The ZYP product and the powders of nine kinds of medicinal materials, namely, CCL, PGC, DIW, EUO, TCD, MOF, CPN, AMK, and ACC were provided by Guangzhou Baiyunshan Zhongyi Pharmaceutical Co., Ltd (Guangzhou, Guangdong Province, China). HPLC-grade acetonitrile, methanol, and formic acid were purchased from Merck (Darmstadt, Germany). Purified water (18.2 MΩ cm at 25°C) was prepared using a Milli-Q purification system (Millipore, Bedford, MA, United States). RPMI-1640 medium, penicillin–streptomycin, fetal bovine serum (FBS), 0.25% trypsin-EDTA, DMSO, chloroform, isopropanol, ethanol, and reagents used for reverse transcription (RT) were purchased from Thermo Fisher Scientific Inc., and 70% t-BHP was purchased from Macklin Inc. Reactive oxygen species assay kit and DEPC-treated water were bought from Beyotime Biotechnology.

The extraction method of ZYP product is the same as previously described (Cao et al., 2020). The UHPLC system (Dionex UltiMate 3000 ×2 Dual RSLC, Thermo Fisher Scientific Inc.) was used for two-dimensional liquid sample analysis and preparation, which included the DGP-3600RS dual system pump, WPS-3000TRS autosampler, TCC-3000RS column thermostat, SRD-3600 degasser, RS variable wavelength UV detector, and Chromeleon chromatography management software (SR7.2).

For first-dimensional (1D) LC sample analysis and preparation, an amino column (Waters XBridge BEH Amide column 130 Å, 4.6 × 150 mm, 3.5 μm) served as the first-dimensional preparation column at 40°C column temperature. Following a gradient elution program, a binary mobile phase consisting of 0.1% aqueous formic acid (A) and acetonitrile (B) was utilized at a flow rate of 1.0 mL/min: 0–10 min: 95%–90% (B); 10–20 min: 90% (B); 20–55 min: 90%–60% (B); and 55–60 min: 60% (B). The UV wavelength was set at three channels, 203, 254, and 280 nm, respectively. The 1D LC collected the fractions every 3 min using the automatic partial collector and took 20 consecutive injections of 40 μL injection volume. The fractions were combined separately, concentrated, reconstituted with 200 μL of acetonitrile, and refrigerated at 4°C as the second-dimensional (2D) LC sample.

The 2D LC analysis used the C18 column (Thermo Fisher Hypersil GOLDTM C18, 150 mm × 2.1 mm, 1.9 μm) at 45°C column temperature. The mobile phase was the same as 1D LC, but the flow rate was 0.4 mL/min and the elution gradient differed as follows: 0–5 min, 5% (B); 5–10 min, 5%–20% (B); 10–55 min, 20%–70% (B); 55–60 min, 70%–100% (B); and 60–65 min, 100% (B). The UV wavelength was set at three channels, 203, 254, and 280 nm, respectively.

To analyze the 20 fractions obtained, a Thermo Fisher LTQ-Orbitrap XL mass spectrometer (Thermo Fisher Scientific) equipped with an electrospray ionization (ESI) source was connected to the UPLC system. The ESI source was operated in positive and negative modes. The other MS conditions were set as follows: ion-spray voltage, 4.5 kV/−4.0 kV; capillary temperature, 380/380°C; capillary voltage, 35/−35 V; and tube lens voltage, 100/−100 V in the positive/negative mode. Xcalibur 2.1 software (Thermo Fisher Scientific) was used for data acquisition and processing.

Compound Discoverer software was used to extract peak area and retention time from mass spectrometry data, and the analysis procedure was carried out as illustrated in Supplementary Figure S1. The data were divided into four groups based on methanol extraction/water extraction and positive/negative ions and then imported into the software for analysis, with the matching mass spectrometry peak areas, retention time, and MW information predicted by the software. After obtaining mass spectral peak area data, the data were imported into the MetaboAnalyst website and processed through the statistical analysis module to produce standardized peak area data.

After analyzing the mass spectrometry data, they were combined with the data from the single pharmacological model and uploaded into SIMCA software. The peak area was taken as x, and the pharmacological model data including the survival rate, scar reduction rate, fluorescence reduction rate in the HTR-8/SVneo cell model, and mean levels of E2, FSH, LH, AMH, Tnf-α, and Ifn-γ in the premature ovarian failure mouse model and mean levels of Mmp9, Timp3, Lif, and Egf in the endometrial receptivity disorder mouse model were taken as y. The PLS model was used to fit the data and determine the VIP and coefficient values. VIP was greater than one, with a significance level of 0.05. The top 15 were screened and compared using Compound Discoverer software with retention time and MW, and the corresponding compounds were inferred from the MS/MS data.

All data were expressed as the mean standard deviation (mean ± SD). GraphPad Prism 8 software was used for data statistics and graphing, and one-way ANOVA was used to compare the two groups. The statistical significance was defined as p < 0.05.

In the creation of a two-dimensional liquid system, the choice of the chromatographic column is critical. To maximize column capacity and improve separation efficiency, the chromatographic column should satisfy orthogonality as much as possible. According to our orthogonal column tests, the amino column was employed in the first-dimension chromatography separation of fractions to increase the retention and separation ability of strong polar compounds, as well as to improve the orthogonality. The classic C18 chromatographic column has a high ability to separate and differentiate weakly polar compounds and is mass spectrometry compatible. As a result, the C18 column was chosen as the second column in this investigation. When comparing the organic phases of acetonitrile and methanol, acetonitrile produces a more stable baseline and detects more mass spectral peaks with better peak shape. In addition, adding 0.1% formic acid to the water phase can improve the ionization intensity, peak shape, and column efficiency. As a result, the acetonitrile–0.1% formic acid water system was chosen for this study.

According to the separation results of first-dimension liquid chromatography, a fraction was taken every 3 min for the whole retention duration of 60 min, yielding a total of 20 fractions, which were concentrated by nitrogen blowing and freeze-dried. As shown in Figure 1, first-dimension liquid chromatography separated 20 fractions, which were re-dissolved in acetonitrile and evaluated using second-dimension liquid chromatography. Supplementary Tables S2, S3 provide the list of 589 chemicals that can be found in ZYP, of which 221 were found in the positive ion mode of the mass spectrometry and the remaining 368 were found in the negative ion mode. Among them, hyperoside, astragalin, quercetin, kaempferol, isorhamnetin, ginsenoside Rg1, ginsenoside Rb2, ginsenoside Rb3, ginsenoside Rd, ginsenoside F2, ginsenoside Rg2, akebia saponin D, sweroside, chlorogenic acid, 3,5-dicaffeoylquinic acid, 3,4-dicaffeoylquinic acid, 4,5-dicaffeoylquinic acid, pinoresinol di-O-β-D-glucopyranoside, geniposidic acid, nystose, atractylenolide I, atractylenolide II, atractylenolide III, ferulic acid, 5-hydroxymethylfurfural, 2,3,5,4′-tetrahydroxystilbene-2-O-β-Dglucoside, emodin, jaceosidin, eupatilin, and protocatechuic acid were confirmed by standard materials; other compounds were speculated by comparing with the literature report. Furthermore, although 139 compounds found by 2DLC-MS are inconsistent with the data of the reported ingredients, they had the same molecular mass as certain known compounds. This is why the word “isomer” is added when they are reported in Supplementary Tables S2, S3, indicating that they may be new compounds.

In addition to that, the spectra of the 27 orthogonal groups of methanol and water extracts were analyzed using UHPLC-MS, and the positive and negative mode results are shown in Supplementary Table S2, S3. Afterward, in vitro and in vivo tests were conducted on 27 methanol extracts and 27 water extracts, respectively, and the spectrum–effect connection was then merged.

HTR-8/SVneo cells were incubated with 100, 50, 25, and 12.5 μg/mL ZYP orthogonal groups for 24 h as pre-protection and then treated with t-BHP, WoundMaker, and ROS kit for model building. The results are shown in Supplementary Figures S3, S4, S5. Since the 50 μg/mL concentration had the best overall efficacy among the four concentration groups, the mean value of the 50 μg/mL result was used for the subsequent spectrum–effect relationship analysis. As demonstrated in Figure 2A, t-BHP modeling reduced cell viability by an average of 19.8%, but incubation with 50 μg/mL ZYP orthogonal groups 24 h before modeling increased cell viability to various degrees in each group. The cell survival of the orthogonal test groups 3, 4, 5, 6, 7, 8, 10, 11, 12, 13, and 14 was considerably improved (p < 0.05) compared to that of the model group.

In the migration model, all orthogonal groups demonstrated higher HTR-8/SVneo cell wound healing than the control group, as shown in Figure 2B, with groups 1, 2, 7, 16, 17, 18, 19, 21, and 23 being significant (p < 0.05). Afterward, the ROS model results shown in Figure 2C indicated that most of the orthogonal groups were able to reduce ROS, except for groups 7–14. In comparison to the model group, the orthogonal groups 20, 21, and 22 showed a considerable effect on reducing ROS (p < 0.05).

Figure 3 depicts the impact of 27 ZYP orthogonal groups on endometrial receptivity disease model mice. Mmp9, Timp3, Lif, and Egf mRNA in the mouse uterine tissue could be decreased after modeling. Following administration, some orthogonal groups demonstrated an inhibition of damage caused by modeling medicines, indicating a protective effect. The experimental results showed that groups 17 (p < 0.001) and 18 (p < 0.01) significantly increased Lif mRNA expression in the uterus of endometrial receptivity disorder model mice, implying that the protective action of this part of the administration group may be associated with Lif-related targets. The amount of Egf mRNA in the uterus of mice in (p < 0.001) administration group 18 was likewise considerably greater than that in the model group, implying that administration group 18 (p < 0.001) may exercise its medicinal effect via modulating growth factor-related pathways.

Figure 4 illustrates the protecting impact of ZYP 27 orthogonal test groups on a mouse model of premature ovarian failure. The levels of E2 and AMH in the serum of mice were dramatically reduced after continuous modeling treatment with cyclophosphamide, whereas the levels of LH and FSH were significantly increased. As shown in Figure 4A, the serum E2 levels of mice in groups 16, 22 (p < 0.01), 26 (p < 0.001), and 27 (p < 0.05) were considerably greater than those in the model group after treatment, implying that ovarian function is protected by inhibiting E2 reduction. Figure 4B indicates that groups 4, 11, 20, 22, and 23 (p < 0.05); 7 (p < 0.01); and 2, 5, 9, 10, 12, and 19 (p < 0.001) were able to significantly increase cyclophosphamide-caused AMH reduction, implying that these groups may be able to prevent premature ovarian failure by maintaining or increasing AMH release. Cyclophosphamide could considerably raise FSH and LH secretion levels, as shown in Figures 4C, D, which is consistent with the symptoms of premature ovarian failure. Most administration groups exhibited an impact of suppressing excessive FSH and LH secretion after delivery. In the mouse ovarian tissue, cyclophosphamide can generate an increase in the expression levels of Tnf-α and Ifn-γ mRNA, as demonstrated in Figures 4E, F, and most of the administration groups showed a downward trend. However, because the expression levels of Tnf-α and Ifn-γ genes in the mouse ovarian tissue were low or not significantly activated, they were below the limit of quantification, or the divergence within the group was too great, and significant differences could not be acquired in the real experimental detection.

After the UHPLC-MS and pharmacological experiments of the 27 ZYP orthogonal groups were performed, the quality markers could be inferred by combining the spectrum–effect relationship. Figures 5A, B, C, D show the results of PCA of 27 orthogonal groupings of mass spectrum peaks. The acquired VIP value data are shown in Figures 5E, F when integrated with one single pharmacological model for PLS fitting. A total of 39 potential quality markers were predicted, including 17 in the positive ion mode and 22 in the negative ion mode. Dipsacus saponin A and codonopsinol B are strong predictors in both the HTR-8/SVneo cell model of t-BHP oxidative damage and the premature ovarian failure mouse model.