Compounds in the three single herbs of LCS were collected from the Chemical Database of the Academy of Sciences (http://www.organchem.csdb.cn/scdb/default.asp), and manually supplemented through text mining (Yue et al., 2017; Zhao et al., 2022). Their chemical structures were obtained from the Chemical Book database (http://www.chemicalbook.com), NCBI PubChem database (http://www.ncbi.nlm.nih.gov/pccompound), and the literature, and drawn by using Chemdraw software (https://www.perkinelmer.com.cn/category/chemdraw, version 3.7.2). Compound formats were saved as mol2 and SMILES, respectively, for further analysis.

Based on the TCMSP database (https://lsp.nwu.edu.cn/gj.htm), ADME parameters such as oral bioavailability (OB) value, Caco-2 permeability, and drug-like index (DL) could be obtained (Ru et al., 2014). The screening thresholds were set as OB ≥ 30%, Caco-2 ≥ −0.14, and DL ≥ 0.1 (Li et al., 2016; Yue et al., 2017). Compounds containing glycosyl groups were further deglycosylated and expressed as compound_qt (Zhao et al., 2022).

Targets of candidate compounds were searched in the Drug Bank database, Similarity Ensemble Approach (SEA) database, STITCH database, and a reverse pharmacophore-based screening platform, Pharm Mapper database (Knox et al., 2011; Kuhn et al., 2014; Liu et al., 2015; Wang et al., 2017; Xu et al., 2018; Zhao et al., 2022). These predicted targets were chosen by the rank of matching score and then input into the Therapeutic Target Database (TTD), Online Mendelian Inheritance in Man (OMIM) database, Drug Bank database, and PharmGKB database (Altman, 2007; Wishart et al., 2008; Jiang et al., 2015; Liu et al., 2015). Combined with the literature, targets related to liver injuries as potential targets for next analysis were retained.

Database for Annotation, Visualization, and Integrated Discovery (DAVID) was used to perform Gene Ontology (GO) enrichment analysis of the potential targets of candidate compounds in LCS. Pathway enrichment was analyzed using pathway data associated with liver injuries obtained from the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway section (Kanehisa and Goto, 2000; Tang et al., 2016).

The candidate compound–potential target (cC-pT) network and potential target–pathway (pT-P) network were established by using Cytoscape (version 3.7.2), open software package for visualizing and analyzing interaction networks, to generate a bipartite network (Smoot et al., 2011).

The contribution index (CI) based on network-based efficacy (NE) and the literature was calculated to estimate the contribution of each candidate compound to the anti-liver injury effects of LCS by the following equations: N E ( j ) = ∑ i = 1 n d i , (1) C I ( j ) = c j × N E ( j ) ∑ i = 1 m c i × N E ( i ) × 100 % , (2) where n is the number of targets associated with ingredient j in the cC-pT network, d _i is the degree of target i associated with ingredient j in the pT-P network, c _i is the number of anti-liver injury-related literature of ingredient i, and m is the number of ingredients. For literature mining, liver injury together with the name of each candidate compound was used as keywords for information retrieval. Articles published in 1995–2022 were obtained.

Molecular docking was performed by using MOE software (Molecular Operating Environment, version 2020) to evaluate the interaction between targets and candidate compounds. To be specific, the X-ray crystal structures of target proteins were downloaded from the Protein Data Bank (www.rcsb.org/). The energy of the enzyme was minimized by deleting the water molecules and the ligands used for crystallization. The 2D structures of candidate compounds (ligand molecules) were drawn in ChemDraw software (version 15.0) and saved in the 3D SDF format. The database was created by transferring it to the MOE system, and then saved as an .mdb file extension. Then the surface of the macromolecule was scanned, and optimization processes were performed to find the active sites on the enzyme. Through generating 30 most stable conformers and electrostatic and van der Waals interactions, the grid scores were calculated to obtain the optimal spatial conformations. All parameters were set to default. Fit values were obtained and ranked to evaluate the possibility of binding of candidate compounds to targets.

Liver tissue samples from normal control, model, and high-dose LCS groups were homogenized in ninefold cold normal saline (w/v). The liver samples (each weighing 50 mg) were deproteinized with 1,000 μL methanol and vortexed for 30 s. Then the mixed solutions were centrifuged at 12,000 rpm at 4°C for 15 min. Afterward, supernatants (each 500 μL) were transferred into new EP tubes, and they were evaporated to dryness at 37°C using a vacuum concentrator. The residues were re-dissolved with 100 μL of extraction solution acetonitrile–water (1:1, v/v), vortexed for 3 min, and centrifuged at 12,000 rpm (4°C) for 15 min. Meanwhile, a quality control pooled (QCP) sample was prepared by mixing 10 μL of each test sample. The supernatants were filtered through a syringe filter (0.22 μm) and transferred into an autosampler vial for the following UPLC-Q-TOF/MS analysis. During the experimental period, all samples were kept at ‒80°C until used.

An Agilent Acquity UPLC system (Agilent Technologies Inc., California, United States) was interfaced with a Triple TOF MS system (AB Sciex, Massachusetts, United States) equipped with an electrospray ionization (ESI) source. A 2 μL aliquot of each vial was injected into an Acquity UPLC@BEH C₁₈ column (1.7 μm, 2.1 × 100 mm) for separating complex components. The column temperature was maintained at 30°C. The mobile phase, which consisted of water with 25 mM NH₄Ac and 25 mM NH₄OH (pH = 9.75) (A) and acetonitrile (B), was pumped at a flow rate of 0.5 ml/min under the following elution program: 0–0.5 min, 5% B; 0.5–7 min, 5%–35% B; 7–8 min, 35%–60% B; 8–9 min, 60%–60% B; 9–9.1 min, 60%–5% B; 9.1–13 min, and 5%–5% B. The MS parameters were as follows: collision energy spread, 30 eV; ion spray voltage, 5.0 kV (+) and 4.0 kV (−); capillary and heater temperature, both 650°C; and sheath and auxiliary gas flow rate, 35 and 60 psi, respectively.

The acquired MS raw data were converted to the mzXML format using ProteoWizard software (version 3.2) and imported to XCMS software (version 3.2) for automatic data processing, including peak extract, peak alignment, and peak picking, to produce peak data matrix containing retention time (R _t), m/z data, and peak intensity. Then, the resultant data matrices were further imported to SIMCA software (version 14.1) for multivariate data analysis, including principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA). PCA was employed to visualize the clustering, trends, and outliers among samples, and OPLS-DA was used to identify the differential metabolites that were responsible for the difference between two groups (Wu et al., 2015; Wu et al., 2017; Yang et al., 2017a).

OPLS-DA was employed to identify differential variables according to their VIP (variable influence on projection) values. Variables with VIP value >1 were selected and further input into Student’s t-test to determine the significance of each variable. Variables with VIP >1 and p < 0.05 were selected as candidate biomarkers. Candidate biomarkers were identified by automated comparison of retention time (R _t)-m/z datasets to a commercially available metabonomics library (provided by Biotree Biotech, Shanghai, China). Metabolic pathway analysis was conducted with MetaboAnalyst 4.0 based on the KEGG database to reveal the crucial metabolic pathways (Kanehisa et al., 2014; Chong et al., 2018).

Serum TNF-α, IL-6, IL-1β, and NO levels were measured by commercial ELISA kits according to the manufacturer’s instructions.

Each supernatant sample from the homogenized liver tissue was used to measure the activities of hepatic antioxidant enzyme SOD, CAT, and GSH-Px, and the lipid peroxidation product MDA level by using commercially available diagnostic kits according to the manufacturer’s instructions.

The data were expressed as means ± standard deviation (SD). One-way analysis of variance and Student’s t-test were used to analyze the differences between groups, and p < 0.05 was considered statistically significant, while p < 0.01 was considered highly significant.

Ingredients in the three single herbs of LCS were manually collected, including 126 ingredients in A. ilicifolius (see Supplementary Figure S2), 50 in P. pulchellum (see Supplementary Figure S3), and 43 in C. cochinchinensis (see Supplementary Figure S4). After removing duplication, the total number of compounds was 213.

TCM prescriptions are usually orally administered, and only when absorbed in blood can compounds show bioactivity (Wang et al., 2011). Oral bioavailability (OB), as one of the most important pharmacokinetic parameters, determines whether a constituent is pharmaceutically active. Compounds with adequate Caco-2 permeability and drug-likeness (DL) often exhibit pharmacodynamic action on target sites (Luo et al., 2016). Moreover, a few active compounds that do not meet all the three criteria were selected for their high amounts and high bioactivities. For example, oleanolic acid (AI01) has a relatively low OB value (<30%), whereas it is a hepatoprotective drug, so it was counted in for further analysis (Zhu et al., 2019). Similarly, other seven compounds were also regarded as candidate components, including (‒)-epicatechin (PP10), (‒)-gallocatechin (PP11), and (‒)-epigallocatechin (PP12), reported with hepatoprotective activities on a cellular level, acteoside (AI89), isoacteoside (AI90), and 4-hydroxy-2(3H)-benzoxazolone (AI37) with anti-liver injury effects, and adenosine (AI53) with anti-inflammatory effects (Cronstein, 1994; Xiong et al., 1998; Kanchanapoom et al., 2001; Liu et al., 2013; Ma et al., 2015; Saranya et al., 2015; Fan et al., 2018). It should be pointed out that five compounds were not available in the TCMSP database, including blepharigenin (AI41), (2R)-2-O-β-D-glucopyranosyl-2H-1,4-benzoxazin-3(4H)-one (AI42), (2R)-2-O-b-d-glucopyranosyl-5-hydroxy-2H-1,4-benzoxazin-3(4H)-one (AI45), 7-Cl-(2R)-2-O-β-D-glucopyranosyl-2H-1,4-benzoxazin-3(4H)-one (AI47), and 2,6-dimethoxy-p-hydroquinone-1-O-β-D-glucopyranoside (AI71), whereas they were frequently reported with significant anti-liver injury effects; thus, they were also reserved as candidate compounds (Kanchanapoom et al., 2001; Huo et al., 2005; Lin et al., 2013; Saranya et al., 2015). Consequently, 57 candidate compounds were selected from the 213 compounds in LCS (ADME parameters of 57 candidate compounds see Supplementary Table S1), with 39 from A. ilicifolius, 17 from P. pulchellum, and 6 from C. cochinchinensis.

Among 57 candidate compounds, only 7 candidate compounds, AI12, AI41, AI49, AI53, AI55, AI71, and AI110, related with no target associated with liver injuries, and the other 50 compounds yielded 87 potential targets. (For information of 87 potential targets, see Supplementary Table S2.)

The platform DAVID provides a comprehensive set of functional annotation tools to understand the biological meaning behind a large list of genes (Huang et al., 2008; Xu et al., 2018). The Gene Ontology (GO) annotation system in DAVID could be used to assign functions to genes (Li et al., 2015). The GO annotation system contains three independent categories: biological processes (BP), molecular function (MF), and cellular components (CC). In the present study, three categories in GO annotation, namely, BP, MF, and CC, accounted for 69.04, 24.55, and 6.41%, respectively, with the top 10 significantly enriched terms in BP, MF, and CC categories, as shown in Figure 1. The key BPs were steroid metabolic process, oxidation–reduction process, bile acid and bile salt transport, drug metabolic process, xenobiotic metabolic process, and negative regulation of apoptotic processes, indicating that LCS could mainly regulate inflammatory, oxidation–reduction, and bile transport processes. The key MFs were carbonate dehydratase activity, oxygen binding, oxidoreductase activity, iron ion binding, drug binding, aromatase activity, monooxygenase activity, zinc ion binding, and heme binding, mainly related to the redox system. As for CC categories, most of the predicted targets were located in the cytoplasmic part, extracellular region, organelle membrane, and organelle, which indicated that the anti-liver injury effects of LCS could be implemented through the processes of preserving cellular structural integrity, repairing hepatic tissue damage, and regulating hepatic cellular homeostasis.

In this study, two visualized networks were constructed, namely, candidate compound–potential target and potential target–pathway networks, which could help decipher the relationships among active compounds, targets, and pathways.

Integrating with the aforementioned two networks, the CI of every candidate compound was calculated based on NE and weighted by the literature. According to CI calculated results (Figure 5A, Supplementary Table S3), eight active compounds displayed the most contribution to the anti-liver injury effects of LCS with a sum of CIs of more than 90%, namely, quercetin (AI27/PP02/CC28), ferulic acid (AI60), naringenin (CC27), luteolin (AI28), gallic acid (AI63), oleanolic acid (AI01), vanillic acid (AI64), and acteoside (AI89) (Figure 5B).

Quercetin (AI27/PP02/CC28) was reported to have radical-scavenging activities and inhibitory activities on lipopolysaccharide-induced inflammatory both in vitro and in vivo (Shen et al., 2002; He et al., 2011). Ferulic acid (AI60) had the ability of inhibiting lipid peroxidation, antioxidation, and anti-inflammatory by reducing the expression of TNF-α and IL-1β in lipopolysaccharide-activated macrophages (Xu et al., 2013; Navarrete et al., 2015). Naringenin (CC27) could attenuate CCl₄-induced hepatic inflammation by activating the Nrf2-mediated pathway in rats (Esmaeili and Alilou, 2014). Luteolin (AI28) could exert anti-inflammatory effects by preventing arachidonic acid synthesis and scavenging hydrogen peroxide and had antioxidant and anticancer properties (Odontuya et al., 2005; Lin et al., 2008). Gallic acid (AI63) could inhibit pro-inflammatory cytokine production and histamine release in mast cells (Kim et al., 2006). Oleanolic acid (AI01) could protect the liver against many hepatotoxicants (Liu et al., 2008). Vanillic acid (AI64) could exert anti-inflammatory activity by regulating the expression of pro-inflammatory cytokines (Ziadlou et al., 2020). Acteoside (AI89) could modulate inflammatory responses through the NFκβ/Iκβ signaling pathway in alcohol-induced hepatic damage and possess many other pharmacological activities, such as antioxidant and antitumor activities, and hepatoprotective effects against CCl₄-induced liver injury (Wong et al., 2001; Khullar et al., 2019; Zhang et al., 2020). The aforementioned summary of the eight key active compounds in LCS further indicated that the anti-liver injury mechanism of LCS was largely related to anti-inflammatory and antioxidant activities.

Molecular docking was used to evaluate the interactions between candidate compounds in LCS and their potential targets. The molecular docking results of 40 representative targets are listed in Supplementary Table S4. According to the molecular docking results of 40 representative targets and their corresponding compounds, it was suggested that these potential targets associated with liver injuries could finely bind to their connected compounds. Target SLC10A1 that is bile acid cotransporter was taken as an example. The 6QD5 protein of target SLC10A1 had the best binding conformation with AI07 (grid score = ‒8.49016). Their binding mode showed that AI07 was well accommodated inside the binding pocket of 6QD5, mainly through hydrogen bond interaction, and the main binding amino acid residues were Asn73 and Asp41 (Figure 6).

The score plots of PCA displayed pattern distinction for the LCS group, model group, and normal control group in ESI⁺ as well as in ESI⁻ modes in their UPLC-Q-TOF/MS metabolic profiles (Figure 7A). Samples in the same group almost clustered together and were within the 95% Hotelling’s T-squared ellipse, confirming that the variables observed in samples were biologically relevant. The samples in the model group could be significantly discriminated against those of the normal control group, indicating that CCl₄ could induce metabolic disorder in the liver. Furthermore, the PCA score plots also showed that the LCS group displayed a returning trend to the normal control group from that of the model group, showcasing the ability of LCS to reverse CCl₄-induced metabolism disorder.

The datasets of samples from the normal control and model groups were extracted respectively as a matrix for OPLS-DA analysis to screen for potential differential metabolites associated with the liver injury induced by CCl₄. The score plots for OPLS-DA presented obvious discrimination between the normal control group and the model group in the ESI⁺ data set as well as in the ESI⁻ data set (Figure 7B), indicating CCl₄ could induce metabolic variations. The goodness-of-fit and predictability of OPLS-DA analysis models were evaluated by calculating R²Y and Q², respectively. The result of the 200-permutation test of the OPLS-DA model indicated that the OPLS-DA model was reliable and not the result of statistical over-fitting (Figure 7C).

The potential metabolites that significantly contributed to the clustering and discrimination were selected based on their variable importance projection (VIP) value and p-value. Metabolites with VIP >1 combined with p < 0.05 were selected as potential differential metabolites. Following the criteria aforementioned, 32 differential metabolites were identified (Table 1).

In order to study the effect of LCS on disturbed metabolites induced by CCl₄, a heat-map was drawn according to the relative intensity of 32 metabolites in LCS, normal, and model groups (Figure 8). The results of the heat-map analysis indicated that metabolites in the LCS group were similar to those in the normal control group. Meanwhile, the relative intensities of these metabolites in different groups were statistically analyzed by one-way ANOVA. As a result, 18 potential biomarkers were notably reversed by LCS treatment (Figure 9).

To investigate the metabolism pathways influenced by LCS treatment, the significant differential metabolites were analyzed using Metaboanalyst 4.0. Ten corresponding pathways were highly related to LCS treatment (Figure 10). Among these metabolic pathways, linoleic acid metabolism, glutathione metabolism, cysteine and methionine metabolism, and glycerophospholipid metabolism pathways were filtered out as significant pathways since these pathways had impact values >0.10 and p < 0.05 (Hoffmann et al., 2014). The detailed information of these 4 significant metabolism pathways is shown in Supplementary Table S5.

Linoleic acid metabolism pathway: As a kind of free polyunsaturated fatty acid (PUFA), linoleic acid, and γ-linolenic acid, the precursors of many polyunsaturated fatty acids, are closely related to human health. The increases of these PUFA could reflect the outbreak of inflammatory response and oxidative stress during liver injury. After treatment with LCS, the increased hepatic linoleic acid and γ-linolenic acid levels could be significantly down-regulated, suggesting that LCS could modulate inflammatory responses and oxidative stress during liver injury by regulating the linoleic acid metabolism pathway.

Glutathione metabolism pathway: Glutathione plays an important role in eliminating free radicals and in defending oxidative stress and therefore decreasing liver injury (Gabele et al., 2009). In this study, we detected many metabolites involved in the glutathione metabolism pathway, such as glutathione and L-glutamate. LCS could increase the glutathione content in the CCl₄-induced hepatic injury rats, suggesting that LCS had the ability to alleviate the disordered glutathione homeostasis.

Cysteine and methionine metabolism pathway: It is recognized that the liver plays an important role in cysteine and methionine metabolism, and marked impairment of this pathway is usually observed in patients with liver injuries. As one of the most important biochemical reactions, the cysteine and methionine metabolism generates approximately 70% methyl donors. In the present study, we detected that two metabolites, S-adenosyl-L-homocysteine (SAMe) and S-adenosylmethionine (SAH), involved in the methionine cycle reaction, were significantly decreased upon CCl₄ inducement. SAMe has the capacity to increase glutathione levels, regulate the hepatocyte apoptotic response and membrane fluidity, and decrease the production of inflammatory factors, which has also been found to protect the liver against various injuries (Mato and Lu, 2007). SAH is a potent inhibitor of SAMe-dependent methylation reactions. It is toxic to immature lymphocytes and can lead to immune suppression. In our study, LCS had the ability to regulate the disordered cysteine and methionine pathway in the liver by reversing the levels of hepatic SAMe and SAH.

Glycerophospholipid metabolism pathway: Glycerophospholipids, storage deposits for lipid mediators, function as integral membrane proteins, transporters, receptors, and ion channels. As important intermediates involved in the synthesis of the characteristic bilayer structure of cells and the maintenance of membrane integrity, their perturbation reflects the disorders of lipid metabolism under severe oxidative stress (Ekroos, 2012). In our study, compared with the normal control group, the levels of glycerophospholipid metabolites (including phosphatidylglycerol, phosphoethanolamine, CDP-choline, and glycerophosphocholine) decreased in the model group, revealing that CCl₄ could cause liver damage by altering the structural integrity and permeability of the plasma membrane in liver tissue. After treatment with LCS, the levels of glycerophospholipid metabolites were significantly reversed, suggesting that LCS could protect the cell membrane against oxidative stress damage.

Free radicals generated from CCl₄ stimulated hepatocytes could secrete signals to activate the innate immune system and Kupffer cells, and then exacerbate liver inflammation by generating a variety of inflammatory cytokines, such as TNF-α, IL-6, and IL-1β (Lin et al., 2012; Wang et al., 2019). TNF-α stimulates the release of cytokines from macrophages and induces oxidative metabolism and NO production (Gao et al., 2014). NO is a highly reactive oxidant and its excessive accumulation could cause damage to the liver. Acute hepatotoxicity could induce inflammation mediated by cytokines. In order to determine the effect of LCS against CCl₄-induced inflammation, the levels of cytokines (TNF-α, IL-6, IL-1β, and NO) in rat serum were measured. Results showed that the levels of these cytokines were significantly increased in the model group compared with the normal control group (p < 0.01, Figures 11A–D). In comparison with the model group, LCS exhibited significant inhibition effects on the secretion of these cytokines, especially IL-6 and IL-1β, with the levels about half of those of the model groups (p < 0.01, p < 0.05) and near to those of normal and positive groups (Figures 11B–C), indicating that LCS could inhibit the production of inflammatory cytokines in rat liver injury induced by CCl₄.

Oxidative stress is the main response of acute liver injury induced by CCl₄, which is associated with the severity of lipid peroxidation and the activities of the antioxidant defense system (Karakus et al., 2011; Ma et al., 2015). A defense system of antioxidant enzymes plays an important role in preventing liver damage induced by CCl₄ (Ma et al., 2015). Lipid peroxides and reactive radicals can easily inactivate antioxidant enzymes under toxic conditions. Therefore, changes in the hepatic levels of these antioxidant enzyme activities are closely related to the ability of the liver to cope with oxidative stress during CCl₄ inducement. SOD, CAT, and GSH-Px are major enzymes in the antioxidant system (Zhong et al., 2009; Gao et al., 2011). Among them, SOD can convert superoxide anions into hydrogen peroxide, and CAT can catalyze the breakdown of hydrogen peroxide into oxygen and H₂O, while GSH-Px can catalyze the reduction of hydrogen peroxide into non-toxic hydroxyl molecules (Hsu et al., 2008). In addition, as an index of the status of oxidative stress and lipid peroxidation, the end-product of lipid peroxidation, MDA, was broadly adopted to assess the level of lipid peroxidation (Abuja and Albertini, 2001). Herein, we assessed the activation of the three antioxidant enzymes (SOD, CAT, and GSH-Px) and the level of lipid peroxide MDA.

As shown in Figures 11E–G, the activities of hepatic SOD, CAT, and GSH-Px in the model group were significantly lower than those in the normal control group (p < 0.01). In comparison with the model group, the activities of SOD, CAT, and GSH-Px in LCS groups were significantly increased (p < 0.01) and were near to normal and positive groups. Specifically, all of the three dosages of LCS groups showed equal or superior abilities to increase the activity of GSH-Px in the positive group. Moreover, the level of MDA significantly increased in the model group compared to that in the normal control group (p < 0.01, Figure 11H). In comparison with the model group, the levels of MDA in the LCS groups were markedly reduced (p < 0.01), especially in the LCS-M and LCS-H groups, showing stronger reduction effects and close to the positive group. These results suggested that LCS could maintain antioxidant enzyme activities and ameliorate lipid peroxidation levels, which could improve the antioxidant defense system to attenuate liver injury induced by CCl₄.