Chemical Profiling and Comparison of Sangju Ganmao Tablet and Its Component Herbs Using Two-Dimensional Liquid Chromatography to Explore Compatibility Mechanism of Herbs

Sangju Ganmao tablet (SGT), a well-known Chinese patent medicine used to treat cold symptoms, is made from eight herbal medicines. In this study, an off-line hydrophilic interaction × reversed-phase two-dimensional liquid chromatography (HILIC × RP 2D-LC) method was developed to comprehensively separate the chemical constituents of SGT. Through optimization of the experimental conditions, a total of 465 peaks were finally detected in SGT, and the structures of 54 selected compounds were fully identified or tentatively characterized by quadrupole time-of-flight mass spectrometry (qTOF-MS) analysis. The established 2D-LC analysis showed high orthogonality (63.62%) and approximate 11-fold improvement in peak capacity (2399 and 1099, obtained by two calculation methods), in contrast to conventional one-dimensional RPLC separation. The eight component herbs of SGT were also respectively separated by using the 2D-LC system, and we found that a total of 12 peaks detected in SGT were not discovered in any component herbs. These newly generated chemical constituents would benefit better understanding of the compatibility mechanism of the component herbs. The strategy established in this study could be used for systematic chemical comparison of SGT and its component herbs, which contributes to exploration of herbal compatibility mechanism.


INTRODUCTION
Most of the Chinese patent medicines are composed of several or even 10s of herbal medicines, and they are very complicated chemical systems containing both hydrophilic and hydrophobic compounds (Li et al., 2010). While profiling the chemical constituents is critically important for modern investigations of a Chinese patent medicine, chromatographic separation of such a complex system is still a big challenge. On the other hand, the chemical profiling of a Chinese patent medicine is usually different from its component herbs, since chemical reactions and physical changes, such as oxidation and precipitation, sometimes occur during decocting together (Kim et al., 2014). Investigating the different chemical constituents between a Chinese patent medicine and its component herbs may be one of effective approaches to explore its compatibility mechanism.
Sangju Ganmao tablet (SGT), a well-known Chinese patent medicine, is currently used in clinical practice to treat cold symptoms, and has been officially recorded in China Pharmacopoeia (Chinese Pharmacopoeia Commission, 2015). It is composed of eight herbs, namely mulberry leaf ( ). The chemical constituents of SGT mainly include flavonoids (free flavonoids and flavonoid glycosides), triterpenoid saponins, phenylethanoid glycosides, and organic acids (Chang et al., 2014). Although several researches have studied the chemical profiling of these eight component herbs using various methods (Huang and Sheu, 2007;He et al., 2013;Chen et al., 2016;Song et al., 2017), systematic chemical analysis of SGT has never been reported so far. Recently, we developed a LC-MS method to profile the chemical constituents of SGT by optimizing different HPLC systems, and only less than 50 compounds were detected (Guo et al., 2017). Comprehensive profiling of chemical constituents in SGT is hindered due to its complex chemical composition, and the significantly different content renders those minor components difficult to be separated and detected. The chemical differences of SGT and its component herbs are also unknown.
In the past decade, two-dimensional liquid chromatography (2D-LC) has been proven to be a powerful tool in rapid separation and detection of chemical constituents in traditional Chinese medicines (TCM) (Shellie and Haddad, 2006;Guiochon et al., 2008;Marchetti et al., 2008). Among all the separation modes, combination of reversed-phase liquid chromatography (RPLC) and hydrophilic interaction chromatography (HILIC) is an effective method to separate complex mixtures with a wide range of polarities (Wang et al., 2008;Jandera and Hájek, 2018). A 2D-LC system can be operated in the on-line or offline mode. Comparing with an on-line 2D-LC system with complex equipment settings, an off-line 2D-LC system can easily accomplish the flexible integration of different separation mechanisms without instrumental limitation . Very recently, an off-line HILIC × RP 2D-LC method was established in our lab to achieve the comprehensive profiling of the chemical constituents of Ginkgo biloba extract .
In this study, a method based on off-line HILIC × RP 2D-LC coupled with qTOF-MS was established to comprehensively profile the chemical constituents of SGT, and the 2D-LC system was systematically optimized to achieve ideal orthogonality and peak capacity. The established method was used to investigate the chemical differences between SGT and its component herbs in order to explore the herbal compatibility mechanism.

Sample Preparation
The simulative solution of SGT was prepared according to its record in China Pharmacopoeia (Chinese Pharmacopoeia Commission, 2015), and the detailed experimental process is described in Supporting information. A total of 23 reference standards were weighed accurately and dissolved in methanol to obtain their individual standard solutions at the concentration of 1 mg/mL, except for 26 (0.25 mg/mL). To optimize chromatographic condition of the 2D-LC system, the solutions of compounds 7, 15, 19, 20, 23, 25, 26, 28, 33, 37, 38, 39, 40, 42, 44, and 51 were mixed, dried under a gentle nitrogen flow, and reconstituted in methanol to produce mixed standard solutions. All the solutions were stored at 4 • C until use.
The D1 LC (HILIC) was conducted on a Waters e2695 HPLC system, and the samples were separated on a Waters XBridge Amide column (2.1 mm × 150 mm, 3.5 µm). The mobile phase consisted of 0.1% formic acid (A) and acetonitrile (B), and a gradient elution program was used: 0 min, 90% B; 10 min, 60% B. The flow rate was 0.4 mL/min, and the column temperature was 35 • C. The detection wavelength was 254 nm. From 1 to 10 min, a total of 10 fractions (Fraction 1 to Fraction 10) were collected with a one-min interval. The sample injection was repeated for five times, and the same fractions were combined, respectively. The resulting fractions (except for Fraction 1) was concentrated to dryness under a gentle flow of nitrogen gas at 37 • C, and were then re-dissolved in 100 µL of 50% methanol for the next D2 LC analysis.
The D2 LC (RPLC) was performed on an Agilent 1290 UHPLC system. The samples were separated on an Agilent Eclipse Plus C18 column (4.6 mm × 100 mm, 3.5 µm), and a binary mobile phase composed of 0.1% formic acid (A) and methanol (B) was used following a gradient elution program: 0 min, 5% B; 30 min, 95% B. The flow rate was 1.0 mL/min, and the column temperature was 35 • C. The detection wavelength was 254 nm.

Mass Spectrometry
An Agilent 6550 qTOF mass spectrometer equipped with an electrospray interface (ESI) was coupled to the D2 LC system, and the ion source was operated in the negative ion mode. The D2 LC eluent from the UV detector was introduced into the mass spectrometer by using a T-splitter, and the post-column splitting ratio was 8:1 (approximate 0.11 mL/min into the mass spectrometer). The MS parameters were set as follows: ionization mode, Dual AJS ESI (-); reference ion, 112.98 and 1033.98; capillary voltage, 3500 V; drying temperature, 280 • C; dry gas flow, 12 L/min; atomizer pressure, 28 psig; sheath temperature, 350 • C; sheath gas flow, 12 L/min; scan range, m/z 50-1500; cataclastic voltage, 300 V; collision energy, 15 and 45 eV.

Data Analysis
The raw DAD data were exported into OriginPro 9.1 software (OriginLab Corporation, United States) to construct the contour plots. Orthogonality (O) and peak capacity (n 2D ) were calculated based on literatures (Gilar et al., 2005;Rutan et al., 2012), as described in our previous publication .

Optimization of the Off-Line 2D-LC System
The separation modes of D1 and D2 in a 2D-LC system are determined based on the types of interactions between samples and stationary phases. The separation capacity of RPLC is mainly associated with non-polar selectivity, while HILIC is dominated by polar interaction. Therefore, the two different separation modes were combined to enhance orthogonality and peak capacity in this study, since the chemical constituents of SGT have a wide range of polarities. RPLC has been selected as the second dimension since it usually shows better peak shape of analytes and better compatibility with mass spectrometry, according to literatures (Jandera and Hájek, 2018). Among the 23 reference standards, 16 representative compounds (7, 15, 19, 20, 23, 25, 26, 28, 33, 37, 38, 39, 40, 42, 44, and 51), which are the major constituents of SGT with different structural types, were chosen and mixed to prepare standards solution (Figure 1). They were used to optimize the 2D-LC conditions.

Optimization of HILIC
Two types of HILIC columns (Agilent Zorbax HILIC plus and Waters XBridge Amide) were evaluated by separating the above standard solution, with the gradient program described in Section "Off-Line Two-Dimensional Liquid Chromatography." As shown in Figure 2, all the standards were co-eluted within 2 min with poor separation when they were separated on the Agilent Zorbax HILIC plus column. On the contrary, the Waters XBridge Amide column exhibited stronger retention capacity to the standards, and their identities were confirmed by mass spectrometry (Figure 2). A total of 14 peaks can be observed, and only four compounds were not fully separated. Therefore, we chose Amide as the stationary phase for the first dimension HILIC in this study, and gradient acetonitrile and 0.1% formic acid were chosen as the mobile phases.

Optimization of RPLC
Four RPLC columns from Agilent Technologies with different packing materials or inner diameters were tested in respect of selectivity and resolution by using the gradient program described in Section "Off-Line Two-dimensional Liquid Chromatography." The 16 reference standards were difficult to be fully separated in one single run, as shown in Figure 3. Apparently, the Eclipse plus C18 (4.6 mm, i.d.) column showed the highest resolution and best peak shape, and only six analytes were not fully separated. In particular, compounds 7, 26, and 39 could be well-resolved, though they were almost overlapped when separated on Zorbax SB-C8, Zorbax SB-C18 and Eclipse plus C18 (2.1 mm, i.d.). Therefore, Agilent Eclipse plus C18 (4.6 mm, i.d.) was tentatively chosen as the stationary phase for the second dimension RPLC. The mobile phases were then optimized, and detailed experimental process was described in Supporting information. As a result, 0.1% formic acid was used as the aqueous phase, and methanol was used as the organic phase.
Subsequently, the orthogonality between Waters XBridge Amide and Agilent Eclipse plus C18 (HILIC × RPLC) was evaluated with Pearson linearity regression correlation coefficient (r) as the index. In addition, the orthogonality of several other combination modes, namely XBridge Amide × Zorbax HILIC plus, Zorbax SB C18 × Eclipse plus C18 (4.6 mm, i.d.), Zorbax SB C8 × Eclipse plus C18 (4.6 mm, i.d.) and XBridge Amide × Zorbax SB C18, was also determined. Retention times of each reference standard separated on these columns were firstly transformed into normalized retention times RT max −RT min , calculated according to our previous publication ], and the r-values of the two groups of RT (i)norm values were then calculated, respectively. As shown in Figure 4, the r-values of Amide × HILIC plus, SB   due to the similar separation mechanisms between D1 and D2. As expected, the combination mode of HILIC × RPLC (Amide × SB C18 and Amide × plus C18) exhibited a higher orthogonality with r < 0.30. Then, the orthogonality (O = bins− √ P max 0.63P max ) and practical peak capacity (n 2D = n 1 × n 2 × 1 β × f ) of different combination modes were calculated according to our previous publication . As a result (Supplementary Table S1), Amide × plus C18 gave a significantly higher practical peak capacity (2303 and 1231) than Amide × SB C18 (1691 and 850), calculated by two different methods, though their orthogonality was the same (0.3969). Based on the above data, the choice of Waters XBridge Amide × Agilent Eclipse plus C18 for the chemical analysis of SGT was further confirmed in this study.

Evaluation of the Off-Line 2D-LC System
The optimized off-line 2D-LC system was used to analyze SGT, which mainly contains flavonoids, triterpenoid saponins, phenylethanoid glycosides, and organic acids. SGT was firstly separated on the XBridge Amide column in D1, and a total of 10 fractions (Fraction 1 to Fraction 10) were consecutively collected with 0.4 mL (1 min) per fraction (Supplementary Figure S1). The resulting fractions (except for Fraction 1) were further separated on the Zorbax Eclipse plus C18 column in D2, respectively, which was successively coupled with DAD and qTOF-MS as the detectors (Figure 5A). All the peaks in the nine chromatograms (Fraction 2 to Fraction 10) were assigned based on their retention time, UV spectra and MS data, and a total of 465 peaks were finally recognized in SGT. To evaluate the orthogonality and peak capacity of the 2D-LC system, the normalized retention times of all the peaks were calculated, and the separation space was divided into 22 × 22 rectangular bins, which (484) is close to P max (465). The rectangular bins were then superimposed with the data points, as shown in Figure 5B, and orthogonality and peak capacity were calculated according to literatures (Gilar et al., 2005;Rutan et al., 2012). Bins containing data plots ( bins) covered about 50% of the separation space, and the orthogonality was calculated as 63.62%. In addition, the Pearson correlation coefficient (r) of the two groups of normalized retention time values was 0.0525, which suggested a good orthogonality.
The average peak width of D1 and D2 was 0.738 and 0.145 min, respectively, and the effective gradient time for D1 and D2 was 10 and 30 min, respectively. Thus, the theoretical peak capacities for D1 and D2 were 14 and 207, respectively. The theoretical peak capacity of the 2D-LC system was 2898, and the practical peak capacity was 2399 and 1099, respectively, calculated following two methods described in our previous publication . The 2D-LC system increased the peak capacity by almost 11-fold comparing with the singledimension D2. Improvement on peak capacity is beneficial to the exposure of more minor components, which could contribute to comprehensive separation and detection of chemical constituents in SGT and its component herbs.

Separation and Characterization of Chemical Constituents in SGT and Its Component Herbs by the Off-Line 2D-LC/qTOF-MS System
Chemical constituents of SGT were globally analyzed by the optimized off-line 2D-LC/qTOF-MS system, and a total of 465 peaks were detected. The eight component herbs were also respectively separated and detected by using the 2D-LC system, and we found that 12 peaks detected in SGT were not  Supplementary  Table S2. The 12 compounds might be produced through chemical reactions of chemical constituents in component herbs during decocting together, such as hydrolysis, oxidation, and dissolution. They might cause the pharmacological activity differences between SGT and its eight component herbs, and contribute to exploration of the herbal compatibility mechanism. Further separation and purification are necessary to fully identify the chemical structures of the 12 compounds through nuclear magnetic resonance (NMR) spectroscopic analysis in future study.

discovered in any component herbs. Their retention times and high-resolution mass spectral data were listed in
Among the 465 peaks, a selected group of constituents (54 compounds) were identified by comparing with reference standards (23 compounds), or tentatively characterized by comparing their high-resolution mass spectral data with previous literatures (31 compounds). These compounds included free flavonoids, flavonoid glycosides, triterpene saponins, phenylethanoid glycosides, organic acids, and others ( Table 1). The qTOF-MS spectra of four representative compounds are illustrated in Figure 6.

Characterization of Flavonoids
Flavonoids (free flavonoids and flavonoid glycosides) in SGT are mainly derived from mulberry leaf, chrysanthemum, Fructus Forsythiae and licorice. In the negative ion mode, free flavonoids in SGT could fragment on ring-C following the retro Diels-Alder (RDA) reaction at a relatively high collision energy, and they could also lose small molecules (like CO) or radicals (like CH 3 ·) to produce diagnostic fragments (Fabre et al., 2001).  (Cuyckens et al., 2001). In contrast, flavonoid C-glycosides could fragment on the sugar ring to lose 60, 90, or 120 Da ( 0,4 X, 0,3 X, or 0,2 X cleavage, respectively) at a relatively high collision energy (Lin et al., 2005). Based on these fragmentation behaviors, a selected group of flavonoids (7 free flavonoids and 16 flavonoid glycosides), including isoflavones, flavones, flavanones and chalcones, were identified or tentatively characterized in SGT. Here we choose liquiritigenin and its analogs, as well as their glycosides, as examples to describe their structural characterization.

Characterization of Triterpene Saponins
Triterpene saponins in SGT are mainly derived from Platycodi Radix and licorice, and most of them are oleanane-type triterpenes, which contain triterpene oleanolic acid or its analogs as the sapogenin and one or two oligosaccharide chains substituted at 3-OH or 28-OH. In the negative ion mode, they could lose sugar residues or sapogenins to produce corresponding diagnostic fragments. Based on this, the structures of 14 selected triterpene saponins were identified or tentatively characterized in this study.

Characterization of Phenolic Acids
Almost all of the component herbs contain phenolic acids, and a total of 10 selected phenolic acids in SGT were identified or tentatively characterized in this study. In the negative ion mode, phenolic acids could lose a carboxy group with the neutral loss of 44 Da (CO 2 ) to generate diagnostic fragments, and the loss of small molecules (like H 2 O) or radicals (like CH 3 ·) sometimes also occurred. For example, the [M-H] − ion of ferulic acid (42) (23), where caffeic acid forms an ester bond with quinic acid, the ester bond fragmented to produce m/z 191.0514 and m/z 179.0460, representing a quinic acid unit and a caffeic acid unit, respectively. Based on these fragmentation patterns, the other selected phenolic acids (1, 7, 13, 15, 22, 38, 41, and 45) were tentatively characterized Yan et al., 2014;Lee et al., 2015;Zhang et al., 2016;Zhu et al., 2017).

CONCLUSION
In summary, we developed an off-line HILIC × RP 2D-LC system to comprehensively separate the chemical constituents in SGT and its component herbs. The 2D-LC system showed high orthogonality (63.62%) and approximate 11-fold improvement in peak capacity in contrast to conventional one-dimensional RPLC separation. As a result, a total of 465 peaks were detected, and the structures of 54 selected compounds were fully identified or tentatively characterized by qTOF-MS analysis. In addition, 12 peaks detected in SGT were not discovered in any component herbs, and they might contribute partly to exploration of the compatibility mechanism of the component herbs. Integration of off-line HILIC × RP 2D-LC and high-resolution mass spectrometry is proven as a promising tool for chemical profiling and comparison of complicated Chinese patent medicines and their component herbs.