Discrimination of Radix Polygoni Multiflori from different geographical areas by UPLC-QTOF/MS combined with chemometrics

Acetonitrile (HPLC grade) and formic acid were purchased from Merck KGaA (Darmstadt, Germany); ultra-pure water was purified by a Milli-Q system (Milford, MA, USA).

Seventeen species of planted or wild RPM samples were collected from 10 counties, 4 provinces of China; and one kind of RPM sample (S13) was purchased from pharmacies (Table 1). All the herbal samples were authenticated by the authors. The corresponding voucher specimens were stored in the laboratory for drug metabolism and pharmacokinetics (DMPK) Research of Herbal Medicines, the First Affiliated Hospital of Henan University of Chinese Medicine.

The 18 RPM samples were sliced, dried, and powdered. The powdered samples were screened trough no. 4 sieve, respectively; and 0.25 g was extracted with 25 mL 70% ethanol for 30 min by reflux extraction method, cooled at room temperature and weighted. The reduced weight was complemented by 70% ethanol and mixed well. After standing, the supernatant was filtered through filter paper. During the process, 18 RPM samples were prepared 3 replicates. Before UPLC-QTOF/MS analysis, the filtered supernatant was filtered through a 0.22 μm microporous membrane and 2 μL aliquot was injected.

In addition, quality control (QC) sample was prepared by mixing 100 μL supernatants of 18 geographical RPM samples to validate stability of LC–MS system. It was injected for 3 times before beginning the whole sample list to condition or balance the system. During the analytical run, QC sample was injected every 9 RPM samples to further monitor and investigate the stability and analytical variability of the system. After that, the change degree of the analytical system in the analysis process could be obtained and determined, which was critical for assessing the variation and reliability of the analytical results.

Samples were analyzed using a Waters ACQUITY UPLC I-Class system (Waters Corporation, Milford, USA). An Acquity UPLC HSS T3 C18 column (2.1 mm i.d. × 100 mm, 1.8 μm) was used for chromatographic separation. All samples were run in a random and non-grouped order. The flow phases consisted of 0.1% formic acid in water (A) and acetonitrile (B). The program of gradient elution was set as follows: 0−16 min, 5−60% B; 16−20 min, 60−100% B. The flow rate was 0.4 mL/min. The temperatures of column oven and auto-sampler were maintained at 35 and 10 °C during the analysis, respectively. The sample injection volume was 2 μL.

MS spectrometry detection was operated on a Waters Xevo G2-XS QTOF/MS (Waters, Manchester, UK) equipped with the UPLC system through an electrospray ionization (ESI) interface in negative and positive ion modes. The ESI source parameters were maintained as follows: capillary voltage 1.0 kV, cone voltage 40 V, source temperature 110 °C, desolvation temperature 450 °C. Nitrogen was used as cone gas and desolvation gas with flow of 50 and 800 L/h, respectively. Argon was used as collision gases. The acquisition range of MS scanning was from m/z 50 to 1200 Da in MS^E continuum mode. By using a collision energy ramp from 10 to 30 V, the MS/MS fragment information was obtained. The mass accuracy and reproducibility of UPLC-QTOF/MS was validated by the reference lock mass of leucine-enkephalin (ESI⁺: m/z 556.2771; ESI⁻: m/z 554.2615) with the concentration of 100 pg/μL and the flow rate of 10 μL/min. Data acquisition was performed using Masslynx™ v 4.1 (Waters, Manchester, UK).

All raw data of RPM samples in the LC–MS runs were loaded on Progenesis QI software (v2.0). By using the “assess all runs in the experiment for suitability”, QC2 was automatically selected as the alignment reference. Next, the peaks of all other runs were aligned by comparison with QC2. After that, the experiment design (QC group and S1–S18 groups) was set, the peaks of all samples were picked and convoluted. And then, all data of the peaks were exported into the EZinfo software (v3.0) for PLS‑DA analysis. The necessary data were filtered and then were imported into Progenesis QI software (v2.0) to identify the compounds by its powerful Metascope search engine in the software according to the accurate mass, isotope distribution, fragment ions, collision cross-sectional area and many other parameters.

Multivariate statistical tools was used to observe all differences among the RPM samples from different geographical origins. Firstly, the 3D LC/MS data acquired by Masslynx™ v 4.1 were converted into a 2D ion intensity map as an exact mass retention time (EMRT) pair by using Progenesis QI. During the process, the RPM QC2 sample was automatically selected as the alignment reference by Progenesis QI, and all other RPM samples were aligned with QC2 as the reference. The representative peak alignment results and chromatograms between QC2 and S14b were shown in Fig. 1. Figure 1a was a vector alignment window, there were 414 vectors; Fig. 1b, c both were 2D ion intensity map. Figure 1c also showed the matching results for peak alignment between QC2 and S9b, the score was 96.1%. The matching score range between QC2 and other RPM sample was from 90.3% (S15c) to 97.9% (S18b). The ordinate of Fig. 1a–c represented the retention time (Rt), and the abscissa of them was m/z. Figure 1d was the total ion chromatograms (TIC), green and purple chromatograms represented QC2 and S9b, representatively.

Then, the experiment was designed, 60 samples were divided into 19 groups, including QC and S1–S18 groups. Next, all peaks of RPM samples were picked and convoluted. The parameters of sensitivity value was set at 3 and the minimum peak width was set at 0.15 min, respectively. Under the condition, the best balance could be obtained with the most true feature ion signals and the least random noise. Total 24,530 peaks were observed in the 2D ion intensity map, which was shown in Additional file 2: Figure S1. The normalization graphs for the RPM samples are shown in Additional file 3: Figure S2.

After that, all data were exported into the EZinfo software (v3.0) for PLS-DA analysis. The outliers and classification trends among the 18 kinds of RPM samples could be observed in PLS-DA results (Fig. 2). In the score plot obtained by PLS-DA, there was a clear differentiation between RPM S1–S12 groups and S13 group, indicating that RPM sample from Changhao Chinese Medicine Development Co., Ltd. was very different from other samples in Guizhou province. RPM samples from Guangdong province (S14–S16 groups) clustered together and separated from Guizhou samples. RPM samples from Henan (S17 group) and Sichuan (S18 group) provinces were closer to Guizhou samples, and located father from Guangdong samples. R ² Y and Q ² of the PLS-DA model were 0.771 and 0.634, respectively, which suggested that the PLS-DA model had good adaptability and predictability. Among the RPM samples from the same place of Guizhou province, the samples between S1 and S2 clustered together respectively, indicating that cultivated and wild RPM samples in Xinzhou town had significant difference; the separation between S5 and S6, S8 and S9, S10 and S11 indicated that cultivated RPM samples for 2 and 3 years in Meitan county, Baiduo village, and Niudachang town also had significant difference.

Identification of potential chemical markers in RPM samples from different geographical origins was carried out on basis of the retention behavior and mass assignment using Progenesis QI software. First, PLS-DA model was constructed from the EZinfo software. From loading plots (Fig. 3a) and VIP plots (Fig. 3b) of that model, the interested potential biomarkers could be extracted. Additionally, an ANOVA P ≤ 0.05, a maximum fold change ≥ 2 and VIP value > 1 were set as the restriction conditions to select the significant changing compounds and reduce the “false discovery rate (FDR)”. Next, the Progenesis MetaScope, ChemSpider (http://​www.​chemspider.​com/​) and Element composition methods of Progenesis QI software was used for preliminary determination of the chemical markers. The mass tolerance between the measured m/z values and exact mass of the interest compounds, and the relative mass error of the performed theoretical fragmentation both were set to within 5 ppm. Then, under the targeted MS/MS mode, the MS/MS spectrum of chemical markers was obtained. Finally, some chemical markers were identified by comparison with the standard reference; and others were identified by MS/MS spectrum, online database, element composition results, and literatures.

According to the protocol detailed above, 9 chemical markers (C1–C9) in RPM samples from different geographical origins were identified (Table 2). Among them, 4 chemical markers including C1, C3, C4 and C9 were identified by comparing with their reference compounds. Other 5 compounds were tentatively identified on basis of their molecular ion information and fragments generated by precursor ions. Herein, the C5 with Rt-m/z of 8.53–407.1347 in negative ion mode was detailed as an example to illustrate the identification process. Firstly, the accurate mass of the marker ([M−H]⁻ at m/z 407.1347) was found from the mass spectrum (Fig. 4). Secondly, specific MS/MS information about fragmentation pattern of the marker was acquired from QTOF system. The main fragment ions of the marker in the negative ion spectrum were observed at m/z 245.0819, 230.0948, 202.0635, and 159.0451, which could be the [M−H]⁻ of lost –C₆H₁₂O₅, –C₆H₁₂O₆, –C₈H₁₄O₅, –C₁₀H₁₉O₆, respectively. C₂₀H₂₄O₉ was located as the candidate due to its high mass accuracy among the possible compounds. Finally, the chemical compound was identified as torachrysone-8-O-glucoside (C5) according to the ChemSpider database and literature [20, 21].

The data of 6 replicates of QC sample were analyzed to evaluate the repeatability of LC–MS method. The relative standard deviations (RSD%) of peak areas, Rt and m/z were 5.73–13.42, 0–0.25 and 0.00012–0.00301%, respectively. QC sample maintained in auto sampler at 4 °C for 4, 8, 12, 24, 28, 32 h were tested to assess the post-preparation stability of samples. The relative errors of peak areas were < 13.42% demonstrating good repeatability and stability of the method.

Furthermore, in order to characterize the differences more clearly, the relative intensity of chemical markers in RPM samples from different geographical origins was shown in Table 3 and Fig. 5. The max fold change of C1–C9 were 4.72, 8.12, 2.01, 31.99, 7.78, 4.42, 4.73, 34.91 and 2567.78, respectively. And the results of Fig. 5 C9 and Table 3 showed that the content of C9 in 18 RPM samples had the most different, and the content of C9 in S18 was the highest, which was about 4 time than S13; the content of C9 in S2 and S10 were similar, which was lower than S13 and S18, but higher than other 15 RPM samples; the content of C9 in S1, S3–S9, S11, S12, S14–S17 were similar. The content trend of C4 (Fig. 5 C4) in 18 RPM samples was similar with C9. The content of C8 (Fig. 5 C8) in S9 was the highest; the content of S1, S3, S13, and S16 were lower than S9, but higher than S2, S4–S8, S10–S12, S14, S15, S17 and S18. The content of S4 and S7 was similar, which was higher than other 16 RPM samples. The content of C5 (Fig. 5 C5) and C7 (Fig. 5 C7) in S5 was the highest, the content of them in S9 was the lowest, but the content of C7 in S9 was similar with that in S9. The content of C1 (Fig. 5 C1) and C3 (Fig. 5 C3) in S9 and S14 were the highest, respectively; but the content of them in S13 both were the lowest. The content of C2 (Fig. 5 C2) and C6 (Fig. 5 C6) in S4 were the highest, but the content of C2 in S7 was similar with S4; the content of them in S16 both were the lowest, but the content of C2 (Fig. 5 C2) in S13 and S15 was almost equal to that of S16; and the content of C6 (Fig. 5 C6) in S14 was almost equal to the content in S16.