Danggui and Yimucao were purchased from Shaanxi Xingshengde Pharmaceutical Co., Ltd. (Xingshengde, Shaanxi, China) and were identified as the root of Angelica sinensis (Oliv.) Diels (No. SNTCM-20200616), and the aerial parts of Leonurus japonicus Houtt. (No. SNTCM-20200617) respectively by Prof. Yu-Ping Tang from the Shaanxi University of Chinese Medicine.

Dried Danggui (100 g), Yimucao (100 g), and DY (200 g, mass ratio was 1:1) were weighed respectively, and break them into powder. 5 times volume of 50% ethanol (v/v) was added to the powdered sample, then soaked for 20 min, ultrasonically extraction for 40 min (temperature 40°C; power 500 w), and then filtered. After extracting the residue twice with the same method, the filtrates were collected. Subsequently, the ethanol in the filtrate was recovered by vacuum distillation to obtain the Danggui, Yimucao, and DY water extract, and the water extract was concentrated for subsequent animal experiments.

Kunming mice (male, 6–8 weeks old, bodyweight 20 ± 2 g; female, 6–8 weeks old, bodyweight 25 ± 2 g) were purchased from Chengdu Dossy Experimental Animals Co., Ltd. (Chengdu, China). All animals were kept in a temperature and light-controlled environment, 12 h light and 12 h dark cycles, and maintained with pathogen-free room with controlled conditions. All experimental procedures were approved by the Animal Care and Use Committee of Shaanxi University of Chinese Medicine. Except for the control group, all other female mice were mated with sexually mature male mice at a ratio of 2:1 overnight to establish pregnancy. The next morning, see if there was a vaginal plug. If there was, it would be designated as the 0.5 days of pregnancy.

All pregnant mice were treated with RU486 (2 mg/kg; diluted with carboxymethyl cellulose; intraperitoneal injection) according to the method in the literature (Lv et al., 2012) on 8.5 days of pregnancy, pregnant mice, and the pregnancy termination rate was 100%. A cotton ball was put into the vagina for monitoring vaginal bleeding, if there was bleeding or embryo excretion, it would be considered as a successful abortion and the next step could be taken. The abortion mice were randomly divided into model group, D (Danggui group 1.04 g/ml), Y (Yimucao group 1.04 g/ml), and DY (Danggui-Yimucao group 2.08 g/ml). From the day of abortion, the mice were sacrificed after continuous administration for 7 days. The serum, spleen, and uterus were obtained respectively, and part of the fresh spleen and uterus were separated for immunohistochemistry, and the rest of samples were stored at −80°C.

The level of progesterone in mouse serum were quantified using ELISA kit according to the corresponding protocol provided by the manufacturer (Elabscience Biotechnology, Wuhan, China), and the absorbance at 450 nm was read in a microplate reader.

T-box expressed in T cells (T-bet) and GATA-binding protein 3 (GATA-3) expression in spleen and uterus tissues were measured by immunohistochemistry. The spleen and uterus tissues of mice were fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned. The sections were processed for immunohistochemical staining with T-bet and GATA-3 monoclonal antibody (Protein Tech Group, Wuhan, China) followed by avidinbiotin based detection kit, then stained with DAB, and re-stained with hematoxylin. The images were observed under a microscope and collected for analysis.

Total RNA from spleen and uterus tissues were extracted using RNAiso plus (Takara, Japan) according to the manufacturer’s instructions, and then cDNA was generated using a reverse transcript kit (Mix) with gDNA remover (SinoMol, China). Quantitative PCR was then performed in an QuantStudioTM3 Real Time PCR System (Thermo Fisher Scientific Inc., Waltham, MA, United States) using TB Green^® Premix Ex Taq™ II (Tli RNaseH Plus) (Takara, Japan). All reactions were carried out in accordance with the manufacturer’s instructions. The GAPDH gene was used as the internal standard gene, and the data were quantitatively analyzed by 2^−ΔΔCT method. The control group was set at 1.0, and all data showed relative mRNA expression (fold change). The sequences of primers used for RT-qPCR were shown in Table 1.

The frozen serum samples were thawed on ice, and each 100 μL of serum was mixed with 300 μL of pre-cooled acetonitrile to precipitate the protein, and then vortexed for 1 min. Each sample was centrifuged at 13,000 rpm/min for 15 min 300 μL of supernatant from each sample was taken and concentrated to dryness by vacuum centrifugation, and then redissolved with 200 μL 10% cold acetonitrile for later use. The quality control (QC) sample was mixed with 20 μL of each serum sample and processed in parallel as above. Before injection, the QC sample was injected in parallel 3 times to adjust or balance the system, and then every 10 samples were injected for further testing of the stability of the analysis system.

UPLC analysis was performed on a ACQUITY UPLC system (Waters Corporation, Milford, MA, United States) that was equipped with SYNAPT G2-Si Q-TOF high-definition mass spectrometer (Waters Corp., Manchester, United Kingdom). Chromatographic separation was carried out on an UPLC BEH C18 column (2.1 × 100 mm, 1.7 µm) at 35°C. We had worked out universal UPLC elution conditions, and the conditions were as follows: mobile phase was composed of A (0.1% formic acid) and B (acetonitrile) under a gradient profile (0–1 min, 90% A; 1–2 min, 90–60% A; 2–8 min, 60–10% A; 8–12 min, 10–5% A; 12–14 min, 5–60% A; 14–15 min, 60–90% A), at a flow rate of 0.3 ml/min and a sample injection volume of 2 μL.

The UPLC system was coupled to a Q-TOF/MS using electrospray ionization (ESI) operated in both positive and negative ion modes. ESI source conditions were as follows: For positive mode, capillary voltage, 3.0 kV; source temperature, 100°C; desolvation temperature, 350°C; desolvation gas flow, 800 L/h; cone gas flow of 50 L/h. For negative mode, capillary voltage, 2.5 kV; source temperature, 120°C; desolvation temperature, 280°C; desolvation gas flow, 600 L/h; cone gas flow of 50 L/h. The MS data were automatically conducted from m/z 50 to 1,000 in the full-scan mode. Leucine encephalin at [M + H]⁺ (m/z 556.2771) and [M − H]⁻ (m/z 554.2615) was used as the lock mass.

All of the raw UHPLC-QTOF/MS data were imported to Progenesis QI software (Waters Corporation) for peak alignment, peak selection, deconvolution and normalization. The statistical analysis was analyzed by principal component analysis (PCA) and orthogonal partial least-squares discrimination analysis (OPLS-DA) with EZinfo 3.0 software (Waters Corporation). The variable importance in the projection (VIP) > 1 in the OPLS-DA model, and t-test (p < 0.05) were used for the screening of potential biomarkers.

The m/z and retention time data provided by UHPLC-MS were analyzed by Progenesis QI software which composed of Progenesis MetaScope, HMDB (http://www.hmdb.ca/), METLIN (http://metlin.scripps.edu/) and ChemSpider (www.chemspider.com) to explore potential biomarkers. MetaboAnalyst (http://www.metaboanalyst.ca), as a web-based tool for visualization of metabolomics, was used to enrich and analyze possible metabolic pathways (Xia and Wishart, 2016).

Comparisons of means were conducted using one-way ANOVA, Student’s t-test and non-parametric test. Data were presented as the mean ± standard deviation (SD). GraphPad Prism version 8.0 (GraphPad Software, United States) was used for graphing and analyses. A value of p < 0.05 was considered statistically significant. Pearson correlation coefficient analysis was used to find the correlation between potential biomarkers and pharmacological indicators. Besides, RT-qPCR data were analyzed using the 2^−ΔΔCT algorithm.

In order to evaluate the recovery of abortion mice, we tested the serum progesterone levels of mice. As shown in Figure 1, the serum progesterone level of abortion mice was significantly higher than that of normal mice. After D, Y, and DY treatments, the progesterone levels of abortion mice were significantly reduced, and the down-regulation degree of DY group was the most obvious. The results showed that both D and Y had a certain regulatory effect on progesterone, and after the two were combined, the regulatory effect on progesterone might rise to a new level.

Immunohistochemical staining showed that D, Y, and DY could all regulate the expression of T-bet and GATA-3 in the spleen of abortion mice. The up-regulation of T-bet in DY group was more obvious, while the down-regulation of GATA-3 in DY and Y groups was more significant (Figures 2A–D). In addition, the DY and Y groups were better in regulating the expression of T-bet and GATA-3 in the uterus (Figures 2E–H). These results indicated that DY might have a better regulating effect on the tilt of Th1/Th2 to Th1.

The results showed that compared with the control group, the mRNA expression of T-bet in the spleen of the model group was significantly decreased, while GATA-3 was significantly increased. After D, Y, and DY treatments, the gene expression of abortion mice had a significant correction. The Y (Figure 3A) group had a more significant up-regulation of T-bet mRNA expression, while the DY (Figure 3B) group had a better inhibitory ability on GATA-3. In addition, compared with the control group (Figures 3D,E), the expression of interferon (IFN)-γ in the model group was suppressed, while the expression of IL (interleukin)-4 was increased. After administration, the IFN-γ in Y group was significantly up-regulated, while the IL-4 in DY group was obviously inhibited. We also tested the expression of T-bet and GATA-3 in the uterus (Figures 3G,H), and the trend was similar to that in the spleen. These results showed that compared with normal mice, the immune balance of the abortion mice was tilted towards Th2, and this tilt could be reversed to Th1 after administration (Figures 3C,F).

Typical base peak intensity (BPI) chromatograms of serum samples, which were collected from the model group and control group in negative and positive modes. As shown in Figure 4, the metabolites with low molecular weight could be separated well within 15 min. The subtle changes between these complex data could be found by using multivariate data analysis techniques, such as PCA and OPLS-DA.

According to the PCA score plot (Figures 5A,B), it could be seen that there was a clear separation between the control, model, D, Y, and DY groups. No matter in positive ion mode or negative ion mode, the model group and the control group had different trends, while D, Y, and DY groups all had different degrees of trends toward the control group, and DY group was closer to the control group.

The OPLS-DA (Figures 5C,D) was performed to further analyze the trend of the control group and the model group in positive ion mode or negative ion mode. It could be found that there were significant differences in the endogenous metabolites between the two groups from S-plots (Figures 5E,F) of OPLS-DA. Potential biomarkers could be screened by the VIP value in the VIP-value plots (Figures 5G,H), and those with a VIP value greater than 1 would be considered qualified. R²Y of the OPLS-DA model in positive and negative modes were 98 and 99%, and Q² were 91 and 93% respectively, which showed that OPLS-DA model was good to fitness and prediction.

First, potential biomarkers with high correlation were extracted from S-plots of OPLS-DA. The VIP value was often used to represent the contribution rate of variable, which was directly proportional to the VIP value. All ions detected by UPLC-MS were ranked in order of VIP value from largest to smallest, VIP > 1 and p < 0.05 were selected as potential biomarkers. The high-precision quasi-molecular ions detected by Q-TOF/MS and MS/MS fragmentation modes were used to calculate the possible molecular formulas of biomarkers. Progenesis MetaScope, HMDB, ChemSpider, and METLIN were performed for the verification of structural information. An endogenous (t_R = 9.16 min, m/z 303.2331) metabolite would be used as the example to illustrate the recognition process: the accurate mass of the potential marker was determined ([M-H]⁻ at m/z 303.2331), the main fragment ions of the marker were observed at m/z 285.2224, 259.2431, 231.2118, 191.1077, which might be representing [C₂₀H₃₀OH]⁻, [C₁₉H₃₁]⁻, [C₁₇H₂₇]⁻, and [C₁₂H₁₇O₂H]⁻. The metabolite was eventually identified as arachidonic acid based on standard references and databases. In the end, a total of 76 metabolites were identified as potential biomarkers, as detailed in Supplementary Table S1 and we listed 20 of these biomarkers in Table 2. Compared with the control group, the level of serum metabolites in the model group increased or decreased to varying degrees, and the level of metabolites in abortion mice was significantly reversed after administration, especially in the DY group.

In order to analyze the effect of DY on the metabolic pathways of medical abortion mice, the metabolites identified above were introduced into MetaboAnalyst (https://www.metaboanalyst.ca/) and then pathways were constructed (Figure 6). Among them, the glycerophospholipid metabolism, Arachidonic acid metabolism, and alpha-linolenic acid metabolism, which had impact values of 0.22, 0.33, and 0.33, respectively, were selected as the most critical metabolic pathways.

As the main lipid component of cell membranes, glycerophospholipids directly affected the physiological functions of cells and were the basis for the formation of dynamic subcompartments in cell membranes, and were considered as a key molecule in cell signaling, homeostasis maintenance, inflammation and immune response (Wang et al., 2021a). They played a vital role in cell proliferation, differentiation and apoptosis. The concentration of glycerophospholipids affected the transformations in cell membrane composition and permeability. Therefore, the fluctuation of glycerophospholipids content reflected the disorder of lipid metabolism and was an important biological indicator (Wang et al., 2019b). Arachidonic acid, also known as eicosa-5, 8, 11, 14-tetraenoic acid, was a ω-6 polyunsaturated fatty acid, which mainly existed in the cell membrane in the form of phospholipids. When cells were in a state of stress, arachidonic acid was released from phospholipids as free arachidonic acids through phospholipase A₂ and phospholipase C, and became the precursor of pro-inflammatory bioactive mediators by three metabolic pathways. Through the cyclooxygenase (COX) pathway, arachidonic acid could be metabolized into prostaglandin (PG) and thromboxane. Arachidonic acid could also be converted into leukotrienes and lipoxins through the lipoxygenase pathway. In addition, arachidonic acid also produced epoxyeicosatrienoic acid or hydroxyeicosatetraenoic acid through the cytochrome P450 pathway. These arachidonic acid metabolites were collectively called eicosanoids, which were effective autocrine and paracrine bioactive mediators and widely participated in a wide range of physiological and pathological processes (Wang et al., 2019c). There were three main metabolic pathways of alpha-linolenic acid, including ATP production and carbon cycle of β-oxidation, incorporating into glycerides within different tissues depots and converting into Long chain n-3 (Picklo and Murphy, 2016). As an arachidonic acid antagonist, increased intake of alpha-linolenic acid could lead to a decrease in arachidonic acid content and might further reduce the biosynthesis of pro-inflammatory eicosanoids, including PGE2, leukotrienes, and thromboxanes (Li et al., 2017).

The Pearson correlation coefficient analysis method was used to find the correlation between potential biomarkers and biochemical indicators, as shown in Figure 7. Blue indicated positive correlation, while red indicated negative correlation, and the stronger the correlation, the darker the color. Progesterone had a strong positive correlation with metabolite 1 (r = 0.71) and GATA-3 (r = 0.51), and a strong negative correlation with metabolite 18 (r = −0.53) and T-bet (r = −0.51). T-bet had a strong positive correlation with metabolite 2 (r = 0.64), and a strong negative correlation with metabolite 1 (r = −0.62). GATA-3 had a strong positive correlation with metabolite 11 (r = 0.79), and a strong negative correlation with metabolite 13 (r = −0.70). The correlation coefficient was detailed in Supplementary Table S2.