Acetonitrile and methanol (HPLC grade) were purchased from Fisher (United States and Trinidad, respectively). Formic acid was purchased from Sigma Aldrich Co. (St Louis, United States). Double-distilled water was purified using a Millipore water purification system (Millipore, Bedford, MA). Other chemicals were of analytical grade. Liquorice (Glycyrrhiza uralensis Fisch., Batch number SAA11) was collected from Neimenggu Province, China and from the different habitats of TwHF (Table 1). The herb was authenticated by Professor Kui-Jun Zhao (Capital Medical University, Beijing, China).

The metabolic profiles of the biofluids were analysed via LC-IT-TOF/MS (Shimadzu Japan). The sample sequence was random and 5 μL of each sample was injected onto an Inertsil ODS-SP C18 (4.6 × 150 mm, 5 μm) column at a column temperature of 40 °C. The mobile phase of solvent A (water with 0.1% formic acid) and solvent B (acetonitrile with 0.1% formic acid) was separated by a 50 min linear gradient. The following gradient was used: a linear gradient of 3% B over an initial 3.0 min, 3%–15% B over 3.0–8.0 min, 15%–25% B over 8.0–15.0 min, 25%–40% B over 15.0–25.0 min, 40%–50% B over 25.0–40.0 min, 50%–75% B over 40.0–45.0 min, and 75%–100% B over 45.0–50.0 min. The flow rate was set as 1 mL/min. The detection wavelength was 275 nm, and the eluent was directly introduced into the mass spectrometer. To ensure the stability and repeatability of the system, 10 μL of each sample was combined to form a quality control (QC) sample, and every five samples were inserted and analysed.

For mass spectrometry, LC-IT-TOF/MS with an electrospray ionization source (ESI) in both positive and negative mode was used. The combined use of both positive and negative ionization LC-MS offers more comprehensive metabolome coverage than the use of a single polarity. The electrospray source parameters were fixed as follows: electrospray capillary voltage 3.5 kV in negative ionization mode and 4.5 kV in positive ionization mode. The mass range was set from m/z 50 to 1,200. The gas temperature was 200 °C in negative ionization mode and 225 °C in positive ionization mode. Gas flow was 1.5 L/min. Nozzle voltage was 500 V in both negative and positive mode.

The initial and final retention times were set for data collection. Data were imported into Metabolists 4.0 (http://www.MetaboAnalyst.ca/) online software to estimate missing values and to filter and normalize data. Then, the obtained data matrix was introduced into SIMCA-P+14.0 (Umetrics, Umeå, Sweden) software for multivariate statistical analysis of principal component analysis (PCA) and orthogonal partial least-squares discriminant analysis (OPLS-DA). Prior to PCA, all variables obtained from the data matrix were centred on the mean and scaled to the variance. The PCA score plot was used to present a natural correlation between the observations. To identify differential compounds, the OPLS-DA model was used to explore differences in depth between raw TwHF (Raw), TwHF combined with liquorice (Com), and TwHF processed by liquorice (Pro). The OPLS-DA model with VIP values (VIP ≥1.0) and |p(corr)| ≥ 0.5 was selected as a differential compound. Data were analysed using the SPSS software program (version 22.0, Chicago, IL, United States). One-way analysis of variance (ANOVA) used a post-hoc test followed by a test to assess significant differences in the results (the difference was considered statistically significant when p < 0.05 and highly significant when p < 0.01).

Three batches of different TwHF samples were analysed, and multivariate analysis of the data matrix in the chromatogram was performed using the MetaboAnalyst (Figure 1A). shows the clustering results in the form of heat maps, with each row representing the expression level of each gene in different samples, and each column indicating the expression level of all genes in each sample. The upper tree diagram shows the results of cluster analysis for different samples from different experimental groups. The results of the cluster analysis for different genes from different samples are presented on the left tree diagram. There were significant differences in the Raw, Com, QC, and Pro groups. The expression patterns observed in the Com and Pro groups were similar, demonstrating that the sample was reproducible. PCA was used as an unsupervised statistical method to analyse metabolic differences between the Com, Pro, and Raw groups. (Figure 1B) shows a score map of the PCA derived from data obtained by the ESI mode. The QC samples were tightly clustered in the PCA score plot, demonstrating the stability of the LC-MS system throughout the analysis. In addition, in the PCA model, a clear separation trend was observed between the Com, Pro, and Raw groups, indicating considerable metabolic differences between the three groups.

To highlight the differences between groups and facilitate the subsequent search for differential compounds, the data were analysed using OPLS-DA. The projection variable weight value (VIP) reflects the contribution of metabolites to the grouping. Variables in the S-Plot far from the origin with a VIP value > 1 were considered to have a significant contribution to the grouping (Chen et al., 2016; De Preter, 2015; Wang and Wu, 2015). Differential compounds were selected based on VIP >1.0 and |p(corr)| ≥0.5 in OPLS-DA models, followed by significance testing (p < 0.05). These variables were pre-selected as differential compounds. To reduce the risk of false positives in differential compound selection, the variable |p(corr)|select ≥0.5 was the most relevant to the OPLS-DA discriminant score (Xie et al., 2018; Yan and Shen, 2015; Zhang et al., 2015). Figures 2A–D shows the results of the OPLS-DA model derived from ESI + analysis data. Figures 2A,C show the results of the OPLS-DA model using data from the Raw and Pro groups. As shown in (Figure 2A), the score map shows good model adaptability and the Raw group can be clearly separated from the Pro group. Similarly, the OPLS-DA model was constructed based on Com and Pro data (Figure 2B), and the Com group could be clearly separated from the Pro group. Figures 2C,D present the S-plots of the two models. By applying this analysis process, the variables responsible for group separation were selected as differential compound candidates.

Next, the metabolites (p < 0.05) that were significantly different in the Com and Raw groups compared with the Pro group, were selected as candidate compounds. The metabolites in the ESI+ and ESI-mode analyses were combined and molecular formulas further identified. The exact mass-to-charge ratio of all compounds was initially determined by the online METLIN database. (http://www.metlin.scipps.edu/). To determine the underlying structure of the ions, targeted MS/MS analysis was applied to identify metabolites. Metabolites were identified using high-resolution MS and MS/MS fragments and database analysis. To identify metabolites, we used ions of 31.47–451.2848 with a retention time of 31.47 min and a molecular weight of 451.2848 as an example. Based upon analysis of its element composition and fractional isotope abundance, the molecular formula was speculated to be C₂₉H₃₉O₄. In the positive ion spectrum, the main fragment ions screened by MS/MS analysis were observed at m/z 433.2743, 405.2794, 335.2011, 213.0916, 201.0930, and 85.0290, which may refer to lost -OH₂, -CH₂O₂, -C₆H₁₂O₂, -C₁₅H₂₆O₂, -C₁₆H₂₆O₂, and -C₂₅H₃₄O₂. Finally, we used the online METLIN database to define its structure and to temporarily identify metabolites as celastrol. The proposed fragmentation pathway is shown in (Figure 3A) and the mass spectrum is shown in (Figure 3B). Using the above protocol, twelve differential compounds were identified, including four metabolites from the ESI + analysis and eight metabolites from the ESI- analysis, which are listed in (Tables 2, 3).

Following the accurate mass determination and comparison with control retention behaviour and mass spectrometry data, only variables presenting significant changes were selected as differential compounds and their molecular formulas were identified. Mass spectrometry revealed 12 different variables between the Raw and Pro groups and eight between the Com and Pro groups (Table 2). All compounds were initially identified by the on-line TCM Database with an accurate mass charge (20 ppm was set to an acceptable mass error). For each mass ion, several candidates were provided by the above database. The candidate was subjected to further MS/MS experiments and the target molecule was verified by the characteristics of its fragment information. The identity was then determined based on the online database and literature.

The results suggested that changes in the composition of TwHF may be due to the processing method. Therefore, differential compounds following the addition of liquorice were analysed. Some compounds differed, and some were the same between the Pro and Com groups. We conducted a comprehensive analysis and identified 51 compounds in the Pro group and 42 compounds in the Com group; 21 were overlapping between the two groups (Figure 4). This indicates that there has a common role in the Pro and Com groups. In addition, some compounds differ between the two groups. This may explain the difference between the efficacy and reduced toxicity of the Pro and Com groups.

In the search for changes in differential compounds caused by liquorice in the production of TwHF, we found significant (p < 0.05) changes in the three major chemical constituents in TwHF; triptolide, celastrol, and wilforlide A. Figure 5A shows that the triptolide content increased significantly after processing; however, there was no significant difference between the Com and Pro groups. Figure 5B shows that the content of celastrol increased significantly after processing, and there was also a significant difference between the Com and Pro groups. Figure 5C shows that the content of wilforlide A increased significantly after processing. There was also a significant difference between the Com and Pro groups. As shown in the figure, the content of celastrol and wilforlide A decreased and that of triptolide increased after combining with liquorice; however, the content of the three main components was significantly increased after being processed by liquorice.

Before and after processing TwHF, we found that there were significant (p < 0.05) changes in nine major chemical constituents in liquorice. Then, we analysed the effect of liquorice on TwHF and found (Figure 6) that isoliquiritin, glycyrrhizic acid, uralsaponin B, uralenin, isolicoflavonol, and glyyunnanprosapogenin D were significantly increased after being processed by liquorice. We analysed the main compositional changes induced by the processing process.