The separation process was performed using column chromatography (CC), including silica gel (300–400 mesh) purchased from Qingdao Marine Chemical Co., Ltd., reverse phase silica gel (RP-18, 40–63 μm) purchased from Merck&Co., Ltd. in New York, United States, and Sephadex LH-20 (GEHealthcare) purchased from Uppsala, Sweden. HPLC was performed on a Gilson PLC 2050 liquid chromatograph equipped with a Gilson PLC 2050 UV detector at a wavelength of 206 nm, using a Hedera ODS-2 column (250 × 20 mm, 10 μm, 10 nm) for semi preparation. The separated chemical reagents were purchased from Tianjin Fuyu Chemical Co., Ltd. in China. The standard samples of D-glucose (D-Glc), L-rhamnose (L-Rha), and L-arabinofuranose (L-Ara) were purchased from Sigma Aldrich Chemical Co. in St. Louis, Missouri, United States.Perform GC-MS analysis on the simadzu GC-MS QP 2010 instrument to obtain information on sugar chains, which select RXI-5 SIL MS column (30 m × 0.25 mm × 0.25 μm). ESI-MS and HRESI-MS were acquired utilizing an Agilent Q-TOF mass spectrometer (Agilent Technologies, Santa Clara, CA, United States). Optical rotation was performed on the INESA SGC-568 polarimeter (Yidian Physical Optics Co., Ltd., Shanghai, China). Conducted on a Bruker Ascend 800 spectrometer (Karlsruhe Bruker GmbH, Germany), TMS was used as an international standard for NMR. Additionally, the use of these advanced instruments and high-quality reagents ensures accurate and reliable results in the analysis and separation of complex chemical compounds. The combination of different chromatographic techniques and spectroscopic methods provides a comprehensive understanding of the chemical structures and properties of the substances under investigation.

In June 2016, the roots and rhizomes were harvested from Taibai Mountain in Shaanxi Province, China, and were authenticated as C. robustum Maxim by Dr. Hai-Feng Tang. A voucher specimen, designated as JB20160605, has been archived in the Herbarium of the Department of Chinese Medicinal Materials and Natural Drugs at the School of Pharmacy, Air Force Medical University in Xi’an, China.

The roots and rhizomes of plants (5 kg) were chopped into pieces and extracted with 15 L of 70% ethanol at a reflux temperature of 85°C for 2 h, repeated three times. The extracts were combined and concentrated using a rotary evaporator to obtain a residue (413.2 g) Equal volumes of water were added to the residue to disperse it, followed by partitioning with petroleum ether and then with n-butanol successively. The n-butanol fraction (202.6 g) was subjected to column chromatography on silica gel, using a solvent system of CHCl₃-MeOH-H₂O with a volume ratio that varied from 100:1:0 to 6:3:0.5, resulting in the collection of seven fractions labeled as Fr.1 through Fr.7. Fr.4–2-1 (470 mg) underwent further purification through semi-preparative HPLC, utilizing a mobile phase of MeOH-H₂O (60:40) at a flow rate of 8.0 mL/min, resulting in the isolation of compound 1 (5 mg, t_R = 26.5 min). Fr.6 (14.77 g) was processed with a Sephadex LH-20 column using CHCl₃-MeOH(1:1) as the eluent, yielding two subfractions, Fr.6–1 and Fr.6–2. From Fr.6–1 (12.66 g), three subfractions (Fr.6–1-1 to Fr.6–1-3) were separated using an ODS column. Fr.6–1-3 (699.7 mg) was then subjected to semi-preparative HPLC with a MeOH-H₂O (70:30) mobile phase at a flow rate of 8.0 mL/min, leading to the purification of compound 2 (23.3 mg, t_R = 23.5 min), 3 (28.3 mg, t_R = 40.1 min), and 4 (11.9 mg, t_R = 20.8 min). Fr.1 (5.11 g) was chromatographed on a Sephadex LH-20 column with MeOH as the eluent to obtain four subfractions (Fr.1–1 to Fr.1–4). Two subfractions (Fr.1–3-1 and Fr.1–3-2) were derived from Fr.1–3 (3.05 g) after passage through an ODS column. Fr.1–3-1 (50 mg) was further purified by semi-preparative HPLC with a MeOH-H₂O (70:30) mobile phase at a flow rate of 8.0 mL/min, affording compound 5 (19.8 mg, t_R = 24.1 min). Compound 6 (20 mg, t_R = 17.8 min) was extracted from Fr.1-3-1-2 (40.02 mg) using HPLC with a MeOH-H₂O (70:30) eluent at a flow rate of 8.0 mL/min. Fr.3 (2.73 g) was purified over a Sephadex LH-20 column with MeOH as the eluent, resulting in two subfractions (Fr.3–1 and Fr.3–2). Two subfractions (Fr.3–1-1 and Fr.3–1-2) were obtained from Fr.3–1 (0.72 g) following passage through an ODS column. Fr.3–1-2 (100 mg) was further purified by semi-preparative HPLC using a MeOH-H₂O (70:30) mobile phase at a flow rate of 8.0 mL/min, yielding compounds 7 (55.9 mg, tR = 30.5 min) and 8 (6.8 mg, t_R = 43.3 min). Fr.4 (5.89 g) was processed on a Sephadex LH-20 column with MeOH as the eluent to yield two subfractions (Fr.4–1 and Fr.4–2). Five subfractions (Fr.4–2-1 to Fr.4–2-5) were separated from Fr.4–2 (4.47 g) using an ODS column. Fr.4–2-5 (60 mg) was then purified by semi-preparative HPLC with a MeOH-H₂O (60:40) mobile phase at a flow rate of 8.0 mL/min, resulting in the isolation of compound 9 (169.8 mg, t_R = 20.7 min).

The gas chromatography-mass spectrometry (GC-MS) technique was employed to analyze the monosaccharide content of compounds 1 and 2, adhering to a procedure akin to those reported in existing scientific literature. Each compound, weighing 5 mg, was subjected to hydrolysis by being treated with 1 mL of 2 M trifluoroacetic acid (TFA) at a temperature of 110°C for a duration of 90 min. Following hydrolysis, acetylation was performed using acetic anhydride. The reaction mixture was then diluted with 2 mL of distilled water and reduced using 100 mg of sodium borohydride (NaBH₄). After reduction, the mixture was re-acetylated with acetic anhydride at 100°C for 60 min. Standard monosaccharide controls—D-glucose (D-Glc), L-rhamnose (L-Rha), and L-arabinose (L-Ara), each at a concentration of 5 mg—were processed using the identical method and analyzed via GC–MS.

The GC-MS analysis was conducted under the specified conditions: the column temperature was ramped from 130°C to 180°C at a rate of 3°C per minute and held at 180°C for 5 min. Subsequently, the temperature was programmed to increase to 310°C at a rate of 10°C per minute, where it was maintained for 15 min. The temperatures of both the injector and the flame ionization detector (FID) were set to 285°C. Nitrogen gas (N₂) with a purity of at least 99.999% was used as the carrier gas, flowing at a rate of 20.0 mL per minute. The retention times for the standard monosaccharides were recorded as follows: D-Glc at 12.89 min, L-Rha at 5.65 min, and L-Ara at 5.15 min.

The cytotoxicity of compounds 1–9 against human cancer cells was detected using the MTT method in 96-well microplates. The DMEM or RPMI-1640 culture medium (Hyclone, United States) was supplemented with 10% FBS (GIBCO, United States), 1% penicillin, and streptomycin (Elabscience Biotechnology Co., Ltd., Wuhan, China), in which the cell lines were cultured at 37°C with 5% CO₂. Cells in the logarithmic phase (5000 cells/well) were seeded in 96-well plates (100 μL/well) incubated for 24 h, then treated with various concentrations of saponins 1–9 (<0.1% DMSO) for 24 h, separately. Then, 20 μL MTT reagent (Sigma Biotechnology Co., Ltd.) was added to each well incubated for 2 h at 37°C with 5% CO₂. The optical density of each well was measured using a microplate reader (BioTek, United States) at a wavelength of 450 nm. The IC₅₀ values of saponins 1–9 were evaluated according to their optical densities. The experiment was conducted three independent replicates, with doxorubicin and nimustine hydrochloride (Sigma, purity ≥99%) used as positive controls.

By means of the SEA platform (https://sea16.docking.org/), the target prediction of compounds with favorable activity was carried out, and common targets were selected for molecular docking simulation.

Molecular docking was employed to depict the interactions between proteins and ligands. The protein CD81, identified by its UniProt accession number P60033, was sourced from the UniProt database (https://www.uniprot.org/). Meanwhile, the ideal molecular configurations for compounds 1–9 were derived from Chem3D. In the molecular docking process, the structure of CD81 was modified using PyMOL 2.5.4 and AutoDock Tools 1.5.6 software. Additionally, the affinity between CD81 and its ligand was assessed using Auto Dock Vina 1.1.2 software. PyMOL 2.5.4 was utilized to visualize the molecular docking outcomes. This comprehensive approach of target prediction and molecular docking provides valuable insights into the potential mechanisms of action of these compounds. It helps in understanding how the compounds interact with specific proteins and may offer clues for further research and development of novel therapeutics.

The Liebermann-Burchard and Molish tests indicated that compound 1 (1) was a saponin (Liu et al., 2023). The molecular formula was identified as C₅₆H₈₈O₂₇ by HR-ESI-MS at m/z 957.5076 [M - H]^- (calcd. For C₄₈H₇₈O₁₉, 957.5059).The NMR spectrum (Table 1) showed six methyl signals at δ _H 0.98 (s, H-29), 1.05 (s, H-30), 1.09 (s, H-24), 1.09 (s, H-25),1.23 (s, H-26),1.80(s, H-27), an ene proton δ _H 5.64 (br s, H-12). Correspondingly, six methyl carbon signals at δ _C 13.2 (C-24),16.3 (C-25),17.8 (C-26),24.7 (C-30),27.3(C-27),33.2 (C-29). Meanwhile, one oxygenated methine δ_C 73.7 (C-3) and δ_C 74.4 (C-16), two olefinic carbons at δ _C 122.9 (C-12) and 144.5 (C-13), as well as one carboxyl at δ _C 176.1 (C-28) can be found in the ¹³C NMR spectrum (Table 1). The NOESY spectrum revealed correlations between H-3/H-23 and H-3/H-5, which pointed to the β-configuration of the oxygen atom at the C-3 position. In the ¹H–¹H COSY spectrum, the proton signal is correlated with the two proton signals of the methylene group at δ _H 2.57 (br d, 1H, J = 12.2 Hz, H-15b) and δ _H 1.77 (15a).Based on the broad single peak of the hydrogen proton on the methylene carbon in the ¹H-NMR, HMBC and DEPT spectra, it is inferred that the hydroxyl group at position 16 should be in the alpha configuration (Shashi and Mahato, 1994). Taken together, the spectroscopic of the aglycone closely matched that of caulophyllogenin (Strigina et al., 1974).

According to the results of GC-MS analysis, after acid hydrolysis treatment three monosaccharides were identified as D-Glc and L-Rha, and the ratio of them is 2:1. By analyzing ¹H NMR, the hydrocarbon signals of three sugar groups were assigned (Table 1). The ¹H NMR spectrum of 1 exhibited three sugar anomeric protons at δ _H 6.26 (d, J = 8.5 Hz, Glc’ H-1), δ _H 4.99 (d, J = 8.0 Hz, Glc’’ H-1) and 5.88 (s, Rha’ H-1), which showed HSQC correlations with the anomeric carbon signals at δ _C 95.9, 105.0 and 102.8, respectively. The trisaccharide chain was established to be Rha (1→4)-Glc ((1→6)-Glc- by the observation of HMBC correlations from Glc-H-1′ to C-28 of the aglycone, Glc-H-1′′ (δ _H 4.99) to Glc-C-6′ (δ _C 69.3) and Rha-H-1‴(δ _H 5.87) to Glc-C-4′′ (δ _C 78.3). This conclusion was confirmed by the NOESY correlations, as shown in Figure 2. Therefore, the chemical structure of 1 was elucidated as caulophyllogenin-28-O-α-L-rhamnopyranosyl-(1→4)-β-D-glucopyranosyl-(1→6)-β-D-glucopyranosyl ester, named C. Spanion A.

Compound 2 (2) is a white, amorphous substance. When we used compound 2 (2) for Lieberman-Burchard and Molish tests, the results of both experiment were showed positively, indicating that compound 2 may be a saponin (Liu et al., 2023). Its molecular formula was established as C₅₉H₉₄O₂₇ by HR-ESI-MS at m/z 1233.5892 [M - H]^- (calcd. For C₄₈H₇₈O₁₉, 1233.5904). As compound 2 are similar to compond 1, indicating that they may gain similar structures (Table 2). The C-3 carbon was observed at δ_C 83.2, which suggested that the sugar linkage was formed at C-3. The main difference existed in the chemical shifts of H-16 and H-23 (δ _H-16 1.90, 2.06 and δ _H-23 9.97 for 2 and δ _H-16 4.29 and δ _H-23 4.15, 3.69 for 1). Furthermore, the aglycone’s spectroscopic details were found to be highly consistent with those of gypsogenin (Zhou et al., 1989).

By using the same method as described above, the results of GC-MS indicating the existence of D-Glc, L-Rha and L-Ara, and the ratio of them is 3:1:1. Comparing the NMR spectra and extensive 2D NMR studies, we found that 2 and 1 have the same linked trisaccharide chain Aglycone C-28. Through the analysis of HMBC spectrum of 2, Ara was connected to C-3 of the aglycone as a cross-peak exists between C-3 of the aglycone and H-1 of Glc I. Key HMBC and NOESY shows in Figure 3. Therefore, the chemical structure of 2 was elucidated as 3-O-α-L-glucopyranosyl-(1→2)-α-L-arabinopyranosyl-gypsogenin-28-O-α-L-rhamnopyranosyl-(1→4)-β-D-glucopyranosyl-(1→6)-β-D-glucopyranosyl ester, named C. Spanion B.

By correlating the NMR data with the information documented in the existing literature, the known saponins were characterized as 3-O-α-L-arabinopyranosyl-hederagenin-28-O-α-L-rhamnopyranosyl-(1→4)-β-D-glucopyranosyl-(1→6)-β-D-glucopyranosyl ester (3) (Wang et al., 2010), 3-O-α-L-glucopyranosyl-(1→2)-α-L-arabinopyranosyl-betulinic acid 28-O-α-L-rhamnopyranosyl-(1→4)-β-D-glucopyranosyl-(1→6)-β-D-glucopyranosyl ester (4) (Liu et al., 2021), HN-saponin F (5) (Fumie et al., 1990), cauloside C (6) (Anisimov and Chaikina, 2015), Kalopanax saponin G (7) (Miyakoshi et al., 1999), echinocystic acid-28-O-α-L-rhamnopyranosyl-(1→4)-β-D-glucopyranosyl-(1→6)-β-D-glucopyranosyl ester (8) (Xia et al., 2014), oleanolic acid 3-O-β-D-glucopyranosyl-(1→2)-α-L-arabinopyranoside (9) (Sobolev et al., 2000). See Supplementary Data Sheet 1 for relevant data.

The MTT assay was used to determine the cytotoxic properties of compounds 1–9, and the findings are outlined in Table 3. Notably, compounds 6 and 9 demonstrated potent cytotoxic effects (IC₅₀ ≤ 10 μM) against a panel of three human cancer cell lines: U251MG, HepG2, and HL-60. Compound 1 exhibits weak activity against HL-60(IC₅₀ = 56.45 ± 5.18 μM), while compound 2 shows weak activity against U-251MG (IC₅₀ = 76.67 ± 4.25 μM).