Liver tissues were obtained intraoperatively from patients undergoing orthotopic liver transplantation or surgical liver resection for primary biliary cirrhosis, primary sclerosing cholangitis, and HBV-related cirrhosis. Demographic and clinical characteristics, laboratory indices, and disease statuses of the patients are shown in Supplementary Table S1. The distance from the obtained liver tissue to the edge of the lesion was at least 5 cm. Written informed consent was obtained from the patients for use of their tissues for research purposes, according to the ethical guidelines of the First Hospital of Jilin University (NO. 2019-356). Subsequently, primary human HSCs were isolated from the wedge sections of human livers using the density gradient centrifugation method (Werner et al., 2015). In brief, collagenase IV (Sigma-Aldrich, St. Louis, MO, United States) was used to perfuse and digest the liver tissue, and a hepatic cell suspension was obtained after blunt separation. Primary hepatocytes (HCs) were separated after centrifugation at 50 g for 5 min, and the supernatant was further centrifuged at 500 g for 5 min at 4°C. Using 8.5% Optiprep gradient medium (Stemcell, Vancouver, Canada), we removed other non-parenchymal cells, including liver sinusoidal endothelial cells and Kupffer cells. Primary HSCs, and the human immortalized hepatic stellate cell line (LX-2) (kindly provided by Dr. Zhengkun Tu) were maintained in Dulbecco’s modified eagle’s medium (DMEM; Gibco, Waltham, MA, United States) supplemented with 10% fetal bovine serum (FBS; Gibco). For TGFβ1-induced activation, LX-2 cells were treated with TGF-β1 (5 ng/μL; R&D, United States) for 24 h after starvation. The cells were then incubated in a 5% CO₂ incubator at 37°C.

HSCs were fixed with 4% paraformaldehyde (Solarbio, Beijing, China) for 15 min at room temperature (25°C), washed with phosphate-buffered saline (PBS), and subsequently blocked with 2% bovine serum albumin for 30 min. HSCs were then incubated for 2 h at 37°C with a primary monoclonal anti-α-SMA antibody (1:200 dilution; Abcam, Cambridge, MA). After washing with PBS, the cells were incubated for 1 h at room temperature with a secondary polyclonal goat anti-rabbit IgG (H + L; 1:200; Earthox, San Francisco, United States). The negative control was obtained by not using the primary antibodies. After incubation with the above antibodies, cells were washed with PBS and counterstained with 4,6-diamidino-2-phenylindole (DAPI; Invitrogen, Carlsbad, CA, United States) for 3 min. Immunofluorescence staining was detected and photographed using a laser scanning microscope (Axiovert 100M; Zeiss, Jena, Germany) at 200× magnification.

Mouse liver tissue sections were cut into 3 μm silces, and embedded in paraffin. Collagen deposition in liver tissue sections was localized using standard histological techniques with Masson’s trichrome staining. Each section was assessed under a light microscopic and photographed at 40× magnification.

Male C57BL/6 mice (6 weeks old) were purchased from Charles River (Beijing, China). All mice were fed a standard rat chow diet and housed under a 12 h light/dark cycle. After acclimatization for 7 days, the mice were randomly divided into four groups: negative control (NC; n = 6), CCl₄ group (n = 6), CCl₄+AAV-NC group (n = 7), and CCl₄+AAV-miR-654-5p group (n = 7). Adeno-associated virus serotype 8 (AAV8) particles encoding miR-654-5p (hereafter referred to as AAV- miR-654-5p) and control AAV particles (AAV-NC) were purchased from Hanbio, Shanghai, China, and were administered to the CCl4+AAV-miR-654-5p and CCl4+AAV-NC groups at a dose of 3 × 10¹¹ viral genomes (vg) per animal via tail vein injection. After 1 week, the mice in the CCl₄, CCl₄+AAV-NC, and CCl₄+AAV-miR-654-5p groups were intraperitoneally injected with a 10% CCl4 (Aladdin, China) dose at 1 ml/kg (diluted with edible olive oil before injection) three times per week. Mice in the NC group were similarly administered the same solvent. After 6 weeks, the mice were sacrificed, and their liver tissues were dissected. Blood samples were centrifuged, and serum was stored at −80°C. Additionally, a portion of the liver tissue samples were fixed in 4% formaldehyde, while the remaining sample was stored at −80°C until use. All experiments involving mice were conducted in accordance with the ethical guidelines of the Animal Ethics Committee of First Hospital of Jilin University (Approval NO. 20220002).

Cells were transfected with 50 nM miR-654-5p mimics or mimics-NC (RiboBio, Guangzhou, China) using Lipofectamine 3,000 (Invitrogen) following the manufacturer’s protocol. Similarly, cells were transfected with pcDNA3.1-RXRα plasmid or its control plasmid pcDNA3.1-NC (Sangon Biotech).

Cells were lysed in RIPA buffer (Beyotime, Shanghai, China) containing PMSF (Solarbio). This assay was performed using standard western blotting techniques with the following primary antibodies: anti-RXRα (Abcam), anti-collagen I (Proteintech, Chicago, United States), anti-MMP2 (Proteintech) and anti-tubulin (YTHX Biotechnology, Beijing, China).

Total RNA was extracted from the cells/liver tissues using Eastep™ Total RNA Extraction Kit (Promega, Madison, United States) and miRNAs were extracted from the cells/liver tissues using the EasyPure miRNA Kit (TransGen, Beijing, China) following the manufacturer’s instructions. Complementary DNA (cDNA) was synthesized using TransScript One-Step gDNA Removal and cDNA Synthesis SuperMix (TransGen) to detect mRNA. cDNA was generated using a Ribo SCRIPT™ Reverse Transcription kit (RiboBio) to detect miRNA. Quantitative real-time polymerase chain reaction (qRT-PCR) was performed using PerfectStart™ Green qPCR SuperMix (TransGen). β-actin was used as an mRNA control. U6 was used as a reference miRNA control. The primers for U6/miR-654-5p were obtained from RiboBio Co., Ltd. All other qPCR primers used are listed in Table 1 qPCR was performed using the Agilent Mx3005P Real-Time PCR System (Applied Biosystems, Foster City, CA).

The wild-type (WT) or mutant (MUT) RXRα 3ʹ UTR was synthesized and subcloned into the pmirGLO Dual-Luciferase miRNA Target Expression Vector (Promega). HEK293 cells were co-transfected with miR-654-5p mimics and pGLO‐WT‐RXRα or pGLO‐MUT‐RXRα using Lipofectamine 3,000. Luciferase activity was measured 48 h after the co-transfection using the Dual-Luciferase Reporter Assay System (Promega) following the manufacturer’s instructions. Relative luciferase activity was calculated by normalizing firefly luciferase activity to Renilla luciferase activity.

LX-2 cells were seeded into 96-well plates (3,000 cells/well) and cultured in serum-free DMEM after transfection; the medium was replaced every 48 h. CCK-8 solution (10 uL; Beyotime, Shanghai, China) was added to each well at 24, 48, 72, 96, 120, 144, and 168 h after transfection, respectively. Absorbance was measured at 450 nm using a microplate reader (Thermo Fisher Scientific, Waltham, MA, United States).

To detect cell apoptosis, LX-2 cells were cultured in serum-free DMEM for 5 days after transfection to induce apoptosis by starvation. Trypsin was used to digest the cells for flow cytometry analysis. Cell apoptosis was measured by staining cells with PE Annexin V and 7-AAD using a PE Annexin V Apoptosis Detection Kit (BD Biosciences, Franklin Lakes, NJ, United States) for 15 min at room temperature in the dark. Apoptosis was detected using a FACS Canto flow cytometer (BD).

To detect the purity of isolated HSCs, freshly isolated HSCs were washed and counted They were then incubated with CD68 and CD146 antibodies (BD Biosciences, San Jose, CA) in the dark at room temperature for 30 min. In addition, other HSCs were cultured in vitro for 14 days and then collected. After washing, cells were fixed and permeabilized with the BD Cytofix/Cytoperm™ Fixation/Permeabilization kit (BD) according to the manufacturer’s introductions. Fixed cells were further incubated with an α-SMA antibody (R&D). After washing, labeled cells were resuspended and analyzed by flow cytometry. Flowjo was used to analyze the flow data.

The levels of serum alanine aminotransferase (ALT), aspartate aminotransferase (AST) and hydroxyproline (Hyp) were measured in the mice using ALT, AST and Hyp measuring reagent kits (Nanjingjiancheng, Nanjing, China) according to the manufacturer’s instructions.

Data are presented as mean ± standard deviation (SD) of at least three independent experiments. The paired t-test was used to analyze the differences in mRNA or miRNA expression from qRT-PCR results, and *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001 from Prism 8.0.1 software (GraphPad Software, San Diego, CA) were defined as statistically significant.

Although there are no known specific markers for quiescent human HSCs, these cells store several cytoplasmic retinoid droplets rich in vitamin A. Therefore, we detected spontaneous fluorescence of these lipid droplets in freshly separated (1 d) HSCs. HSCs were cultured in plastic dishes for 14 days to induce spontaneous cellular activation. The activated HSCs underwent morphological changes and expressed α-SMA (Figures 1A,B). Moreover, since HSCs are similar in size to liver endothelial cells and macrophages, it is likely that the layer obtained following density gradient centrifugation contained all three of these cell subsets. Therefore, we identified liver endothelial cells via CD146 and macrophages via CD68. The remaining CD68^– CD146^– cells were considered to be HSCs. The results showed that the purity of freshly isolated HSCs was >90% (Figure 1C). Subsequently, we labeled HSCs with α-SMA and analyzed them via flow cytometry, confirming that the purity of isolated HSCs was >90% (Figure 1D).

To confirm the spontaneous activation of HSCs following in vitro culture, the expression levels of activation-related genes in HSCs were measured. After 14 days of in vitro culture, the mRNA expression levels of collagen type 1-α1 (col1α1), α-SMA, and matrix metalloproteinase 2 (MMP2) were significantly upregulated compared to those at on day 1 (Figure 2A). Similarly, after 48 h of TGFβ1 stimulation, the mRNA expression levels of col1α1 and MMP2 in LX-2 cells were significantly upregulated, while that of α-SMA decreased (Figure 2B). After in vitro culturing for 14 days, miR-654-5p and RXRα expression was evaluated in activated primary HSCs. Our results showed that the relative expression of miR-654-5p was significantly increased in activated HSCs compared to quiescent HSCs (freshly isolated; p < 0.05), while the relative expression of RXRα decreased (p < 0.001; Figure 2C). In addition, we treated LX-2 cells with TGF-β1, an HSCs activator. A similar trend was observed in TGF-β-induced LX-2 cells compared to the NC group (p < 0.05 and p < 0.001, respectively; Figure 2D).

To explore the function of miR‐654‐5p in LF, we modulated the expression of miR‐654‐5p by transfecting LX-2 cells with mimics. First, we overexpressed miR‐654‐5p in LX-2 cells by transfection with miR-654-5p mimics (Figure 3A) and found that overexpression of miR‐654‐5p promoted the mRNA expression levels of HSC activation markers col1α1 and MMP2 compared to the control group (mimics-NC; Figure 3B; p < 0.01 and p < 0.05). Subsequently, LX-2 cells were treated with TGFβ1 after miR-654-5p mimics/mimics-NC transfection. The resulting expression levels of col1α1 and MMP2 proteins were increased in miR-654-5p-overexpressing LX-2 cells (Figure 3C). These results validated the functional relevance of miR‐654‐5p in the activation of HSCs in vitro.

LX-2 cells transfected with the miR-654-5p mimic were then starved in an FBS-free medium for 5 days. The resulting quantity and morphology of LX-2 cells differed from those of the NC group (Figure 3D). Moreover, the CCK‐8 assay results showed that overexpression of miR‐654‐5p significantly promoted the proliferation of LX-2 cells (Figure 3E).

In addition, flow cytometry analysis results suggested that LX-2 cells treated with the miR‐654‐5p mimic exhibited changes in apoptosis compared to mimic-NC. Specifically, three repeated experiments showed that Annexin V-positive cells were significantly increased after miR-654-5p transfection; therefore, miR-654-5p reduced early apoptosis of LX-2 cells (Figures 3F,G). Overall, these results indicate that upregulation of miR‐654‐5p promotes the activation and proliferation of HSCs while inhibiting apoptosis.

To identify the potential mRNAs targeted by miR-654-5p, bioinformatics analysis using TargetScan (http://www.targetscan.org/vert_71/) and miRwalk (http://mirwalk.umm.uni-heidelberg.de/) databases were carried out. Based on our lab RNA-seq data from the previous period, RXRα was noted to be a candidate involved in miR‐654‐5p regulation of HSCs. The speculative binding region between hsa‐miR‐654‐5p and RXRα, based on TargetScan results, is shown in Supplementary Figure S2. A dual-luciferase reporter assay was then used to validate the association between the two. To this end, luciferase reporter plasmids of WT‐RXRα- and MUT‐RXRα 3ʹ UTR were constructed and are shown in Figure 4A. Co-transfection of the luciferase reporter plasmid containing WT‐RXRα with miR‐654‐5p mimics in HEK‐293T cells decreased reporter activity. Conversely, the luciferase reporter plasmid co-transfection containing MUT‐RXRα with miR‐654‐5p mimics did not alter the luciferase activity (Figure 4B). In addition, the levels of both RXRα mRNA and protein were markedly inhibited by miR‐654‐5p upregulation in LX-2 cells (Figure 4C). These findings indicate that miR-654-5p directly targets RXRα.

To verify that miR-654-5p regulates the activation, proliferation and apoptosis of HSCs by targeting RXRα, we transfected LX-2 cells with pcDNA3.1-RXRα plasmid to ectopically overexpress RXRα and confirmed the elevated levels of RXRα (Figure 5A), LX-2 cells were then co-transfected with miR-654-5p mimics and pcDNA3.1-RXRα/pcDNA3.1-NC. Subsequently, the transfected cells were treated with TGFβ1 to induce col1α1 expression. The results of western blotting showed that the overexpression of RXRα suppress the miR-654-5p mimics-induced expression of col1α1 protein (Figure 5B).

The co-transfected cells were also obtained to assess apoptosis using flow cytometry analysis. Results showed that following co-transfection with pcDNA3.1-RXRα and miR‐654‐5p mimics, the proportion of LX-2 cells undergoing early apoptosis had increased compared to those transfected with pcDNA3.1-NC and miR‐654‐5p mimics (Figures 5C,D). Similarly, CCK-8 assay results showed a decline in the proliferation in cells co-transfected with pcDNA3.1-RXRα and miR‐654‐5p mimics (Figure 5E).

To further investigate the role of miR-654-5p in vivo, we established a CCl₄-induced LF mouse model. The mRNA levels of α-SMA, col1α1, and MMP2, were higher in the livers of CCl₄-treated mice than in those of control mice (Figure 6A). Masson’s trichrome staining further revealed increased ECM deposition in the livers of CCl4-treated mice compared to those of control mice. In addition, Hyp content was upregulated in the liver tissues of mice treated with CCl₄ (Figure 6B). These results confirmed that CCl₄ induced LF in the mouse model.

Subsequently, the levels of mmu-miR-654-5p and mmu-RXRα between the CCl₄ model and NC groups were estimated. Results showed that mmu-miR-654-5p expression was upregulated in fibrotic mouse, while mmu-RXRα mRNA level was consistently downregulated (Figure 6C). The targeting relationship between mmu-miR-654-5p and mmu-RXRα was then predicted using TargetScan (Supplementary Figure S3). To assess the delivery of AAV into the liver, the level of miR-654-5p was quantified in mouse livers by qRT-PCR analysis. Expression of miR-654-5p was significantly upregulated in the CCl₄+AAV-miR-654-5p group compared to that in the CCl₄+AAV-NC group. Consistent with this, the mRNA expression of RXRα was decreased (Figure 6D). In addition, serum ALT and AST levels were compared between the CCl₄+AAV-miR-654-5p and CCL4+AAV-NC groups. No significant differences were observed in AST and ALT content levels between the two groups (Figure 6E). Finally, we also observed a significant increase in fibrosis within AAV-miR-654-5p overexpressing livers, as evidenced by increased collagen deposition in the liver, detected via Masson’s trichrome staining, and increased Hyp levels (Figure 6F).