Comparing the antidiabetic effects and chemical profiles of raw and fermented Chinese Ge-Gen-Qin-Lian decoction by integrating untargeted metabolomics and targeted analysis

The information regarding the experimental design, statistics, and resources used in this study is attached in the minimum standards of reporting checklist (Additional file 1).

Acetonitrile (HPLC and MS grade) and methanol (HPLC grade) were purchased from Tedia (Fairfield, USA) and Hanbon (Nanjing, China), respectively. Formic acid (analytical grade) was provided by Aladdin Chemistry Co. Ltd (Shanghai, China). De-ionized water was prepared in-house by a Milli-Q water purification system (Millipore, MA, USA). Other chemicals and reagents were analytical grade. The chemical reference substances (purity > 98%, HPLC–DAD) of 3′-hydroxypuerarin, puerarin, daidzin, daidzein, baicalin, wogonoside, baicalein, wogonin, coptisine, berberine, palmatine, magnoflorine, genistin, genistein, ononin and formononetin were purchased from Chengdu Wei ke-qi Bio-Technology Co., Ltd. (Chengdu, China). Liquiritin, isoliquiritin, liquiritigenin, isoliquiritigenin and glycyrrhizic acid were purchased from Chunqiu Bio-Technology Co., Ltd. (Nanjing, China). Scutellarein (purity > 98%, HPLC–DAD) was isolated, purified and identified in our lab.

Puerariae Lobatae Radix (Gegen), Scutellariae Radix (Huangqin), Coptidis Rhizoma (Huanglian) and Glycyrrhizae Radix et Rhizoma Praepapata Cum Melle (Zhigancao) were purchased from Wan Min pharmacy (Taiyuan, China) and authenticated by Associate Professor Chenhui Du, according to the standard of the Chinese Pharmacopeia (2015 edition). Voucher specimens were deposited in the Modern Research Center for Traditional Chinese Medicine of Shanxi University. SC (CICC 1205) was purchased from the China Center of Industrial Culture Collection (CICC).

Herb pieces of 3200 g (Gegen:Huangqin:Huanglian:Gancao = 8:3:3:2) were immersed in a 10-fold volume of distilled water (w/v) for 0.5 h and then extracted by refluxing two times (40 min, 30 min). For each extract, the decoction was filtered through eight layers of gauze to remove the herbal residue. The two filtrates were combined, condensed under reduced pressure with a rotary evaporator at 70 °C and evaporated to dryness (yield: 28.6%).

Freeze-dried spores of SC were recovered in 25 mL of potato dextrose (PD) medium and then incubated at 28 °C on a rotary shaker at 180×g for 24 h. A 20-mL volume of GQD (0.5 g mL⁻¹, crude drug per g mL⁻¹) was mixed with 30 mL of distilled water in a 250-mL flask. The substrates of GQD were subjected to autoclaving at 121 °C for 20 min, then shook evenly and allowed to cool naturally. The sterilized substrates of GQD were inoculated with 2% (v/v) recovered SC and incubated at 28 °C in a shaking incubator (180×g). GQD samples were fermented for 48 h and then evaporated to dryness.

The concentrations of GQD and FGQD were approximately 2 g mL⁻¹ (crude drug per g mL⁻¹) for the animal experiments. In addition, the GQD and FGQD extracts for LC and LC–MS analysis were also prepared using the same protocol mentioned above in triplicate.

Male Sprague–Dawley rats (200–220 g) were purchased from Beijing Vital River Laboratories Co., Ltd. (SCXK (Jing) 2014-0013, Beijing, China). The rats were housed at a controlled room temperature of 23 ± 2 °C, 55 ± 10% humidity and a 12-h dark–light cycle for 10 days with free access to food and water. Then, 70 rats were randomly divided into two groups: the normal control group (NC, n = 10) and the diabetic rats group (n = 60). The NC group was fed a regular diet. The diabetic rats group was fed a high-sugar and HFD containing 5% sucrose, 10% lard, 5% yolk powder, 1% cholesterol, 0.1% sodium cholate and 78.9% regular diet. After 4 weeks of dietary intervention, the diabetic rats were fasted for 24 h and then received STZ (35 mg kg⁻¹) dissolved in citrate buffer (0.1 M, pH 4.5) by intraperitoneal injection. The rats in the NC group received an equivalent volume of citrate buffer vehicle. One week after injection, fasting blood glucose (FBG) levels were determined using a drop of blood from the tail vein. Rats with FBG level above 11.1 mM were randomly subdivided into four groups (n = 13 for each group): the diabetic model group (DM) and three treatment groups. The treatment groups were fed 0.67 mg kg⁻¹ of metformin hydrochloride (HM), 20 g kg⁻¹ of GQD, or 20 g kg⁻¹ of FGQD (crude drug per g kg⁻¹ of body weight) every day for 8 weeks. Body weights were recorded every week, and FBG levels were measured every 2 weeks throughout the experiment.

At the end of the experimental period, the rats were sacrificed under anaesthesia, and blood was immediately collected. Total serum cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C) and low-density lipoprotein cholesterol (LDL-C) levels were measured by an ELISA kit (Nanjing jiancheng Bioengineering Institute, Nanjing, China). The fast serum insulin (FINS) concentration was measured using commercial kits (Wa Lan Biotechnology, Shanghai, China). The insulin sensitivity index (ISI) was calculated according to FBG and FINS. The following formula for ISI was used: Ln (1/FBG * FINS) [24]. Homeostasis model assessment-insulin resistance (HOMA-IR) was calculated to measure the insulin sensitivity of the rats fed the experimental diets using the following formula: [FINS × FBG] 22.5⁻¹ [25].

For HPLC quantification, a mixed stock solution of ten reference substances was prepared at concentrations ranging from 1.0 to 2.5 mg mL⁻¹ in 70% methanol. A standard working solution of the mixtures was obtained by diluting the stock solutions to the desired concentrations. All solutions were stored at 4 °C before use.

An HPLC Ultimate™ 3000 instrument coupled with a Q Exactive MS (Thermo Scientific, Bremen, Germany) was used for untargeted metabolomics in this study. Chromatographic separation was performed on an Agilent Poroshell 120 EC-C₁₈ column (3 × 100 mm, 2.7 µm, Agilent, CA, USA). The mobile phase consisted of water containing 0.1% (v/v) formic acid (A) and acetonitrile (B). The following gradient was used: 0–10 min, 5% B to 17% B; 10–12 min, 17% B; 12–14 min, 17% B to 22% B; 14–19 min, 22% B; 19–29 min, 22% B to 32% B; 29–30 min, 32% B to 50% B; 30–34 min, 50% B to 90% B. The column was equilibrated for 5 min prior to each analysis. The flow rate was 0.3 mL min⁻¹, and the column temperature was maintained at 30 °C. The mass spectrometer was operated in both positive and negative ESI full MS–dd-MS/MS acquisition mode with the use of the following parameter settings: spray voltage, 3.5 kV; sheath gas: 35 arbitrary units; auxiliary gas: 10 arbitrary units; capillary temperature: 320 °C; S lens RF level: 55; heater temperature: 300 °C. Full scan data were recorded for ions with m/z 100–1500 at a resolution of 70,000 (FWHM defined at m/z 200) in profile format. The automatic gain control (AGC) target values were set at 1 × e⁶ and 3 × e⁶ ions, respectively. The injection time was set to 250 ms in ESI⁻ mode and 100 ms in ESI⁺ mode. The MS/MS event was triggered when the given precursor ion was detected in an isolation window of m/z 2.0. The stepped normalized collision energies (NCE) of the analytes were 10, 30 and 50.

Targeted metabolite quantification was performed on a Waters ACQUITY UPLC H-Class system (Milford, MA, USA). Samples were separated on an Agela-MP C₁₈ column (2.1 mm × 250 mm, 5 μm, Agela, Tianjin, China) maintained at 30 °C. The binary mobile phase consisted of water containing 0.1% formic acid (A) and acetonitrile (B) at a flow rate of 1.0 mL min⁻¹. The optimized gradient elution program was set as follows: 5–20% B (0–25 min), 20% B (25–30 min), 20–22% B (30–35 min), 22–40% B (35–55 min), 40–63% B (55–65 min), 63–80% B (65–70 min). The UV signals from two separate channels of 254 nm and 276 nm were recorded.

Data from the HPLC Q Exactive MS acquisition and processing were used for chemical profile analysis using Xcalibur™ 2.2 (Thermo Fisher). The untargeted metabolomics analysis was conducted by using Compound Discovery (version 1.2.1, Thermo SCIEX), and the detailed workflow is shown in Additional file 2: Figure S1. The multivariate data matrix was introduced into SIMCA-P (Version 13.0, Umetrics AB, Umea, Sweden) software for “unsupervised” principal component analysis (PCA) and “supervised” orthogonal projection to latent structure-discriminant analysis (OPLS-DA). All variables were UV-scaled for PCA and Pareto-scaled for OPLS-DA.

As shown in Fig. 1, the body weight of the diabetic rats decreased significantly compared with the NC group after STZ injection (p < 0.01). HM reversed the diabetes-induced body weight decrease from the 6th week (p < 0.05), whereas FGQD significantly reversed the body weight decrease from the 7th and 8th weeks (p < 0.01, p < 0.05). However, no significant (p > 0.05) effect was observed for the GQD group, suggesting that GQD had no significant effect on weight gain. As shown in Additional file 2: Figure S2, the FBG level was significantly increased in the diabetic rats compared to the NC group (p < 0.01) and was decreased in all drug-treated groups from the 4th week (p < 0.01, p < 0.05) after the injection of STZ. Although no significant difference was observed among the drug-treated groups (p > 0.05), the diabetic rats in FGQD showed a better trend of recovery. Rats in the model group had significantly higher levels of TC and TG (p < 0.01) than those in the NC group, and these levels were reduced in all drug-treatment groups (p < 0.01) (Fig. 2). Notably, the levels of TC and TG were significantly lower in the FGQD group than in the GQD group (p < 0.01) (Fig. 2). In addition, the treatments with HM and FGQD reversed the up-regulation of LDL and down-regulation of HDL in the diabetics rats group to the control level, whereas no significant (p > 0.05) effect was observed for GQD (Fig. 2). As shown in Table 1, the diabetic rats showed significant increases in FINS and HOMA-IR (p < 0.01) and a decrease in ISI (p < 0.01) compared with the NC group. After 8 weeks of drug administration, the levels of FINS, ISI and HOMA-IR were reversed compared with the DM group (p < 0.01). In addition, a notable difference in FINS level was observed in the FGQD group (p < 0.01) compared with the GQD group. In short, the body weight gain and the regulation of the levels of FINS, TC, TG, LDL and HDL in the FGQD group were significantly better than those in the GQD group (p < 0.01), but there were no significant differences in FBG, ISI and HOMA-IR levels between GQD and FGQD. These results suggested that FGQD had better therapeutic effect against diabetes than GQD.

Since herbal medicines are generally taken as a decoction, we focused on boiled water extracts of GQD and their fermentation. The structural characterization of compounds in GQD is an essential step in identifying and matching those compounds with their secondary metabolites obtained through biotransformation. All known compounds were identified by comparison with chemical standards. For unknown compounds, structures were tentatively characterized based on retention time and MS spectra by referring to the previous literature. Finally, assignments of all compounds were further conducted by comparing the corresponding extracted ion chromatography (EIC) of GQD with those of the individual herbs. In total, 133 compounds were rapidly identified or tentatively characterized; these compounds were divided into six structural types. The detailed information, including retention times, accurate m/z, ppm errors, characteristic fragment ions, identified names and formulas, are summarized in Table 2, Additional file 2: Figure S3. Notably, two compounds were identified for the first time in GQD: 6-d-xylose-genistin and kuzubutenolide A.

To identify chemical markers distinguishing GQD and FGQD samples, the negative and positive ion mode data detected by HPLC Q Exactive MS were simultaneously used for global analysis. Visual inspection of the chromatograms for GQD and FGQD indicated that the fermentation process induced obviously different peak intensities; that is, FGQD contained more daidzein, liquiritigenin, genistein, and biochanin A and less daidzin and liquiritin than GQD (Fig. 3). Multivariate statistical analysis was subsequently applied to further reveal the minor differences between GQD and FGQD. In the PCA score plot (Additional file 2: Figure S5A, B) generated by PC1 (46.2%) and PC2 (17.9%) for positive ion mode and PC1 (51.1%) and PC2 (17.9%) in negative ion mode, clear separation can be observed between GQD and FGQD. Then, OPLS-DA was further performed to process the secondary metabolome data between the GQD and FGQD groups by S-plot and VIP-value analysis. The model fit parameters were 0.999 for R²Y (cum) and 0.971 for Q² (cum) for positive ion mode and 0.999 for R²Y (cum) and 0.987 for Q² (cum) for negative ion mode, respectively, suggesting that the OPLS-DA model exhibited good fitness and predictability. In the S-plots, each point represented an ion t_R-m/z pair, whereas the distances of the pair points from the mean centre indicate the contribution of the variables in discriminating the GQD and FGQD groups (Fig. 4a, b). The VIP-value threshold cut-off of the variables was set to one, and thus 83 and 117 variables were finally screened in LC/MS (ESI⁺) and LC/MS (ESI⁻), respectively. Among them, 25 variables were identified in both ion modes. Three variables and two variables were identified in negative ion mode and positive ion mode, respectively. Thus, 30 compounds that had different intensities between GQD and FGQD were detected.

To maximize the understanding of the effect of fermentation on GQD, the mean peak areas and the t-test results for the significant differences in the 30 compounds from GQD and FGQD are shown in Figs. 5, 6. As shown in Fig. 5a1, the mean peak areas of free flavones (P35, P37, P40 and G12) were larger in FGQD than in GQD (p < 0.001), whereas the mean peak areas of their corresponding O-glycosides (P5, P18, P20, P26, G2 and G3) were smaller in FGQD than in GQD (p < 0.001, p < 0.05), indicating that O-glycoside hydrolysis occurred during fermentation processing (Fig. 5a2). P23 could also be transformed to P35 by O-glycoside hydrolysis. In addition, P10 and P34 contained abundant hydroxyl and methyl and were deduced to possibly produce P18 by dehydroxylation or demethylation. Actually, a marked decline in the level of P34 was also observed (p < 0.01) (Fig. 5a1), however, its corresponding aglycone P41 was not obviously altered in FGQD, which might be due to a dynamic equilibrium between their formation (from O-glycoside hydrolysis) and further transformation (e.g., demethylation). By contrast, C-glucosides appeared to be more difficult to transform by SC, since five C-glucosides (P6, P11, P13, P14 and P24) were detected in FGQD (Fig. 5b1). Their significant increasing trend was probably caused by the hydrolysis of low contents of puerarin C-glucoside-O-glucoside derivatives, such as P1, P2, P3, P4, P8, P12 and P15 (Fig. 5b2). O-C glycoside bonds have been reported to be the main effective target of β-glucosidase [13], in agreement with our results that puerarin (P11) and its derivatives were difficult to hydrolyse by β-glucosidase.

As shown in Fig. 6a1, the remarkable increase in the level of flavone aglycone (S43) was potentially due to hydrolysis of the corresponding flavone O-glucuronide (S28), which contains a 6-OCH₃ group (p < 0.001). S31, which contains an 8-OCH₃ group, was more difficult to transform by hydrolysis by SC but was easier to produce from S25 by dehydroxylation (Fig. 6a2). Although a different strain of yeast was used, the current findings are still in agreement with those in a previous study [39]. Notably, the increasing trend of S37 is likely partially responsible for the hydrolysis reactions of the corresponding compound (S19) (Fig. 6a2). A previous study demonstrated that Escherichia (E.) coli β-glucuronidases could hydrolyse glucuronic acid at the 7-position if the structure contains a 6-OH group [39]. Other metabolic reactions for flavone-O-glucuronides, including demethylation and dehydroxylation, were also deduced.

Due to the lack of a free hydroxyl group, alkaloids are demethylated to form free hydroxyl groups by SC [36]. In this study, a significant increase in demethyleneberberine (C9) was observed in FGQD compared to GQD (p < 0.05), which probably contributed to the demethylation of C19 during fermentation processing (Fig. 6b1, b2). There were no significant differences in the other benzylisoquinoline alkaloids between GQD and FGQD (p > 0.05), thus indicating that the contents of these molecules remained stable during the fermentation process.

As mentioned above, the untargeted metabolomic studies indicated that isoflavone O-glycosides, flavone O-glycosides, flavone O-glucuronides and alkaloids were potential chemical markers for distinguishing GQD and FGQD. Thus, three O-glycosides (daidzin, baicalin and liquiritin), one C-glycoside (puerarin), three flavones (daidzein, liquiritigenin, and baicalein), and three alkaloids (coptisine, berberine and palmatine) were quantitatively determined as examples to illustrate the effects of processing (Additional file 2: Figure S3, Table S1). Their content changes in GQD and FGQD are summarized in Table 3. As expected, fermentation processing significantly depleted liquiritin (O-glycoside) from 0.80 ± 0.06 mg g⁻¹ to 0.48 ± 0.02 mg g⁻¹ (p < 0.05), whereas daidzin was not even detectable in FGQD (p < 0.001) after fermentation with SC. Interestingly, the concentrations of daidzein and liquiritigenin (free flavones) in FGQD were greatly enhanced (p < 0.001, p < 0.05, respectively). In addition, an obvious increase in the level of puerarin (isoflavone C-glycoside) was observed until the end of fermentation. Regarding alkaloids, the contents of coptisine, palmatine and berberine remained relatively stable (p > 0.05). Moreover, there was a slight increasing trend for baicalin (flavone O-glucuronide), whereas no significant difference was found between GQD and FGQD. Interestingly, the quantitative results revealed an increasing trend for baicalein (p > 0.05) did not correspond to the results of the untargeted studies, which showed a significant increase in the content of baicalein in FGQD compared with GQD (p < 0.05).