Last, the sections were mounted using VectaMount AQ (Vector Laboratories Inc., Burlingame, USA) and covered with a glass coverslip.

Oil Red O staining

Tissue sections were first washed twice with PBS and once with Milli-Q water, followed by incubation in 0.3% Oil Red O working solution (Sigma-Aldrich, O0625) for 1 hour. After staining, slides were drained and briefly washed in 60% isopropanol until the background stain disappeared. The sections were then rinsed twice with Milli-Q water and counterstained with Mayer’s hematoxylin for 1 min to visualize nuclei. Slides were washed under running tap water for 5 min and mounted with a coverslip using VectaMount AQ (Vector Laboratories).

Masson’s Trichrome Staining

Paraffin-embedded tissue sections were deparaffinized in xylene and rehydrated through a graded ethanol series (100, 95, and 70%). Sections were then rinsed in distilled water and incubated in Bouin’s solution at 56°C for 1 hour to enhance staining. After thorough rinsing under running tap water to remove excess picric acid, the sections were stained with Weigert’s iron hematoxylin for 10 min. Excess stain was washed off under running tap water, followed by brief differentiation in acid alcohol. Slides were subsequently stained with Masson’s trichrome solution (Solarbio, G1346) for 15 min, rinsed in distilled water, and differentiated in 1% acetic acid solution. Last, slides were dehydrated through graded ethanol, cleared in xylene, and mounted with a coverslip using VectaMount AQ (Vector Laboratories).

Immunohistochemistry staining

Paraffin-embedded sections were subjected to antigen retrieval by boiling in sodium citrate buffer (pH 6.0). Endogenous peroxidase activity was quenched by incubation in 3% hydrogen peroxide solution (Sigma-Aldrich, 516813). Nonspecific binding was blocked using 3% BSA for 30 min at room temperature. Sections were incubated with primary antibodies overnight at 4°C, followed by horseradish peroxidase (HRP)–conjugated secondary antibodies. Detection was carried out using 3,3′-diaminobenzidine (Sigma-Aldrich, D12384) and counterstaining with hematoxylin. Slides were then mounted with a coverslip using VectaMount AQ.

Serum ALT and AST measurement

Mouse serum levels of ALT and AST were quantified colorimetrically using commercial assay kits (Abcam, catalog no. ab105134 for ALT and catalog no. ab105135 for AST), following the manufacturer’s instructions.

qRT-PCR

Total RNA was extracted from cultured cells or tissue (liver and spleen) using the RNeasy Mini Kit (QIAGEN, Hilden, Germany), and RNA concentration was measured with a NanoDrop 2000 spectrophotometer (ThermoFisher Scientific, Delaware, USA). Complementary DNA (cDNA) was synthesized using the QuantiTect Reverse Transcription Kit (QIAGEN). Quantitative PCR was performed in triplicate using iTaq Universal SYBR Green Supermix (Bio-Rad) on a QuantStudio 7 Real-Time Thermal Cycler (Thermo Fisher Scientific, CA, USA). Target genes include Sptlc2, Degs1, Degs2, Cers2, Cers6, Smpd1, Smpd2, Smpd3, Cd25, Cd60, Cd86, Igm, Mhc-II, Ox40, TNF-α, IL-1β, Ccl2, Col1a1, Col3a1, Timp-1, and Tgf-β. Gapdh was used as the housekeeping gene, and mRNA expression levels were normalized to control.

Western blot

Proteins were extracted from liver tissue using radioimmunoprecipitation assay buffer supplemented with protease and phosphatase inhibitors (Thermo Fisher Scientific, IL, USA). Following reduction with 1,4-dithiothreitol, 15 μg of total protein per sample was separated by SDS–polyacrylamide gel electrophoresis and transferred onto a polyvinylidene difluoride membrane using the iBlot 2 Gel Transfer Device (Invitrogen, CA, USA). Membranes were blocked for 1 hour at room temperature in TBS-T containing 5% Blotting Grade Blocker (Bio-Rad). Subsequently, membranes were incubated overnight at 4°C with primary antibodies, except for glyceraldehyde-3-phosphate dehydrogenase, which was incubated for 1 hour at room temperature. After washing, membranes were incubated with HRP-conjugated secondary antibodies for 1 hour at room temperature. Signals were detected using Immobilon Western Chemiluminescent HRP Substrate (Millipore), and images were acquired with the Bio-Rad ChemiDoc Touch Imaging System.

LC-MS/MS for lipidomics

The procedures for sample preparation, lipid extraction and quantification, and pooled quality control (QC) were described previously (77). In brief, frozen liver tissue (10 mg) was homogenized with 5-mm stainless steel beads (QIAGEN, Hilden, Germany) in 100 μl of PBS on ice using a TissueLyser LT (QIAGEN) at 25 oscillations/s for 2 min, repeated three times. Protein concentrations were determined using the Pierce Detergent Compatible Bradford Assay (Thermo Fisher Scientific), following the manufacturer’s protocol. All samples were stored at −80°C until further analysis.

For lipid extraction and quantification, 10 μl of tissue homogenate or 10 μl of plasma were mixed with 100 μl of butanol/methanol (1:1, v:v; Merck Millipore, Massachusetts, USA) spiked with the following internal standards: 400 nM PE 17:0/17:0, 2.4 nM Cer d18:1/8:0, 200 nM PC 13:0/13:0, 1.7 nM HexCer d18:1/8:0, 400 nM LPC 13:0, 400 nM LPE 14:0, 400 nM PS 17:0/17:0, 3.6 nM SM d18:1/6:0, 0.8 nM SPH d18:1 d7, and 0.8 nM SPH d18:0 d7 from Avanti Polar Lipids and 400 nM triglycerides (TG) 12:0/12:0/12:0 from Sigma-Aldrich. Samples were sonicated for 30 min followed by centrifugation at 14,000g for 10 min at 22°C. The supernatant was used for analysis. The samples were analyzed using an Agilent 1290 series UHPLC system connected to an Agilent 6495 QQQ mass spectrometer after separation on a ZORBAX Eclipse plus C18 column (2.1 mm × 50 mm, 1.8 μm, 95 Å, Agilent) at 40°C. The injection volume was 2 μl. Solvent A consisted of 60% water/40% acetonitrile (v/v) with 10 mM ammonium formate; solvent B consisted of 90% isopropanol/10% acetonitrile (v/v) with 10 mM ammonium formate. The gradient started with a flow rate of 0.4 ml/min at 20% B and increased to 60% B at 2 min, 100% B at 7 min, held at 100% B until 9 min, followed by equilibration with 20% B from 9.01 min until 10.8 min. The column effluent was introduced to the mass spectrometer via an AJS-ESI ion source operating under the following conditions: gas temperature, 200°C; gas flow, 14 liter/min; nebulizer, 20 psi; sheath gas temperature, 250°C; sheath gas flow, 11 liter/min; capillary, 3500 V. Mass spectrometry analysis was performed in positive ion mode with dynamic multiple reaction monitoring (dMRM). Mass spectrometry settings, liquid chromatography–mass spectrometry (LC-MS) gradient, and MRM transitions for each lipid class were adapted from a published method (78). Relative quantification of lipids was based on one-point calibration with class-specific internal standards and was further normalized to the total protein content within each sample. Liver ceramides were presented as fold changes relative to ceramide levels in healthy control mice, and the relative concentrations of each detected ceramide species to a single internal standard for each lipid class were provided for Figs. 2G, 5C, and 6C and fig. S8C in data S1.

Pooled QC samples were prepared and analyzed every 10 samples. The average signal-to-noise (S/N) ratio and the coefficient of variation (CV) of each individual lipid in the QC samples were then calculated. Lipids with S/N ratio below 10 and CV higher than 30% were excluded. A dilution series of QC samples were prepared and measured at the start and end of the experiment, respectively. The Pearson correlation test was used to assess the linearity of the dilution curves. Lipid peaks with R² lower than 0.8 were excluded.

Statistical analysis

Statistical analyses for LC-MS/MS data were conducted with R version 4.0.2. Pearson’s correlation coefficients were calculated for all pairs of lipid species. Statistical significance was assessed with one-way analysis of variance (ANOVA), followed by Tukey’s post hoc test. To control for false discovery rate, P values were corrected for multiple comparisons using the Benjamini-Hochberg procedure. All other experimental data were analyzed using unpaired Student’s t test or one-way ANOVA, followed by Tukey’s post hoc test for multiple comparisons, unless otherwise stated, with GraphPad Prism version 8.4.0 (GraphPad Software, CA, USA). Semi-quantitative densitometry of Western blots was performed using Fiji (ImageJ) and GraphPad Prism. All data are presented as means ± SEM unless otherwise stated. P value < 0.05 was considered statistically significant.

Supplementary Materials

The PDF file includes:

Figs. S1 to S12

Legend for data S1

Other Supplementary Material for this manuscript includes the following:

Data S1