Acetonitrile and formic acid (MS-grade) were purchased from Fisher Scientific (Loughborough, United Kingdom). Leucine-enkephalin and Major Mix calibration fluid were purchased from Waters Corp. (Milford, MA, United States), and distilled water was purchased from Watson (Guangzhou, China). Other analytical-grade reagents were purchased commercially (Beijing, China).

One hundred gypenoside reference standards (purity >95%) were previously isolated in our laboratory, and their structures were confirmed by MS and NMR (some of which were literature data) (Li et al., 2021; Zhang et al., 2021) (Supplementary Figures S1 and S2; Supplementary Table S1).

Forty-four batches of G. longipes were collected from the Maobaling Planting Base in Pingli county, Shaanxi province, China, and identified by Prof. Baolin Guo of the Institute of Medicinal Plant Development, Chinese Academy of Medical Sciences, and Peking Union Medical College, Beijing, China. In total, 15 rhizome samples had been grown within a year, labeled R1–R15; the 15 stem samples were all uniform (20 cm removed from the top and cut down to 30 cm in length) from the longest branch in the plant, labeled S1–S15; and 14 leaves from same branches, labeled L1–L14.

Stock solutions of the 100 reference standards were prepared in methanol at a concentration of 0.3 mg/mL and pipetted in the same amount, and 14 mixed reference solutions were prepared according to the grouping information in Supplementary Table S2. All the solutions were stored at 4 °C for further study.

Forty-four batches of samples were crushed evenly until they passed through a 40-mesh sieve. Then, a fine powder sample was accurately weighed to 0.25 g and added to 10 mL of 70% aqueous methanol. The conical flask was tightly plugged, shaken, and weighed. After sonication for 30 min, the sample was cooled to room temperature, and the weight loss was compensated for using extraction solution. All the solutions were filtered through a 0.22-μm filter membrane before PCA analysis. One milliliter of the supernatant solution from each batch of samples was blended for use as a quality control sample (QC) and applied between every six injections during the UHPLC-Q-TOF-MS analysis to monitor system stability.

Simultaneously, 1 mL of the supernatant solution was mixed into different parts to prepare the QC samples for each plant part, labeled QC-R, QC-S, and QC-L. Then, QC-R, QC-S, and QC-L were concentrated at reduced pressure to dryness (distillation temperature ≤45 °C), dissolved in 1 mL of 70% methanol, and passed through a 0.22-μm filter membrane before one-dimensional analysis.

One-dimensional hydrophilic chromatography (1D HILIC) and two-dimensional reverse-phase chromatography (2D RPC) were combined to establish an offline 2D-LC system for separating the chemical compounds in G. longipes. 1D HILIC was performed using a Vanquish Core system (Thermo Fisher Scientific, Germany) equipped with a dual pump, autosampler, column compartment, charged aerosol detector (CAD), and diode array detector (DAD). A YMC-Triart Diol-HILIC column (4.6 mm × 250 mm, 5 μm) was used at a column temperature of 40 °C. The mobile phase comprised water with 0.2% formic acid (A) and acetonitrile (B). The gradient protocol used was as follows: 0–5.0 min, 95%–90% B; 5.0–20.0 min, 90%–77% B; 20.0–25.0 min, 77%–70% B; 25.0–30.0 min; 70%–55% B; 30.0–32.0 min, 55% B; 32.0–33.0 min, 55%–95% B; and 33.0–40.0 min, 95% B. The flow rate was 1 mL/min, the sample injection volume was 5 μL, and the experiment was repeated 20 times. CAD was used to guide separation, the evaporator temperature was set to 35 °C, and the nitrogen gas pressure for the nebulizer was set at 0.6 MPa. The data collection rate was set at 5 Hz using a filter constant of 5 s. The DAD was used to monitor the ultraviolet signals at 254 and 203 nm to guide fractionation. The sub-fractions were collected via valve switching: fraction 1: 0–6 min, fraction 2: 6–8.9 min, fraction 3: 8.9–11.6 min, fraction 4: 11.6–15.2 min, fraction 5: 15.2–19 min, fraction 6: 19–26.8 min, and fraction 7: 26.8–48 min. When combining the same fractions, the distillation temperature was maintained under 45 °C, and the samples were dried to remove the solvent at reduced pressure. The seven fractions were sequentially dissolved in 1 mL of 70% methanol and passed through a 0.22-μm filter membrane before two-dimensional analysis.

The 2D RPC measurements were performed using a Waters ACQUITY I-Class system (Waters Corp., Milford, MA, USA). A Waters ACQUITY UPLC CSH C18 column (2.1 mm × 100, 1.7 μm) was used at a column temperature of 40 °C. The linear mobile phase was water containing 0.1% formic acid water(A) and 0.1% formic acid acetonitrile (B). The gradient protocol was as follows: B was started at 10%, increased to 25% at 2 min, 32% at 7 min, 33% at 10 min, 35% at 12 min, 46% at 22 min, 56% at 27 min, and finally increased to 95% at 30 min. The flow rate was 0.5 mL/min. The sample injection volume was 1 μL.

A VION IMS-QTOF system (Waters Corp., Wilmslow, United Kingdom) was used to acquire the HDMS^E data in both negative- and positive-ion modes, and the data were acquired from 50 to 2,000 Da. The electrospray ionization (ESI) source temperature was set to 110 °C, the desolvation gas was nitrogen (flow rate: 850 L/h), and the temperature was 450 °C. The capillary voltages were 3 kV (positive-ion mode, ESI⁺) and 2.5 kV (negative-ion mode, ESI⁻), and the cone voltage was 50 V. In a low-collision energy (CE) scan, the CE was 4 eV; in a high-CE scan, the ramp CEs were 50–70 and 15–25 eV in negative- and positive-ion modes, respectively. Leucine-enkephalin was used as the lock mass at a flow rate of 10 μL/min for calibration, whereas Major Mix was used to calibrate the CCS values. The instrument was controlled using UNIFI 1.9.4 (Waters Corp., Milford, MA, USA).

Major Mix calibration kit was used for CCS calibration and was injected at the beginning of the sequence and between every six injections to monitor the variation in the measured CCS values and ensure that it was less than 2%. One hundred reference standards were detected using HDMS^E data acquisition mode, to measure the CCS values for [M + H]⁺, [M + Na]⁺, [M + NH₄]⁺ in positive-ion mode, and [M−H]⁻ and [M + HCOO]⁻ in negative-ion mode. Each compound was measured three times, and the average value was obtained as the CCS value of the compound. CCSondemand (https://ccs.on-demand.waters.com/) was used to predict the CCS values of the gypenosides in the “Herba Gynostemma mass spectrometry database” (Zhou et al., 2020).

The analysis method was as follows. The maximum allowed number for peak detection was 10,000 for 2D peak detection, and the peak intensity threshold was 100 counts for high-energy and 1,000 counts for low-energy 4D peak detection. Mass and fragment errors were both set at 10 mDa for compound identification. Adducts included + HCOO⁻, −H and +H, +Na, and +NH₄ in negative- and positive-ion modes. Leucine-enkephalin was used to ensure mass accuracy, for which [M−H]⁻ 554.2620 and [M + H]⁺ 556.2766 were used in negative- and positive-ion modes, respectively. The “Common Fragment Settings” and “Common Neutral-Loss Settings” were based on the fragments of gypenosides in the mass spectrum.

Forty-four batches of sample data were reprocessed through deconvolution, alignment, and data reduction to provide a list of mass and retention time (Rt) pairs, along with the corresponding peak areas for all the detected peaks from each file in the dataset. The processed data list was then imported into SIMCA 14.1 (Sartorius Stedim, Germany) for PCA, HCA, and OPLS-DA analyses. The parameters used in the analysis were 0–30 min for the range of Rt, 50–2,000 Da for the mass range, 0.02 Da for the mass tolerance, and 0.10 min for Rt tolerance. All test groups were discriminated using PCA and HCA to determine whether the different groups could be separated. Then, OPLS-DA was carried out to discriminate the ions that contributed most to the classification of the samples, and a VIP-plot was constructed to show the variables that contributed to the classification. Heatmap visualization was performed using TBtools to show changes in the variable marker content (Chen et al., 2020).

Two-dimensional liquid chromatography significantly improves the separation ability and peak capacity of ions and is helpful for the total component analysis of natural products. Reverse-phase chromatography (RPLC) is suitable for the separation of saponins, and HILIC and RPLC show good orthogonality. Therefore, an offline 2D-LC condition was established by optimizing the 1D method in the HILIC column and the 2D method in the RP-C18 column. The gypenosides are the main components of G. longipes; therefore, six RP-C18 columns were used to evaluate the separation ability for the gypenosides of G. longipes, including a Waters ACQUITY UPLC BEH C18 column (2.1 mm × 100, 1.7 μm), ACQUITY UPLC HSS C18 column (2.1 mm × 100, 1.8 μm), ACQUITY UPLC CSH C18 column (2.1 mm × 100, 1.7 μm), ACQUITY UPLC HSS T3 column (2.1 mm × 100, 1.8 μm), ACQUITY UPLC BEH Shield RP18 column (2.1 mm × 100, 1.7 μm), and Phenomenex Kinetex XB-C18 column (2.1 mm × 100, 1.7 μm). Comparing the separation ability and chromatographic capacity, the CSH C18 column was the most suitable for G. longipes. At the same time, the gradient elution conditions, mobile phases (water–acetonitrile; 0.1% formic acid in water–acetonitrile, 0.1% formic acid in water–0.1% formic acid in acetonitrile, 10 mM ammonium formate in water–acetonitrile, and 10 mM ammonium formate in water–0.1% formic acid in acetonitrile), and column temperature (25, 30, 35, 40, and 45 °C) were also optimized to achieve better peak shape and separation. Four HILIC columns were used to evaluate the orthogonality between the 1D and 2D methods by calculating the linear regression correlation coefficient (R ²) (Qiu et al., 2015; Yang et al., 2016) of the relative retention times of the 1D and 2D separations for 39 reference standards (numbered as GL1-21 and JGL1-20 in Supplementary Table S1) of G. longipes. As shown in Supplementary Figure S3, the separation performance of the four HILIC and CSH C18 columns was similar. Of the obtained values, the R ² value (R ² = 0.1159) of the YMC-Triart Diol-HILIC column (Figure 1) was the lowest when using a mobile phase composed of water with 0.2% formic acid (A) and acetonitrile (B). Finally, the YMC-Triart Diol-HILIC column having a column temperature of 40 °C with 0.2% formic acid water–acetonitrile and the CSH C18 column having a column temperature of 40 °C with 0.1% formic acid water–0.1% formic acid acetonitrile was selected as optimum conditions for the 1D and 2D methods, respectively.

To obtain a better MS response and richer mass fragments, different capillary voltages (1, 1.5, 2, 2.5, 3, and 3.5 kV), ESI source temperatures (100, 110, 120, 130, 140, and 150°C), desolvation gas temperatures (350, 400, 450, and 500°C), and CEs in both negative- and positive-ion modes were optimized. Using the chromatographic peak area as an index, nine reference standards (JGL-19, JGL-6, JGL-4, GL-11, GL-20, GL-7, GL-21, GL-14, and GL-15) were selected to evaluate the responses of the gypenosides. Each parameter was determined in triplicate for reliability. The responses of the main gypenosides were the best when the source temperature was 110°C, desolvation gas temperature was 450 °C, and capillary voltages were 2.5 and 3 kV in negative- and positive-ion mode, respectively. The CE was found to have a significant impact on the mass fragments. Based on fragment richness, CEs of 10–80 eV were investigated sequentially in 10-eV steps in negative-ion mode. However, triterpene saponins are prone to fragmentation in positive-ion mode; thus, the CE was investigated as 5–40 eV in ESI⁺ in 5-eV intervals. Finally, the ramp CE was set to 50–70 eV in negative-ion mode but 15–25 eV in positive-ion mode.

Studying the mass fragments of reference standards in both negative- and positive-ion modes and summarizing the fragmentation patterns is helpful for identifying unknown components. Thirty-nine reference standards for triterpene saponins isolated from G. longipes (numbered as GL1-21 and JGL1-20 in Supplementary Table S1) were classified according to the structure of the C17 branched chain, and components with representative structures were selected to study their mass fragments.

GL-1 is a triterpene saponin containing a type-A aglycone. In ESI⁻ mode, it produced an adduct ion at m/z 959.5186 [M + HCOO]⁻ at low CE and a deprotonated molecular ion at m/z 913.5141 [M−H]⁻, as well as a series of glycosyl cleavage fragment ions at m/z 781.4735 [M−H−Xyl]⁻, 635.4166 [M−H−Xyl−Rha]⁻, 473.3635 [M−H−Xyl−Rha−Glc]⁻, and 389.3058 [M−H−Xyl−Rha−Glc−C₅H₈O]⁻ at high CE. In ESI⁺, the main fragment ions of GL-1 were m/z 937.5163 [M + Na]⁺, 932.5606 [M + NH₄]⁺, 457.3679 [M + H−Xly−Rha−Glc−H₂O]⁺, 439.3575 [M + H−Xly−Rha−Glc−2H₂O]⁺, and 421.3466 [M + H−Xly−Rha−Glc−3H₂O]⁺. Based on the mass spectra (Figure 2), the [M + HCOO]⁻ adduct ions were the main ion fragments in low-CE ESI⁻. In high-CE ESI⁻, the deprotonated precursor ion was detected, and a xylose, rhamnose, and glucose unit connected at C-3 were successively removed from the outside to the inside to obtain the aglycone ion. The cyclic side chain of C-17 was broken and formed a neutral-loss fragment at m/z 84.0575. In positive-ion mode, adduct ions of [M + Na]⁺ and [M + NH₄]⁺ were obtained. Even at a low CE, aglycone fragments and a series of dehydrated ions were detected, but the response of glycosyl cleavage fragments was relatively low.

Aglycones B and C are gypenosides whose C-17 side chains were converted into five-membered oxygen-containing heterocyclic structures. The aglycone of JGL-4 belongs to type B. As shown in Supplementary Figure S4A, the cyclic side chain of C-17 was broken and formed a neutral-loss fragment at m/z 100.0524 (C₅H₈O₂) because of a hydroxyl group substituted at C-26 in high-CE ESI⁻. In positive-ion mode, the characteristic aglycone ions were at m/z 487, 469, 451, and 433, and the fragment in which the side chain of C-17 was completely cleaved was observed at m/z 313 Da. In ESI⁻, GL-14 (the type-C aglycone) produced an adduct ion at m/z 999 [M + HCOO]⁻. Owing to an acetyl group connected to glucose, it initially formed a neutral-loss fragment of 42 Da. The glycosyls were then cleaved to the aglycone ion at m/z 471, and the side chain of C-17 was broken and formed a neutral-loss fragment at m/z 112.0524 (C₆H₈O₂). In ESI⁺, a series of aglycone ions were observed at m/z 455, 437, and 419 (Supplementary Figure S4B).

Aglycones of D and E were analyzed using the same method: in negative-ion mode, the aglycone ion was observed at m/z 475 for both D and E, whereas in positive-ion mode, the aglycone ions were observed at m/z 459, 441, and 423 for D and m/z 441, 423, and 405 for E.

The F-aglycone is a gypenoside with a straight chain on C-17. As shown in Supplementary Figure S4C, GL-7 produced an adduct ion at m/z 1107 [M + HCOO]⁻ and a doubly charged ion at m/z 576 at a low CE in ESI⁻. In high-CE ESI⁻, the fragment ions were observed at m/z 929, 783, 621, 459, and 375, attributed to the loss of one xylose, one rhamnose, two glucose residues, a formula of C₆H₁₂, and a neutral-loss fragment of 84.0939 Da because of the side-chain cleavage of C-17. In ESI⁺, GL-7 produced adduct ions at m/z 1085 [M + Na]⁺ and 1080 [M + NH₄]⁺ and a series of aglycone ions at m/z 443, 425, and 407. Owing to the different substituents on C-10 and the side chain of C-17, the neutral-loss fragment and aglycone ion also changed. For instance, JGL-22 (type G) is a gypenoside with hydroxyl substitutions at C-23 and C-24, respectively. Thus, the neutral-loss fragment was observed at m/z 118 (C₆H₁₄O₂) in ESI⁻, and the aglycone ions were observed at m/z 459, 441 and 423, in ESI⁺.

JGL-18 is a triterpene saponin with a type-H aglycone, whose C-17 side chain is a six-membered heterocyclic oxygen-containing ring. In high-CE ESI⁻, the characteristic neutral-loss fragment was observed at m/z 84.0211 (C₄H₄O₂). In high-CE ESI⁺, the aglycone ions were observed at m/z 471, 453, and 435 Da, and the fragment ion was identified at m/z 341(C₇H₁₀) because of the cleavage of the oxygen-containing heterocycles (Supplementary Figure S4D).

In ESI⁻, gypenosides often produce a [M + HCOO]⁻ adduct ion, and a doubly charged ion can also be obtained at low CE when a gypenoside is connected to two glycosyl chains. In high-CE ESI⁻, the deprotonated molecular ion, neutral-loss fragment ions of the glycosyls, and side-chain cleavage of C-17 were detected. In ESI⁺, aglycone ions were the major fragments, always exhibiting a high response. Therefore, based on the fragment ions obtained in negative-ion mode, the molecular formula, type of sugar moieties, glycosyl connecting position to the aglycone, and structure of C-17 were determined. Further, combined with the fragment ions obtained in both negative- and positive-ion modes, the aglycone could be identified.

Forty-four batches of G. longipes comprising different parts of the plant were analyzed using 1D-RPLC/IM-QTOF-MS. Through peak extraction, alignment, and normalization, the MS data were preprocessed and imported into SIMCA for HCA and PCA. HCA was used to divide the samples into specific groups according to similarity criteria, as shown in Figure 7, where the samples are clearly divided into three categories: the rhizomes, stems, and leaves were classified as different categories, indicating that the samples from different plant parts showed significant variation. The PCA analysis is shown in Figure 8, where each point represents an individual sample. Samples were clustered into three groups representing the different parts, and the results are similar to those of HCA.

To clarify the differences between the underground and aerial parts, rhizomes (Group 1) and stems and leaves (Group 2) were selected for OPLS-DA. Groups 1 and 2 were distributed on the negative and positive sides of the t1 axis, respectively, indicating significant differences between the two groups (Figure 9). The parameters were R2X = 0.642, R2Y = 0.966, and Q2 = 0.961, indicating that 64.2% of the variables (R2X) in the model explained 96.6% of the difference (R2Y) between the two groups, and the predictive ability of the model was 96.1%. After 200 permutations tests, the intercept of R2 was 0.283 and the intercept of Q2 was −0.393, both lower than the theoretical values (R2 ≤ 0.3, Q2 ≤ 0.05), indicating that the model was not over-fitted. Based on the OPLS-DA and VIP plots, the variables with VIP values greater than 6.5 were considered as potential chemical markers. Thirty-six differential components were identified, including two flavonoids and 34 triterpenoid saponins, 20 of which were substituted with a malonyl group and (or) acetyl group. A heatmap (Figure 10) was used to visualize the trends in these 36 differential components, and 32 gypenosides were found to be higher in content in the rhizomes. These results combined with the results of total component identification indicate that the composition and content of chemical components in the rhizomes are both higher than in the other parts of the plant, and the structures of these differential compounds are shown in Figure 11. Overall, these findings indicate that the rhizomes are a medicinally important part of G. longipes. As reported in the literature (Teng et al., 2022), the total saponins in the underground part of G. pentaphyllum from Shaanxi province also have good ability to reduce blood lipids; therefore, both the aerial and underground parts of G. longipes are recommended for medicinal use.