Methanol and acetonitrile of HPLC grade were bought from Tedia Company (Fairfield, United States). 3-(trimethylsilyl) propionic-2, 2, 3, 3-d₄ acid sodium salt (TSP) and Deuterium oxide (D₂O, 99.9%) were purchased from Sigma-Aldrich (St. Louis, MO, United States). Deionized water was ordered from Watsons (Watsons, Hong Kong, China). The three standards, including Cucurbitacin B (CuB), Isocucurbitacin B (IsocuB), and Cucurbitacin E (CuE) were laboratory-made and identified by¹H and ¹³C NMR. Hu-lu-su-pian (HLSP, the main ingredients are CuB and CuE, 0.1 mg/tablet) was used as a positive drug and was obtained from Hunan Dinuo Pharmaceutical Co., Ltd. (Changsha, China). The Nanjing Jiancheng Bioengineering Institute (Nanjing, China) supplied the assay kits for the enzymes aspartate aminotransferase (AST) and alanine aminotransferase (ALT). HPP were provided by Beijing Tibetology Research Center (Linzhi City, Tibet Autonomous Region, China). Other reagents were all analytically pure.

5 kg HPP was crushed to 30 mesh with a multifunctional crusher, and then extracted twice with ethanol (1:10, w/v) under reflux for 2 h each time. The two filtrates were collected and concentrated under reduced pressure to obtain 390 g of an ethanol extract. The extracts were extracted using petroleum ether and ethyl acetate after being suspended in water. The extract (48.5 g) of ethyl acetate was successively eluted with petroleum ether-acetone (8:2, 7:3, 6:4, 5:5, 4:6, v/v) on a silica gel column (50 cm × 8 cm i.d.), and the thin fractions were checked by thin layer chromatography (TLC). Fractions 11-14 were collected, concentrated under reduced pressure, and fully dried in a vacuum drying oven to obtain 3.42 g of TCs.

ICR mice (20.0 ± 2 g) were obtained from Qinglongshan Animal Breeding ground in Nanjing. All animals were kept in a specified pathogen-free (SPF) laboratory with unrestricted access to food and water at a temperature of 25°C, with a humidity of 55% ± 5%, and a 12-h light/dark cycle. The animals used in this study have been handled in accordance with the National Institutes of Health (NIH) guidelines, and the experimental protocol has been approved by the Animal Care and Use Committee of Nanjing University of Science and Technology.

For histopathology, liver tissues were quickly removed, and then fixed with 10% formalin to prepare paraffin sections with 5 μm sections for H&E staining, and the remainder were stored at −80°C for other analyses.

For serum biochemical analysis, ALT and AST levels were assessed using a commercial kit in accordance with the manufacturer’s recommendations.

Liver tissue sample preparation for ¹H NMR analysis (Ruan et al., 2018): Frozen liver tissue samples were crushed in a mortar and pestle in the presence of liquid nitrogen and immediately weighed, then homogenized (5 mL/g) using pre-cooled acetonitrile-water (50:50, v/v). The homogenate was centrifuged for 10 min at 16,000 g at 4°C, and the supernatant was then transferred to a centrifuge tube. With the removal of acetonitrile under a nitrogen blower, it was then lyophilized and kept at −80°C for future use.

Serum sample preparation for ¹H NMR analysis (Fu et al., 2019): Slowly thawing serum samples on ice was followed by the addition of twice as much methanol, the mixture was vortexed and left to stand at −20°C for 20 min. The supernatant was evaporated with a nitrogen blower to remove methanol following centrifugation at 14,000 g for 15 min at 4°C. It was then freeze-dried and kept at −80°C for future use.

The dried liver and serum extracts were added with a new mixture of 600 μL99.8% D₂O phosphate buffer (0.2 M, Na₂HPO₄ and 0.2 M NaH₂PO₄, pH 7.0) containing 0.05% (w/v) sodium 3-(trimethylsilyl) propionate-2,2,3,3-d₄ (TSP). After vortex and 14,000 g centrifugation for 10 min, the supernatant was transferred to the 5 mm NMR tube for the ¹H NMR analysis.

We acquired ¹H NMR spectra of liver and serum samples using a Bruker Avance 500 MHz spectrometer (Bruker GmbH, Karlsruhe, Germany). The pulse sequence was edited with a lateral (perpendicular to the main magnetic field direction) relaxation-edited Call-Purcell-Meiboom-Gil (CPMG) sequence (90 (τ-180-τ) n-acquisition) and a spin echo delay of 10 ms (2 nτ) in order to suppress residual macromolecular protein signal in the sample. The number of scans was 32 (32 K data points) and the spectral width was 20 ppm. The spectra were Fourier transformed after multiplying the free induction decay (FID) curve by an exponential weighting function (corresponding to a 0.5 Hz linewidth).

Using the software Topspin 3.0 from Bruker BioSpin, all NMR spectra were manually pre-processed for baseline, phase, and TSP zero (0.0 ppm). MestReNova (version 6.1.0, Mestrelab Research SL) was used to convert the files to ASCII format, which was then imported into “R” (http://cran.r-project.org/) for multivariate statistical analysis. To minimize the data dimension, segmental integration is carried out in the chemical potential range of 0.5 ppm–9 ppm using an adaptive binning approach with an average bin width of 0.015 ppm. Between 4.47 and 5.50 ppm, the remaining signals of water and its affected regions were eliminated. Finally, the data were subjected to Pareto scaling (Karaman, 2017) and probability quotient normalization (Dieterle et al., 2006) to eliminate systematic differences between sample concentrations.

Orthogonal signal correction-partial least squares discriminant analysis (OSC-PLSDA) was then performed on the processed data, revealing differential metabolic changes in serum and liver tissue. The orthogonal signal correction (OSC) filter is applied to remove uninteresting variations, such as systematic variation, from spectral data before performing PLS-DA. To evaluate the fitting ability and predictive ability of the established OPLS-DA model, repeated two-fold cross-validation (2CV) and permutation test (n = 200) were performed. R ² (total explained variance) and Q² (model predictability) values were used to assess the model’s quality. Differences and clusters between groups were displayed using score plots, and metabolites that changed between groups were displayed using colored loading plots.

Metabolites of liver tissue and serum were identified using the commercial software Chenomx NMR Suite v.8.1 (Chenomx, Edmonton, Canada) and statistical total correlation spectroscopy (STOCSY) techniques, and then by querying public metabolite databases such as the Human Metabolite Database (HMDB, http://www.hmdb.ca) and the Madison-Qingdao Metabolomics Database (MMCD, http://mmcd.nmrfam.wisc.edu).The preliminary identification of metabolites was further confirmed by using 2 dimensional (2D) heteronuclear single quantum correlation (HSQC) and a ¹H–¹H total correlation (TOCSY) NMR spectroscopy.

For univariate analysis, a parametric test (t-test) and a non-parametric statistical test (Wilcoxon signed rank test) were employed to validate key increased or decreased metabolites across groups using “R”. The fold change (FC) values of metabolites adjusted by the Benjamin-Hochberg modification approach and their correlated p values were calculated and exhibited as colored tables. The significance threshold for all tests was set at p < 0.05.

All data are expressed as mean ± SD, and statistical analysis was performed using GraphPad Prism 8.0 software (GraphPad Software, CA, United States).

Molecular networking is a powerful complement to screening and identification of undescribed compounds from natural products (Yang et al., 2013). The molecules exhibiting similar fragmentation patterns were clustered together while the molecules with dissimilar MS/MS spectra were displayed as independent nodes (Wang et al., 2016). The visualized molecular networking of TCs and HLSP using Cytoscape 3.6.0 software is presented in Figure 2. Each node represents a compound, inside which the molecular mass of parent ion of the compound is indicated. The nodes of compounds are linked by lines indicating structural similarity, with thicker lines representing greater similarity between the structures. The red and blue node represents TCs and HLSP, respectively. The Pink area represents major cucurbitacin clusters. Clusters that formed by red nodes and blue nodes represent the unique compounds of TCs (Figure 2B) and HLSP (Figure 2C), respectively. Compounds 1 and 2 showed identical [M + FA-H]⁻ ions at m/z 603.3 in negative ion mode, suggesting that they were isomers. Furthermore, their MS/MS spectra exhibited the typical daughter ions at m/z 497.3, corresponding to the loss of CH₃COOH (60 Da). In addition, the other fragment ions were nearly alike except for some disparities in relative abundance (Supplementary Figures S1, S2). This revealed that they were two configurational isomers. Thus, Compounds 1 and 2 were therefore tentatively identified as cucurbitacin B and isocucurbitacin B by comparing their retention times and MS/MS spectra with those of authentic standards.

Compound 3 showed a [M + FA−H]⁻ ion at m/z 601.3, as well as the typical daughter ion at m/z 495.3 which were 2 Da less than compound 1 and 2 (Supplementary Figures S1, S2), indicating that the core structure of compound 3 contained one additional double bond compared to Compound 1 and 2. Therefore, by comparing its retention time and MS/MS spectra with those of an authentic standard, compound 3 was tentatively identified as cucurbitacin E. The retention time, molecular formula, mass errors, MS/MS fragment ions, and detected monoisotopic masses in the four-decimal format provided by QTOF-MS for the major cucurbitacins and the other nine derivatives are summarized in Table 1. Other untargeted compounds (with an MS² spectrum) were also annotated by matches to the GNPS spectral libraries (data not shown).

In order to further investigate the amount of each component in TCs, three primary components were quantified by HPLC technique. CuB and IsocuB could not be completely separated by 65% methanol (Figure 3Aa), whereas the two overlapping isomers could be completely separated by 50% acetonitrile (Figures 3Bb, c, d) with a resolution R > 1.5, therefor, an isocratic elution with 50% acetonitrile was used to determine CuB, IsocuB, and CuE in TCs and HLSP (Figure 3B). The contents of CuB, IsocuB, and CuE were respectively 70.3% ± 0.4%, 26.1% ± 0.2%, and 3.6% ± 0.3% in TCs and 41.0% ± 1.1%, 42.8% ± 0.7% and 16.2% ± 0.6% in HLSP, as shown in Table 2.

LD₅₀ and NOAEL values are two metrics commonly used to assess the safety of drugs (Vilariño et al., 2018). Following gavage administration of TCs, all mice died within 5 hours at a dosage of 80 mg/kg. Therefore, the second level dosage was reduced to 40 mg/kg, leading to the death of three out of five mice. At the third level, a dosage of 20 mg/kg resulted in no deaths, prompting a further reduction to 10 mg/kg for the fourth level. None of the nine mice who received this dose died. An additional dose level of 30 mg/kg was introduced between the second and third levels, resulting in the death of two out of seven mice. Lastly, nine mice were given a dosage of 15 mg/kg, with no deaths observed (Table 3). Symptoms of apathy and piloerection were observed within 30 min across all groups, except at 15 and 10 mg/kg. Convulsions increased in frequency over time, occurring every 15 s. Semi-closed eyes were observed an hour before death. At a dosage of 20 mg/kg, 1 mouse showed signs of apathy and another had a minor piloerection. At 15 and 10 mg/kg, none of the mice died and there were no side effects over the following 2 weeks (Table 4). Based on these findings, the oral NOAEL for mice was established at 15 mg/kg, and this modified four-level “up and down” procedure revealed dose-dependent mortality. The results, presented as a percentage of mouse deaths in relation to TCs dosage, are shown in Figure 4. Nonlinear regression fitting software was used to calculate the oral LD₅₀ of TCs to be 36.21 mg/kg (GraphPad Prism, version 8.0, GraphPad Software, CA, United States).

Throughout the treatment period, the body weight of mice in all groups exhibited an upward trend (Figure 5A). This suggests that both TCs and HLSP had minimal effect on the animals’ eating behavior at treatment doses. Hepatomegaly, as evidenced by the mean relative liver weight (also known as liver index), was induced by CCl₄, with the CCl₄ group having a significantly higher liver index than the NC group (p < 0.01). Treatment with TCs and HLSP, on the other hand, resulted in a significant decrease in liver index compared to the CCl₄ group (p < 0.01), with no significant difference observed between the HLSP and MD groups (Figure 5B). The disturbance of the hepatocyte membrane structure caused by CCl₄ led to a substantial rise in serum AST and ALT levels (p < 0.0001) (Figures 5C,D). Dose-dependent treatment with TCs resulted in a significant and substantial drop in ALT and AST levels in the MD and HD groups (p < 0.01 and p < 0.0001) (Figures 5C,D), with no significant difference observed between the HLSP and MD groups. It can be concluded that TCs can prevent hepatomegaly in CCl₄-induced mice.

The purpose of histopathological observation is to reveal morphological alterations in liver cells, as well as inflammatory cell infiltration and other pathological conditions, which cannot be fully captured by serum biochemistry. Histological sections of liver tissue showed that the NC group had a morphologically normal liver lobular and cellular structure (Figure 6A). In contrast, the CCl₄ group exhibited clear localised necrosis of hepatocytes, loss of hepatocyte structure, and infiltration of inflammatory cells, particularly around the central vein (indicated by the red arrow). Inflammatory cell infiltration was most common in the portal area and manifested as a decrease in nuclear volume, loss of cell structure, or aberrant cell shape. The HLSP group exhibited considerable improvement in hepatocyte shape and inflammatory infiltration. Perivascular inflammatory cell infiltration was lower in the LD group than in the CCl₄ group. Increased dosages of TCs reduced inflammatory cell infiltration and improved cell morphology, resulting in a decrease in liver tissue necrosis area in a dose-dependent manner (Figures 6B,C). The HD group had the greatest hepatoprotective effect, reducing the necrotic area of liver tissue by 77.9% compared to the CCl₄ group (p < 0.0001). The hepatoprotective effect showed a significant dose-response relationship, with the MD group performing marginally better than the HLSP group (p < 0.001), and the HD group performing the best.

The main metabolic differences across groups were investigated using the OPLS-DA model, and the CCl₄ and NC groups were clearly separated in liver and serum samples (Figure 7A; Figure 8A). Following TCs therapy, both the MD and HD groups distinguished themselves from the CCl₄ group (Figures 7Ba and 8Ba), and the HD group nearly overlapped with the NC group (Figures 7Bb and 8Bb). The HD group exhibited greater separation from the CCl₄ group than the HLSP group (Figures 7Bc and 8Bc), suggesting that TCs can dose-dependently reverse CCl₄-induced metabolic abnormalities in mice.

Using the OPLS-DA model, different metabolites were represented by various-sized points of different colors and shapes in the S-plot (Figure 7 bottom left). Among these metabolites, a total of 19 endogenous metabolites were identified (Figure 9) in liver and serum, 11 metabolites were chosen for potential biomarkers to distinguish between the NC and CCl₄ groups. Lactate, fumaric acid, succinic acid, inosine, O-phosphocholine, ascorbate, taurine, tyrosine, phenylalanine, valine, and alanine were chosen based on fold changes and associated p-values between groups (see Supplementary Tables S3, S4 for details).