Chemical Differentiation and Quantitative Analysis of Different Types of Panax Genus Stem-Leaf Based on a UPLC-Q-Exactive Orbitrap/MS Combined with Multivariate Statistical Analysis Approach

Two quantitative methods (−ESI full scan and −ESI PRM MS) were developed to analyze ginsenosides in ginseng stem-leaf by using UPLC-Q-Exactive Orbitrap/MS. By means of −ESI PRM MS method, the contents of eighteen ginsenosides in Asian ginseng stem-leaf (ASGSL) and American ginseng stem-leaf (AMGSL) were analyzed. The principal component analysis (PCA) model was built to discriminate Asian ginseng stem-leaf (ASGSL) from American ginseng stem-leaf (AMGSL) based on −ESI PRM MS data, and six ginsenosides (F11, Rf, R2, F1, Rb1, and Rb3) were obtained as the markers. To further explore the differences between cultivated ginseng stem-leaf and forest ginseng stem-leaf, the partial least squares-discriminant analysis (PLS-DA) model was built based on −ESI full scan data. And twenty-six markers were selected to discriminate cultivated ginseng stem-leaf (CGSL) from forest ginseng stem-leaf (FGSL). This study provides reliable and effective methods to quantify and discriminate among different types of ginseng stem-leaf in the commercial market.


Introduction
Asian ginseng (Panax ginseng C. A. Meyer) and American ginseng (P. quinquefolium L.) are two different medicinal herbs highly valued all over the world [1][2][3][4]. Asian ginseng (ASG) has been used to cure various diseases for thousands of years in Asian countries [5]. Recently, American ginseng (AMG) has also been well known in Asian countries because of its therapeutic functions [2]. ese two related and similar-looking functional herbs, however, show certain differences in their biological activities and pharmacological effects. e root of Panax genus is the most commonly used part of the plant. And the stem-leaf is attracting more and more attention because of its similar pharmacological activities to the root [6,7]. e quality and chemical composition of ginseng vary widely depending on the species, variety, geographical origin, cultivation method, environment, and harvesting time [8]. According to the cultivation environment, ASG could be divided into cultivated ginseng (CG) and forest ginseng (FG). e former is cultivated in artificial conditions and grown for 3-7 years, while the latter is produced by sowing seeds in forest and grown in natural environment for over 10 years. In the commercial market, the specifications and grades of commercial products are numerous, and the prices fluctuate widely with the types of ginseng. Since there is no practical criterion for differentiating different CG, FG, and AMG, it is valuable to find analytical markers and provide methods for discriminating different types of ginseng. Much effort has been devoted to the quantitative and qualitative analysis of ginseng roots which are cultivated in different areas and ages [9]. However, little is known about the difference of stem-leaf in chemical compositions between different types of Panax genus.
Ginsenosides are the major pharmacologically active constituents of Panax genus [10,11]. ey are considered to be responsible for the activities of antioxidant, anti-inflammatory, antiapoptotic, and immunostimulant properties of ginseng [12,13]. More than 70 ginsenosides have been isolated and identified in Panax genus [14]. Few analytical methods, including high-performance liquid chromatography (HPLC) [15], have been developed to exclusively determine ginsenosides in stem-leaf. However, in most of these methods, ginsenosides in minor or trace amounts cannot be detected due to the limited resolution and sensitivity of the detectors. Comparing with HPLC, LC-electrospray-mass spectrometry (ESI-MS) can achieve much higher sensitivity and selectivity and quantify chemical markers in the complex matrixes with only a small amount of sample. Q-Exactive Orbitrap mass spectrometry, with extremely high resolution, sensitivity, and mass accuracy, is a powerful technique for the detection of ginsenosides, especially for those in minor amounts [16,17]. It provides multiple scan modes, including ESI parallel reaction monitoring (PRM) and ESI full scan MS, which are efficient for the quantification of ginsenosides [18]. Compared with the full scan mode, the PRM mode shows better sensitivity and selectivity without establishing the baseline chromatographic separation of target analytes, but the PRM mode only can be carried out by using standards to achieve accurate quantification. Ginsenosides could be detected in both positive and negative ion modes [19]. eir ionization efficiency and limit of detection are different in each mode. erefore, it is necessary to evaluate the quantitative performance using PRM in both positive and negative ion modes.
LC-ESI-MS combined with multivariate statistical analysis has been widely applied for the systematic identification and quantification of all metabolites in a given organism or biological sample in the study of metabolomics [9,20]. e enhanced resolution provided by mass spectrometry, along with powerful chemometric software, allows the simultaneous determination and comparison of   [21,22]. It also has been used to differentiate ginsengs which were cultivated in different areas [23]. In the present study, a UPLC-Q-Exactive Orbitrap/MS method was developed to accurately quantify and identify the ginsenosides extracted from forest ginseng stem-leaf (FGSL), cultivated ginseng stem-leaf (CGSL), and American ginseng stem-leaf (AMGSL). Both −ESI PRM MS and +ESI PRM MS were validated [24], then the −ESI PRM MS was chosen to quantify the ginsenosides in FGSL, CGSL, and AMGSL. e multivariate statistical analysis was further employed to differentiate the Panax genus stem-leaf of various types and also to discover the chemical markers in terms of the detected ginsenosides.

Sample Preparation.
A certain amount of Noto R1, Noto R2, Rb1, Rb2, Rb3, Rc, Rd, Re, Rf, Rg1, Rg2, Rg3, Rh1, Rh2, Ro, F1, F2, and F11 were dissolved in methanol, respectively, to get eighteen standard solutions. ey were mixed and diluted by 70% methanol to get stock solutions and concentrations of each standard in stock solutions are shown in Table 1. en the stock solutions were diluted to be 1.0, 3.3, 10.0, 33.3, 100.0, 333.3, 1000.0, 3333.3, 10,000.0, 33,333.3, 100,000.0, and 333,333.3 fold dilutions by 70% methanol, respectively, for method validation. All the prepared solutions were stored at 4°C. e stem-leaf of Panax genus samples were dried and pulverized to powder and then passed through a 10-mesh sieve. e obtained powder was weighed (0.1 g) and extracted with 5 mL of 70% methanol in an ultrasonic waterbath for 60 min. e extract was filtered through a syringe filter (0.22 μm) and injected directly into the UPLC system.

Method
Validation. An external calibration method was used for the quantitative analysis. And one of the CGSL samples was used in method validation. e linear calibration curves were constructed by plotting the concentrations of six to eleven mixed standard solutions against the corresponding peak areas. e limit of detection (LOD) and limit of quantification (LOQ) were measured with the signal-to-noise ratio of 3 and 10, respectively. e precision was performed by analysis of the standard solution six times and the results were expressed as the relative standard deviation (RSD). For repeatability test, six independent sample solutions were prepared using the procedures described in the last section. e recovery of this method was achieved using the standard addition method. Different concentration levels of the standard solutions were added to the sample six times. e average recoveries were determined by the following formula: (1)
Mass spectrometric detection was carried out on a Q-Exactive Orbitrap/MS ( ermo, San Jose, CA, USA) equipped with an electrospray ionization (ESI) ion source operated in the positive or negative ion mode. e parameters of ion source were set as follows: sheath gas flow of 40 Arb, aux gas flow of 12 Arb, and sweep gas flow of 1 Arb. e S-Lens RF level was 55%. Capillary voltage was set to +3.5 kV or −3.5 kV with a capillary temperature of 333°C and an aux gas heater temperature of 317°C. Full scan MS data were acquired at the centroid mode from m/z 200 to 2000, 70,000 resolution, automatic gain control (AGC) target of 1e6, and maximum injection time (IT) of 100 ms. e mass range of MS 2 spectra varies depending on the precursor ions. e precursor and product ions used for the quantification in the PRM mode are listed in Table 3, together with the corresponding normalized collision energy (NCE). PRM MS data were acquired at the centroid mode, 17,500 resolution, AGC target of 1e5, and maximum IT of 50 ms.

Multivariate Statistical Analysis.
For the −ESI PRM MS, the contents of eighteen ginsenosides in ginseng stem-leaf were used as a dataset containing sample code, ginsenoside, and content. For the −ESI full scan MS, the raw data of samples were processed by the Sieve software (version 2.1, ermo, San Jose, CA, USA), which could detect the mass, retention time, and intensity of the peaks in each chromatogram. e maximum retention time shift was set at 0.25 min and the m/z width was 10 ppm to align the features. e base peak minimum intensity and background were set at 10 5 and 3, respectively. After being aligned, the intensity of each ion was normalized by the total ion intensity of each chromatogram. e resultant dataset, containing m/z value @ retention time, the normalized intensity and the sample code, was used to perform the multivariate statistical analysis. en, both the datasets were saved as .csv files and imported into SIMCA-P software 11.5 (Umetrics, Umea, Sweden) to conduct the multivariate statistical analysis including principal component analysis (PCA) and partial least squaresdiscriminant analysis (PLS-DA). In the PLS-DA model, ions with variable importance in projection (VIP) 1 and VIP 2 values larger than 1 were highlighted and were further filtered by Student's t-test (SPSS19.0, Chicago, IL, USA). e components with p < 0.05 were considered significant and were selected as analytical markers.   Figures  1(a) and 1(b), the total ion chromatograms (TIC) of eighteen ginsenoside standards were obtained in −ESI full scan and +ESI full scan MS. Most of the standards were separated distinctly within 30 min, with an exception of Rf and F11. Rf and F11 were structural isomers and the characteristic components in Asian ginseng and American ginseng, respectively. eoretically, they cannot be present in one ginseng sample. Although they exhibited a poor chromatographic separation in the full scan mode, their quantification could be achieved in PRM mode, in which they were separated by their distinct ion pairs. e ion pairs used for the quantitative analysis of ginsenosides were firstly optimized. Generally, the base peak in the full scan spectra (usually the adduct ion) was taken as the precursor ion, while the base peak in the MS 2 spectra was selected as the product ion. Taking Rd as an example, the [M + HCOO] − ion at m/z 991.55 was the main adduct ion of Rd in the −ESI full scan MS and was chosen as the precursor ion. Its corresponding MS 2 spectrum is shown in Figure 2 respectively. e NCE which ranged from 10% to 50% was further optimized to get the maximum intensity of the product ions. As shown in Figure 2  , respectively, which were obtained by the successive losses of the two saccharide moieties. ese distinct ions could be used as the characteristic product ions for their differentiation and quantification. erefore, the ion at m/z 475.38 and 653.43 were selected as the product ions of Rf and F11, respectively, as their intensities were higher than the counterparts. In addition, the NCE was optimized to be 30% and 25% for Rf and F11 (Figure 3(c)). In the positive ion mode, the fragment ions were also unambiguously identified, as shown in Figures 3(d) and 3(e). All the optimized PRM conditions are shown in Table 3. Table 4, the precisions for all the eighteen ginsenosides from two scan modes were in the range from 1.17% to 4.67%, while the repeatability ranged from 2.18% to 10.05%. e results showed that the developed UPLC-Q-Exactive Orbitrap/MS method had a good precision and repeatability for quantifying ginsenosides in all the tested modes. e LOD and LOQ of eighteen ginsenosides are shown in Table 5. In total, the lower LOD and LOQ were obtained in −ESI PRM MS instead of +ESI PRM MS. In addition, the −ESI PRM MS has the lower LOD and LOQ less than 0.0276 and 0.092 μg/mL for all the ginsenosides.

Validation of the Method. As shown in
As shown in Table 6, all the compounds showed good linearity (r ≥ 0.99) when the method was operated in the negative ion mode. e concentration range could reach up to 3 orders of magnitude, indicating the developed method a good capability to quantify the ginsenosides. Only a few ginsenosides showed good linearity in +ESI PRM MS. It is observed that the linearity of ginsenosides detected in the negative ion mode was better than that in the positive ion mode. e recoveries detected in both the modes were in the range from 65.42% to 137.71% with RSD less than 10.60%, as shown in Table 7.   e validation results indicated that the −ESI PRM MS was better than +ESI PRM MS in LOD, LOQ, linearity, concentration range, and mean recovery. erefore, the described −ESI PRM MS method was subsequently applied to the analysis of all the samples. e developed UPLC-Q-Exactive Orbitrap/MS methods showed lower LOD and LOQ, larger concentration range than HPLC methods [18,25].

Quantification Analysis Based on −ESI PRM MS Data.
e TIC plots of the extracted AMGSL, CGSL, and FGSL are shown in Figures 1(c)-1(e). e differences between AMGSL and Asian Ginseng stem-leaf (ASGSL) could be clearly observed, as Rf and F11 were only present in ASGSL and AMGSL, respectively. e contents of eighteen ginsenosides in every type of samples were detected using the developed PRM/− method, and the quantitative results p [2] (f ) are shown in Table 8  of contents of eighteen ginsenosides. To discriminate among different ginseng stem-leaves and find analytical markers, the multivariate statistical analysis was performed in the next section.

Multivariate Statistical
Analysis. e dataset of the ginsenoside contents was subjected to the multivariate statistical analysis to differentiate the samples from different origins and to discover the chemical markers. Firstly, a PCA mode, an unsupervised pattern recognition technique, was established with R 2 (X) (cum) � 0.850 and Q 2 (cum) � 0.561. As shown in Figure 4(a), the samples are separated distinctly into two groups. CGSL and FGSL were clustered together, suggesting their similar content of ginsenosides. And the clear separation between AMGSL and ASGSL was achieved, indicating the significant difference between these two species. To reveal the ginsenosides that contribute most to the separations of samples from different origins, the loading plot was used to select the chemical markers. As shown in Figure 4(b), the ginsenosides far from the origin were considered to contribute most to the separation and were selected as chemical marker candidates. ese ginsenosides were further filtered by Student's t-test. e variables with contents which are statistically significant different (p < 0.05) between two groups were selected as the analytical markers. Finally, six ginsenosides were obtained as the markers and marked with red boxes in the loading plot. ey were F11, Rf, R2, F1, Rb1, and Rb3 respectively.
To further explore the differences of targeted ginsenosides present in ASGSL samples, the PLS-DA model [26] was established by setting CGSL as group I and FGSL as group II. e PLS-DA scores plot is established with the model parameters of R 2 (Y) � 0.742 and Q 2 (cum) � 0.615 and shown in Figure 4(c). A clear separation is still not observed, since the limited ginsenosides could not provide enough characteristic information for their differentiation. erefore, the PLS-DA model was established based on the data obtained in −ESI full scan MS method was used to separate CGSL and FGSL. As shown in Figure 4(e), clear separations are observed with R 2 (Y) � 0.890 and Q 2 (cum) � 0.824, suggesting good fitness and prediction ability of the established PLS-DA model. Permutation test (n � 200) was further performed to validate the model. No overfitting was found because the permutated R2 and Q2 values on the left are lower than the original point on the right (Figure 4(d)).
ese results indicated that the established PLS-DA mode has the high goodness of fit and predictability. en loading plot (Figure 4(f )) and VIP were used to reveal the potential analytical markers that contribute most to the separation between CGSL and FGSL. After t-test, twenty-six chemical components showed statistical differences between groups and considered to be the chemical markers for the separation of ASGSL and marked with red boxes in the loading plot (Figure 4(f )).
Excepting Rb1, Rg1, and R1, the identification of rest analytical markers was conducted by means of accurate mass-to-charge ratio and MS 2 information obtained by MS. e marker 975.54@20.84 was assigned to Vina-ginsenoside R3 [27] whose MS/MS spectrum is shown in Figure 5 Figures 5(b) and 5(c). e chemical information of analytical markers is shown in Table 9. All the contents of Rg1, Rb1, R1, Vinaginsenoside R3, ginsenoside F3, and ginsenoside F5 in the FGSL were less than in the CGSL. e reason may be related to cultivation.
is method provides an excellent quantitative tool for analysis of ginsenosides in stem-leaf due to its high capacity, high sensitivity, and high selectivity. F11, Rf, R2, F1, Rb1, and Rb3 are analytical markers which could be used to identified ASGSL and AMGLS. Rf was found only in ASGSL, and F11 was found exclusively in AMGLS, which were similar to the difference between Asian ginseng root and American ginseng root [28]. Furthermore, our results suggest that the approach in the present study could be effectively applied to discriminate CGSL from FGSL. Based on UPLC-Q-Exactive Orbitrap/MS combined with multivariate statistical analysis, a reliable and effective approach aimed to discriminate among different types of Panax genus stem-leaf has been successfully developed.

Conflicts of Interest
e authors declare that they have no conflicts of interest.