Ferulic acid (CAS: 1135-24-6, batch number: 110773-201614) was obtained from the National Institute for Food and Drug Control (Beijing, China). Triton WR-1339 was obtained from Sigma-Aldrich (Shanghai, China). Twenty-four male Kunming (KM) mice (18–22 g body weight) were purchased from the Guangdong Medical Laboratory Animal Center (Guangzhou, China) (Permit number: 44007200106576). All animal experiment procedures were performed strictly with the Laboratory Animal Guideline for Ethical Review of Animal Welfare (GB/T 35892-2018, China) and were approved by the Guangdong Provincial Engineering Technology Institute of Traditional Chinese Medicine (Guangzhou, China).

The standard conditions of temperature, humidity, and light were used to acclimatize all animals for a week. The KM mice were randomly divided into the following 4 groups of 6 mice in each group: control group (CG), model group (MG), fenofibrate group (FG, 26 mg/kg) (Li et al., 2020), and ferulic acid group (FA, 100 mg/kg) (Kesh et al., 2013). The administration groups were given corresponding drugs by gavage and once a day for 5 days. On the third day of administration, all groups were intramuscularly administered triton WR-1339 (480 mg/kg) to build an acute hyperlipidemia model excluding the CG. On the fifth day, after administrating for 1 h, all mice were anesthetized with isoflurane and sacrificed through inner canthus artery exsanguination. Blood samples were preserved at room temperature and centrifuged at 3,000 rpm at 4°C for 15 min to isolate the serum. Serum and liver were saved at −80°C until the next steps. The flow chart is shown in Figure 1.

The serum’s TC, TG, HDL-c, and LDL-c contents were detected using commercial reagent kits (Nanjing Jiancheng Bioengineering Institute, Nanjing, China) according to the manufacturer’s instructions.

Mice liver tissues were immobilized with 4% paraformaldehyde solution, then dehydrated, embedded in paraffin, and sliced into 4 μm thick sections for H&E staining, and the histomorphology of the livers was observed and recorded using a microscope.

Serum (100 μL) was mixed with 400 μL of pre-cooled acetonitrile-water mixture (1∶1, v/v) and vortexed for 30 s. The samples were incubated at −20°C for 1 h to precipitate the protein, then centrifuged for 15 min (4°C, 12,000 rpm), transfered the supernatant (400 μL) to a new 1.5 mL tube and blown dry with nitrogen. 100 μL of pre-cooled acetonitrile-methanol mixture (1∶1, v/v) was added into samples and vortexed for 30 s, centrifuged for 15 min (4°C, 12,000 rpm), and accurately absorbed 80 μL of the supernatant into the injection bottle for the next analysis.

Quality control (QC) samples were prepared by mixing 10 μL aliquots of each analyzed sample. To test the stability of UPLC-Q-TOF-MS/MS, QC samples were injected once every six samples during the metabolomics (Xiao et al., 2023).

The serum metabolomics analysis was carried out by Agilent 1,290 II coupled to AB SCIEX X500R Q-TOF-MS/MS. The chromatographic separations were achieved on a Waters UPLC BEH C18 column (2.1 mm × 100 mm, 1.7 μm), flow rate 0.3 mL/min, injector temperature 4°C, column temperature 35°C, injection volume 1 μL. Mobile phase: gradient elution of acetonitrile (A) - 0.1% formic acid water (B) (0–1 min, 2% A; 1–3 min, 2%–10% A; 3–7 min, 10%–40% A; 7–16 min, 40%–75% A; 16–20 min, 75%–98% A; 20–23 min, 98% A). MS was performed in positive and negative ion modes with electrospray ionization (ESI). The optimization source parameters were set as follows: ion voltage −4,500 V and +5,500 V, Gas1 55 psi, Gas2 55 psi, curtain gas 35 psi, de-clustering potential voltage 60 V, ion source temperature 500°C, collision energy 35 V, collision energy spread 15 V, full scan m/z 50-1,000.

Multivariate statistical analysis was conducted using SIMCA 14.1 software with unsupervised principal component analysis (PCA) and orthogonal partial least discriminant analysis (OPLS-DA) models, and the results of OPLS-DA needed to be further confirmed by 200 times permutation evaluations. Statistical analyses were then performed using the online tool Metaboanalyst 6.0 for univariate student t-test and finally for difference multiple (FC value) analysis. The metabolites with significantly different (Variable important in projection (VIP) > 1, p < 0.05, Fold change (FC) < 0.8, or FC > 1.2) between MG and CG and between FA and MG were screened for preliminary identification and analysis. Further analysis of these metabolites was performed to screen for significant differences between FA and MG metabolites. It showed a recovery trend compared with CG, which was screened as the differential metabolite for treating hyperlipidemia. Potential biomarkers were characterized according to the METLIN and Human Metabolome databases, and pathways of differential metabolites were analyzed using the online software MetaboAnalyst 6.0 (http://www.metaboanalyst.ca/) (Pang et al., 2024).

To further explore the hypolipidemic mechanism of FA on hyperlipidemia from upstream to downstream signals, MetaboAnalyst 6.0 was used to perform an integrated analysis of metabolomics and network pharmacology. By constructing a compound-target-metabolite (C-T-M) network, we identified core targets shared between metabolite-associated targets and potential therapeutic targets. These core targets represent key entities through which FA exerts its therapeutic effects in treating hyperlipidemia. The conjoint analysis for fenofibrate was the same as above.

We utilized molecular docking to elucidate the interactions between FA and core target proteins by predicting their binding modes and affinities. The crystal structure format files for both FA and core targets were downloaded from the PubChem database and the RCSB Protein Data Bank (PDB, https://www.rcsb.org/), respectively. Removed water molecules from core proteins with AutoDockTools 1.5.6 software, added nonpolar hydrogens, and calculated the Gasteiger charges. The ligands and receptors were docked in PDBQT format via AutoDock Vina (Trott and Olson, 2010).

Herein, the results were shown as the mean ± SD. Statistical analysis was carried out by one-way ANOVA, student’s t-test was applied to compare two groups, and significance was established by GraphPad Prim 9.0. Metabolomics data are normalized by MetaboAnalyst 6.0 and imported into SIMCA 14.1 for further steps. We used the following annotations to indicate statistical significance: ^# p < 0.05, ^## p < 0.01, *p < 0.05, and **p < 0.01.

We determined serum levels of total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-c), and low-density lipoprotein cholesterol (LDL-c) as depicted in Figure 2A. The serum TC, TG, and LDL-c levels were significantly elevated in the MG compared to the CG. However, both the FG and FA groups exhibited significant reductions in TG, TG, and LDL-c levels post-treatment. Conversely, HDL-c levels were significantly lower in the MG than in the CG, but markedly higher in both the FG and FA groups. These findings indicate that FA effectively mitigated dyslipidemia in hyperlipidemic mice.

As observed by H&E staining results, the hepatocytes of CG were well arranged with intact nuclei and no inflammatory cell infiltration or vacuolar lesions, while MG had disorganized hepatocytes with unclear cell borders, some missing nuclei, diffuse steatosis of hepatocytes, and increased vacuolar lesions and inflammatory infiltration. Compared with MG, FG and FA had well-arranged hepatocytes, significantly fewer fat vacuoles, and improved cell morphology, although a small amount of inflammatory infiltration was still present (Figure 2B).

QC samples are mixtures of each sample and are primarily used to assess the accuracy, stability, and reproducibility of experiments and to ensure the reliability of results. As shown in Figures 3A, B, the QC samples displayed good clustering in the PCA score plot, demonstrating that the preparation and experimental conditions of the samples were consistent and reliable, affirming the accuracy of the acquired data.

The PCA analysis (Figure 3) delineated distinct clusters for CG, MG, and FA in both positive and negative ion modes. Oral administration of FA effectively attenuated serum biochemical abnormalities induced by triton WR-1339 in hyperlipidemic mice (MG). Subsequently, OPLS-DA models were constructed for CG vs MG and MG vs FA comparisons, respectively. Figures 4A–H depict the OPLS-DA score plots along with results from the 200 permutation tests. Table 1 presents detailed model parameters including R²Y, Q², and CV-ANOVA (p-values), which assess the model’s goodness of fit and predictability (Yan et al., 2022). Figure 4 showed that the serum metabolomics variables were evaluated with a suitable fit degree and high prediction ability. Additionally, differential metabolites were identified using volcano plots (Figures 4I–L). A total of 31 differential metabolites were discerned based on criteria set at VIP >1, p-value <0.05, FC > 1.2, or FC < 0.8 (Table 2). These findings highlight specific metabolic changes associated with FA treatment in hyperlipidemia, providing valuable insights into its therapeutic effects.

To investigate potential FA biomarkers for the therapy of hyperlipidemia and associated metabolic pathways, MetaboAnalyst 6.0 was utilized. Ten relevant metabolic pathways were identified using metabolic pathway enrichment analysis. The metabolic pathways with the strongest correlations (raw p < 0.05 and pathway impact > 0.05) (Xiao et al., 2023) were as follows: linoleic acid metabolism, arachidonic acid metabolism, and ether lipid metabolism. The metabolic pathway showed that FA exerts its lipid-lowering effect mainly regulates lipid and amino acid metabolism.

The heatmap analysis visualized the differential metabolites in the three groups and revealed the correlation between them. Our results showed that 2 metabolites (L-acetylcarnitine and palmitoleic acid) were significantly downregulated in the MG compared to the CG. In addition, 29 metabolites were significantly upregulated in the MG compared to the CG, including 9-OxoODE, 12,13-EpOME, α-linolenic acid, hippuric acid, linoleic acid, etc. The levels of palmitic acid, α-linolenic acid, 5-HETE, N-oleoyl phenylalanine, LysoPC(18:0), N-acetylglutamic acid, janthitrem C, 12,13-EpOME, CE (16:0), L-methionine, arachidonic acid, and hippuric acid were significantly decreased in the CG, and the levels of these metabolites in FA were reversed and returned to normal or near-normal levels compared to MG. Therefore, they are considered potential biomarkers for the hypolipidemic effects of FA. Figure 5 and Table 2 showed the trend of biomarker levels in the three groups, which indicated a significant difference in the potential biomarkers between CG and MG, and FA could reverse these differences. Furthermore, metabolic pathways associated with these biomarkers were constructed based on the KEGG database (Figure 6). These analyses underscore FA’s ability to modulate specific metabolic pathways crucial for lipid regulation, offering mechanistic insights into its therapeutic potential for hyperlipidemia.

To further explore the hypolipidemic mechanism of FA on hyperlipidemia, we integrated metabolomics and network pharmacology to find core targets for treating hyperlipidemia with FA. We constructed a C-T-M network using Cytoscape 3.7.2, the results were shown in Figure 7G. A total of 18 were considered as core targets, including PTGS2, PPARA, MAPK8, CYP1A2, PARP1, NOS2, PTGS1, MMP2, ABCB1, STAT3, HMGCR, CYP1B1, MMP9, TLR4, ACLY, CTNNB1, APP, and RELA. Furthermore, arachidonic acid, palmitic acid, linoleic acid, ursolic acid, α-linolenic acid, and 5-HETE were recognized as potential biomarkers for FA treatment of hyperlipidemia. These findings were supported by KEGG pathway enrichment analysis, which highlighted FA’s role in modulating lipid and atherosclerosis-related pathways, consistent with our metabolomics results. This integrated approach underscores FA’s multifaceted therapeutic effects in addressing hyperlipidemia by regulating key metabolic pathways and targets, providing a comprehensive understanding of its therapeutic potential.

Moreover, we also constructed a C-T-M network of fenofibrate, the results were shown in Figure 7H. A total of 16 were considered as core targets, including ACACA, MDM2, SRC, NOS2, CASP3, MAPK14, ACE, AGT, ACHE, MMP9, MAOB, PPARG, CYP1A2, MMP2, SLC5A2, and MAPK1. Furthermore, citric acid, ornithine, L-glutamine, L-valine, D-sorbit, glycine, creatinine, phenylalanine, linoleic acid, and D-glucose were recognized as potential biomarkers for fenofibrate treatment of hyperlipidemia.

A comparative analysis of the combined effects of FA and fenofibrate revealed four common targets: CYP1A2, MMP2, MMP9, and NOS2. Furthermore, FA was found to influence the expression of the PPARA gene, while fenofibrate modulated the expression of the PPARG gene. These findings elucidate that FA targets PPARα, whereas fenofibrate targets PPARγ, to exert their lipid-lowering effects.

We explored the interaction between FA and core targets (PTGS2, PPARA, MAPK8, CYP1A2, PARP1, NOS2, PTGS1, MMP2, ABCB1, STAT3, HMGCR, CYP1B1, MMP9, TLR4, ACLY, CTNNB1, APP, and RELA) using molecular docking analysis. Using AutoDock Vina software, 18 docking results were obtained and showed low binding energies, indicating a strong affinity between the FA and core targets. It was worth noting that all core targets demonstrated high affinity binding to FA. The results were shown in Table 3, and all of them were less than −3 kcal/mol, and the lower binding energy indicates a more stable binding of FA to the target. The results suggested that FA can spontaneously bind to each target via hydrogen bonding, van der Waals forces, etc., indicating their important roles in the molecular mechanism of FA with hyperlipidemia.