Rosa roxburghii Tratt. fruits were obtained from Guizhou Saisi Rosa roxburghii Tratt Health Industry Co., Ltd. (Qiannan, Guizhou, China). A DNA extraction kit was purchased from Omega Bio-tek Co., Ltd. (Omega Bio-tek, Norcross, GA, USA). The library building kit was procured from Bioo Scientific Co., Ltd. (Bioo Scientific, Norcross, GA, USA). A sequencing kit was obtained from Illumina Co., Ltd. (Illumina, Norcross, GA, USA). The bicinchoninic acid (BCA) protein concentration determination kit was received from Beyotime Biological Reagent Company (Guangxi, China). Polyvinylidene fluoride membrane (PDVF) was sourced from Millipore Co., Ltd. (Millipore, Bedford, MA). AMPK monoclonal antibody, AMPK phosphorylated monoclonal antibody P-AMPK monoclonal antibody, SREBP-1c monoclonal antibody, FASN monoclonal antibody, and β-actin monoclonal antibody were supplied by Abcam Co., Ltd. (Abcam, Cambridge, USA). Horseradish peroxidase (HRP)-labeled goat anti-rabbit/mouse IgG secondary antibody was provided by Servicebio Co., Ltd. (Wuhan, China).

RFV was prepared as previously described (31). In brief, the fresh Rosa roxburghii Tratt. fruits were cleaned and beaten thoroughly sequentially, and water was added at a ratio of 1:3 (g:mL). The Rosa roxburghii Tratt. juice (RRJ) was obtained. The mixed fermentation starter (Lactobacillus plantarum, Acetobacter pasteurianus, and Saccharomyces cerevisiae at a ratio of 1:1:1) was activated with warm water and added at a mass ratio of 0.02%. Finally, at a constant temperature of 25–27°C, the fermentation broth was supplemented with 10% (m/m) sucrose for aerobic fermentation for up to 20 days. The fermentation broth was stirred every 3 days. The fermentation was completed with a solid content of 7–8%, and a pH value of approximately 4. RFV was obtained by filtration and used for subsequent experiments.

UPLC–MS was used to identify the compounds of RFV. The specific method has been previously described (32). According to the previous experimental results, 70 compounds were identified in RFV through UPLC–MS analysis, including 27 phenolic compounds, 14 terpenoids, 5 organic acids, 4 sugars, 3 lactones, 3 amino acids, 2 lipids, 1 vitamin, and 11 other compounds (32). Based on the active compounds in RFV, network pharmacology methods were employed to predict their targets. In brief, the network pharmacology methods primarily included the following steps: first, potential target genes for the compounds in RFV and NAFLD were obtained through the Swiss Target Prediction¹ and GeneCards databases (see text Footnote 1), respectively. The intersecting targets were screened using the Venny 2.1 online tool.² The intersecting targets were then imported into the STRING database to construct a protein–protein interaction (PPI) network.³ Cytoscape software was used for visualization. The CytoNCA plugin was used to select and visualize core targets. Subsequently, gene ontology (GO) and Kyoto encyclopedia of genes and genomes (KEGG) analyses were performed using the DAVID database.⁴ The results were visualized using the Wei Sheng Xin Cloud Platform.⁵ Finally, the compounds, targets, and pathways were integrated to construct a network relationship diagram, revealing the mechanism by which RFV improves NAFLD.

The spatial data files (SDF format) of the active compounds were retrieved from the PubChem database.⁶ Three-dimensional structures of the core target proteins were obtained from the protein data bank (PDB, https://www.rcsb.org/). Active compounds and target proteins were prepared through dehydration, hydrogenation, and structural refinement using PyMOL. PyMOL software was utilized to visualize and analyze the protein–ligand interactions. Binding energy values were used to evaluate molecular docking results.

Experimental grouping: Animal studies were performed in accordance with the Chinese regulations on the management of laboratory animals (State Council Decree No. 676, 2017) and the animal research: reporting of in vivo experiments (ARRIVE) guidelines. All protocols were approved by the Animal Ethics Committee of Guizhou Medical University (SYXK (Gui) 2023–0002). Forty C57BL/6 J male mice (8-week-old, 20–22 g) were obtained from Beijing Xiao Shu You Tai Biotechnology Co., Ltd. (Beijing, China). These mice were housed in a standard animal facility maintained at 23–25°C with 55–60% humidity. They underwent an adaptive feeding period for 1 week. Subsequently, 40 mice were randomly divided into five groups (n = 8): a normal diet (Control) group, a high-fat diet (NAFLD) group, an RRJ group, a low-dose RFV (RFV-L) group, and a high-dose RFV (RFV-H) group.

One week later, a NAFLD model was established based on the method described in this reference (33). Except for the Control group, the mice in the other groups were fed a high-fat diet for 12 weeks. During the modeling period, mice in the RRJ, low, and high-dose RFV groups were intragastrically administered with RRJ and RFV, respectively. The Control and NAFLD group received saline. All treatments were administered at 0.01 mL/10 g body weight, during which the body weight of mice was recorded every 5 days. Based on the previous studies (34) and results of our pre-experiments, RRJ was diluted 2-fold (50.0%, v/v), and RFV was diluted 8-fold and 4-fold to prepare as low-dose RFV (12.5%, v/v) and high-dose RFV (25.0%, v/v), respectively.

Total DNA was extracted using the E.Z.N.A.® Soil DNA Kit (Omega Bio-tek, Norcross, GA, USA). The concentration and purity were determined using NanoDrop 2000 (Thermo Fisher Scientific, San Jose, CA, USA). DNA integrity was assessed using 1% agarose gel electrophoresis at a voltage of 5 V/cm for 20 min (Biowest Co., Ltd., Spain). The Illumina MiSeq platform (Microeco Tech Co., Ltd., Shenzhen, China) was used to amplify and sequence the highly variable region of the 16S rRNA gene V3–V4 using the following primers: 338F: ACTCCTACGGGAGGCAGCAG and 806R: GGACTACHVGGGTWTCTAAT.

Chromatographic conditions: The chromatographic column was ACQUITY UPLC® HSS T3 (100 × 2.1 mm, 1.8 μm; Waters, Milford, USA). The parameters were set as follows: mobile phase A consisted of 95% water and 5% acetonitrile (containing 0.1% formic acid), and mobile phase B consisted of 47.5% acetonitrile, 47.5% isopropyl alcohol, and 5% water (containing 0.1% formic acid). The injection volume was 3 μL, and the column temperature was 40°C.

Mass spectral conditions: Electrospray was used to ionize the sample. The spectra were obtained using electrospray ionization with both positive and negative ionization modes, using a Q-Exactive Focus (Thermo Scientific, Waltham, MA, USA). The parameters were set as follows: The negative ion spray voltage and positive ion spray voltage were set to 3.50 kV, the sheath gas was 50 arb, the auxiliary gas was 13 arb, the scanning range was 70–1,050, the capillary temperature was 325°C, the resolution was 3,500, and collision voltage was set to 40 eV.

The expression levels of AMPK, P-AMPK, SREBP-1c, and FASN in the livers of each group were detected by Western blotting. Proteins were extracted from the protein lysate and denatured at 100°C for 10 min. Protein samples were subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), and the proteins were transferred to polyvinylidene fluoride (PVDF) membranes, which were then blocked with a 5% skimmed milk solution. Subsequently, the membranes were incubated overnight at 4°C with rabbit anti-AMPK, rabbit anti-P-AMPK, rabbit anti-SREBP-1c, rabbit anti-FASN, and rabbit anti-β-actin antibodies at a dilution of 1:1000. Finally, the membranes were incubated with the corresponding anti-mouse secondary antibody (1:3000) at room temperature for 1 h. The signal was detected using a chemiluminescent imaging system (XRS; Bio-Rad, Fort Hershey, USA). Protein expression was determined by grayscale analysis of the target protein bands using ImageJ software. The relative expression levels of SREBP-1c and FASN were expressed as the gray ratio of SREBP-1c, FASN, and β-actin, and the relative expression level of P-AMPK was expressed as the gray ratio of P-AMPK and AMPK.

GraphPad Prism (version 5) and Origin (version 2022) software were used for mapping analysis. The t-test was used for comparisons between two groups. If p < 0.05, the difference between the two groups was considered statistically significant.

The compounds in RFV were identified as previously described. A total of 70 compounds were identified in RFV through UPLC-MS analysis, including 27 phenolic compounds, 14 terpenoids, 5 organic acids, 4 sugars, 3 lactones, 3 amino acids, 2 lipids, 1 vitamin, and 11 other compounds. Based on these findings, network pharmacology was used to predict the targets and pathways by which RFV regulates NAFLD. A total of 1,816 potential targets of RFV were identified using the ITCM database (with a probability > 0). Simultaneously, 1,399 targets associated with NAFLD were identified using the GeneCards database. Among 1,399 targets, 377 with a relevance score > 0.88 (twice the median relevance score) were selected for further analysis. A Venn diagram indicated that 44 common targets were identified across these datasets (Figure 1A). Using the CytoNCA plugin, a PPI network was constructed for the top 20 target genes based on their degree (Figure 1B). Protein nodes, such as PPARG, TNF, IL6, AKT1, ALB, and FASN, exhibited higher connectivity, indicating their central roles within the network.

The 20 intersecting targets were analyzed for GO functional annotation and KEGG pathway using the DAVID (version 6.8.0) database. The results were visualized on the Wei Sheng Xin Cloud Platform. GO results indicated that biological process (BP) included inflammatory responses, fatty acid metabolism, and liver regeneration (Figure 1C). The proteins encoded by the intersecting genes were primarily located in the nucleus, cytoplasm, and nucleoplasm. Molecular function (MF) primarily involves the binding of proteins, enzymes, and lipids. The results of KEGG pathway analysis are illustrated in Figure 1D. The following 11 relevant pathways were selected: lipid and atherosclerosis pathway, AGE-RAGE signaling pathway, AMPK pathway, C-type lectin receptor pathway, insulin resistance, HIF − 1 signaling pathway, toll-like receptor signaling pathway, TNF signaling pathway, alcoholic liver disease, NAFLD, and pathways in cancer. A compound-gene-pathway network diagram was created to visually illustrate the intricate relationships among compounds, targets, and pathways (Figure 1E). The lines in the diagram reflect the relationships among compounds, targets, and pathways. The network diagram further revealed that the active compounds of RFV might improve NAFLD through multiple targets and pathways.

Molecular docking results revealed that the key active compounds (quercetin, ellagic acid, and gallic acid) were molecularly docked with core targets associated with the AMPK pathway (AMPK and FASN). Molecular docking results indicated that all three compounds could dock with AMPK and FASN proteins (Figures 1F,G). They all exhibited binding energies of <−5 kcal/mol, indicating a strong binding affinity. The AMPK pathway may be an important target in NAFLD regulation. This was experimentally validated below.

The samples were analyzed using Illumina MiSeq high-throughput sequencing to investigate the effect of RFV on the intestinal microbiota composition (n = 6). The Venn diagram illustrates the total and overlapping numbers of operational taxonomic units (OTUs) in each group (Figure 2A). Control, NAFLD, RRJ, RFV-L, and RFV-H groups had 755, 465, 428, 404, and 490 OTUs, respectively. All five groups shared 152 OTUs, indicating the presence of at least 152 symbiotic bacteria common to each group. Furthermore, the data revealed that the Control group had 453 unique OTUs, the NAFLD group had 39 unique OTUs, and RFV-L and RFV-H groups had 39 and 53 unique OTUs, respectively. Alpha diversity estimation analyses revealed that the Chao1 indices (391) in the NAFLD group decreased significantly (p < 0.001) compared to the Control group, whereas the Simpson indices (0.08) did not indicate any significant change. After administration of high-dose RFV, Chao1 indices (465) and Simpson indices (0.30) increased significantly (p < 0.05, p < 0.001) (Figures 2C,D).

The microbial composition significantly changed after HFD and RFV intervention. There was a clear separation between the Control and NAFLD groups. Each administration group was significantly different from the NAFLD group (Figure 2B). The above results indicate that an HFD can alter the structure of the intestinal microbiota in mice, and RFV can improve the situation to a certain extent; therefore, the microbiota structure is closer to that of the Control group. The composition of the intestinal microbiota was analyzed at the phylum and genus levels (Figures 3A,B) and (Supplementary Figures 1A,B). At the phylum level, Firmicutes, Bacteroidetes, and Actinobacteria were dominant in each group. The relative abundances of Firmicutes in Control, NAFLD, RRJ, RFV-L, and RFV-H groups were 55.71, 0.57, 0.32, 1.44, and 4.86%, respectively. The relative abundances of Bacteroidetes were 55.71, 0.57, 0.32, 1.44, and 4.86%, respectively. The NAFLD group exhibited a significant increase in Firmicutes and a significant decrease in Bacteroidetes compared to the Control group (p < 0.001) (Figure 3C). As a result, the ratio of Firmicutes to Bacteroidetes (F/B) was highest in the NAFLD group. After the administration of high-dose RFV, the relative abundance of Firmicutes decreased significantly (p < 0.01), Bacteroidetes increased, and F/B decreased significantly (p < 0.01) (Figure 3D), indicating that RFV could improve the imbalance in microbial abundance caused by NAFLD, and the community composition was reversed to be similar to that of the Control group. Compared to the Control group, the relative abundance of Allobacterium, Faecalibaculum, Coriobacteriaceae_UCG-002, Dubosiella, Blautia, and unclassified_f_Lachnospiraceae increased in the NAFLD group (p < 0.05). Compared to the NAFLD group, RFV-H significantly reduced the relative abundance of Allobaculum, Faecalibaculum, Dubosiella, Blautia, Coriobacteriaceae_UCG-002, and unclassified_f_Lachnospiraceae (Figures 3E-J). Overall, RFV reversed intestinal microbiota disruption caused by NAFLD.

Partial least squares discriminant analysis (PLS-DA) demonstrated that liver metabolites were significantly different due to HFD and RFV intervention. The induction of NAFLD significantly changed liver metabolites. After the administration of RRJ and RFV, the metabolite composition was different from that of the NAFLD group and closer to the Control group (Figure 4A). A replacement test was performed using the PLS-DA model (R²Y = 0.64, Q² = 0.256, p < 0.05), indicating that the model was stable and reliable (Figure 4B).

Enrichment analysis of differential metabolites was performed using metaboAnalyst enrichment analysis. A total of 490 different metabolites were screened among the five groups (p < 0.05). These metabolites included 191 lipids and lipid-like molecules, 84 organic acids and their derivatives, 55 organic heterocyclic compounds, 31 benzenoid compounds, 29 organic oxygen compounds, 22 phenylpropanoids and polyketones, 10 nucleosides, nucleotides, and analogs, 7 alkaloids and their derivatives, 5 organic nitrogen compounds, 5 organic sulfur compounds, 2 hydrocarbon derivatives, 1 hydrocarbon compound, 1 lignin, new signaling, and related compounds, 1 organic polymer, and 13 other metabolites (Figure 4C).

There were 441 and 47 differential metabolites in the Control versus NAFLD group and NAFLD versus RFV-H group (p < 0.05, VIP value > 1, Fold change = 1), respectively. In both positive and negative ion modes, the volcano plot revealed that 315 metabolites were upregulated and 26 metabolites were downregulated in the Control versus NAFLD groups, whereas 30 metabolites were upregulated and 17 metabolites were downregulated in the NAFLD versus RFV-H groups (Figures 4D,E).

Metabolic sets were created for metabolites of the two comparison groups. KEGG topology analysis was used to examine the pathways associated with the metabolites (Figures 5A,B) and subsequently screened for key pathways based on p-values and impact values (Table 1). Except for metabolic pathways with an impact of 0, 21 metabolic pathways were involved between the Control and NAFLD groups. Arachidonic acid metabolism, glycerophospholipid metabolism, tryptophan metabolism, and the biosynthesis pathways of ubiquinone and other terpenoid quinones were the most significant. The impact values were 0.320, 0.314, 0.216, and 0.007, respectively, with p-values < 0.05.

In the RFV-H group, RFV-H significantly regulated NAFLD through multiple pathways, including valine, leucine, and isoleucine biosynthesis, as well as glycerophospholipid metabolism. The impact values were 0.107 and 0.094, with p < 0.05. The glycerophospholipid metabolic pathway was common to NAFLD and Control, RFV-H groups, suggesting that it is a potential target for RFV intervention in NAFLD. Therefore, the related metabolites are discussed in detail.

The glycerophospholipid metabolic pathway is illustrated in Figure 5C, where the red portion represents significantly different metabolites between the Control and NAFLD groups. The relative abundance of these metabolites was significantly regulated by RFV (Figure 5D) (n = 6). The high-fat diet significantly upregulated the levels of PCs, PS, and LysoPCs (p < 0.001). Following the intervention across various treatment groups, these metabolite levels varied during recovery, with RFV-H exhibiting the most significant effect (p < 0.01).

To investigate the correlation between intestinal microbiota and superoxide dismutase (SOD), triglyceride (TG), cholesterol (TC), malondialdehyde (MDA), high-density lipoprotein (HDL-C), and low-density lipoprotein (LDL-C) in NAFLD mice, species were selected for correlation network analysis based on Spearman’s correlation, |r| > 0.6, p < 0.05 (Figures 6A,B). The heatmap of the correlation analysis demonstrated that Firmicutes exhibited a significant positive correlation with MDA, TC, LDL-C, and TG and a negative correlation with SOD. Patescibacteria was negatively correlated with MDA (p < 0.05), TC, LDL-C, and TG (p < 0.05) and positively correlated with SOD (p < 0.05). Bacteroidetes were negatively correlated with MDA (p < 0.001), and Unclassifiedk_norank_d_Bacteria were positively correlated with SOD (p < 0.05). TC and LDL-C exhibited a positive correlation with Allobaculum, Faecalibaculum (p < 0.01, p < 0.05), and Blautia (p < 0.01). TC was positively correlated with Dubosiella (p < 0.01), whereas LDL-C was negatively correlated with norank_f_Muribaculaceae (p < 0.05). MDA levels were positively correlated with Collinsella, Romboutsia, Family_XIII_AD3011_group (p < 0.05), Faecalibaculum, unclassified_f_Lachnospiraceae, Allobaculum, Ruminococcus_torques_group (p < 0.01), and Blautia (p < 0.001), whereas they were negatively correlated with norank_f_Muribaculaceae (p < 0.01). SOD exhibited a negative correlation with Romboutsia, Faecalibaculum, Blautia (p < 0.01), unclassified_f_Lachnospiraceae (p < 0.05), and Allobaculum (p < 0.001) and a positive correlation with norank_f_ Muribaculaceae (p < 0.05). Consequently, a correlation was found between the intestinal microbiota, lipid metabolism, and oxidative stress levels in mice with NAFLD. The main bacterial genera were Allobaculum, Faecalibaculum, Dubosiella, Blautia, unclassified_f_Lachnospiraceae, and norank_f_ Muribaculaceae.

Spearman was used to analyze the correlation between the related bacteria and the differential metabolites affected by RFV (Figure 6C). Glycerophospholipids such as phosphatidylethanolamine: PE (22:5/0:0), phosphatidylcholine: PC (18:1(6Z)/0:0), PC (22:5/0:0), PC (18:2(9Z, 12Z)/18:0), PC (16:0/20:4(8Z, 11Z, 14Z, 17Z)), PC (20:4(8Z, 11Z, 14Z, 17Z)/15:0), lysophosphatidylcholine: LysoPC (22:5(4Z, 7Z, 10Z, 13Z, 16Z)/0:0), LysoPC (18:1(11Z)/0:0), LysoPC (20:4(5Z, 8Z, 11Z, 14Z)/0:0), and LysoPC (16:1(9Z)/0:0) had a significant negative correlation with Bacteroidetes (p < 0.05). Firmicutes had a significant positive correlation with LysoPC (15:5/0:0), LysoPC (18:1(11Z)/0:0), and PC (18:1(6Z)/0:0) (p < 0.05). Faecalibaculum had a significant positive correlation with PC (18:3/18:1), LysoPC (22:5(4Z, 7Z, 10Z, 13Z, 16Z)/0:0), and LysoPC (15:0/0:0) (p < 0.05). Blautia had a significant negative correlation with phenylalanine (p < 0.05), while Allobaculum had a significant positive correlation with LysoPC (18:1(11Z)/0:0) and PC (18:1(6Z)/0:0) (p < 0.05) (see Table 2).

To further investigate the effect of RFV on the AMPK/FASN/SREBP-1c signaling pathway, Western blotting was performed to measure the expression levels of AMPK, FASN, and SREBP-1c in the liver.

Western blotting results demonstrated that compared to the Control group, the expression of P-AMPK in the NAFLD group (n = 3) was significantly inhibited (p < 0.05), whereas the expressions of SREBP-1c and FASN were significantly increased (p < 0.05) (Figure 7). After administration of RFV, AMPK expression was activated, and P-AMPK was significantly increased (p < 0.05, p < 0.01). Conversely, SREBP-1c and FASN expression levels were significantly reduced. In summary, RFV inhibited the expression of SREBP-1c and FASN by activating P-AMPK in the AMPK signaling pathway, thereby reducing lipid accumulation and improving NAFLD.