Identification of Ginseng Radix et Rhizoma, Panacis Quinquefolii Radix, Notoginseng Radix et Rhizoma, and Platycodonis Radix Based on UHPLC-QTOF-MS and “Matrix Characteristics”

Abstract

Ginseng Radix et Rhizoma (GRR), Panacis Quinquefolii Radix (PQR), Notoginseng radix et rhizoma (NRR) and Platycodonis Radix (PR) are often confused in the material market because of similar appearances and characteristics. Moreover, chemical identification methods tend to characterize the whole herb with regard to a single or a few components, which is an inaccurate representation and does not demonstrate the effective utilization of unknown components, and the result is unconvincing. In order to strengthen quality control, improve identification efficiency, and realize digital identification at the individual level of traditional Chinese medicine (TCM), we have put forward the “matrix characteristics” of TCM, combined with a UHPLC-QTOF-MS analysis to explore and realize the digital identification of GRR, PQR, NRR, and PR. The mass spectrometry was quantized to extract common data from different batches of the same TCMs as their matrix characteristics, and the matching credibility (M) was given by matching the “matrix characteristics” with unknown Chinese medicines. The results show that within a reasonable parameter threshold range, the M of four TCMs was higher than 92.00% compared with their own “matrix characteristics”, which was significantly higher than the M ranked second. Furthermore, the digital identification of four TCMs can be successfully realized based on the UHPLC-QTOF-MS analysis and “matrix characteristics”. This has important reference significance for developing the digital identification of GRR at an individual level based on UPLC-QTOF-MS and “matrix characteristics”.

1. Introduction

The dried roots and rhizomes of the plant Panax ginseng C.A. Mey from the Araliaceae family are known as Ginseng Radix et Rhizoma (GRR, Renshen) and the King of Herbs in TCMs [1,2]. It has a variety of effects, such as tonifying vital energy and the spleen and lungs, promoting the production of fluids and nourishing the blood, calming the spirit, and benefiting the intellect. In clinical practice, it is used to treat fatigue, type II diabetes, and inflammation and to develop immunity [3,4,5,6]. Panacis Quinquefolii Radix (PQR), known as Xiyangshen in China, refers to the dried root of Panax quinquefolium L., which is in the family Wujiaceae, and has the effects of invigorating qi and nourishing yin, clearing away heat and promoting fluid production. It is often used for qi and Yin deficiencies, heat exhaustion, asthma and phlegm, dry mouth, and pharynx [7,8]. The Notoginseng Radix et Rhizoma (NRR, Sanqi) is the dried root and rhizome of Panax Notoginseng (Burk.), F.H. Chen of Araliaceae, and has the effects of removing blood stasis, stopping bleeding, reducing swelling, and relieving pain. In clinical use, it can be used to treat cardiovascular diseases, pain, inflammation, trauma, and internal and external bleeding due to injury [9,10]. In addition, Platycodonis Radix (PR, Jiegeng) is the dried root of Platycodon grandiflorum (jacq.) A.DC., a plant of the family Campanulaceae, and has the effects of dispersing lung qi, relieving sore throat, eliminating phlegm, and discharging pus [11]. It is often used in clinical settings for coughing with excessive phlegm, chest tightness, sore throats, and hoarse voice [12].

For the above purpose, numerous research studies have been conducted by various researchers from different perspectives. For example, Zhang et al. simultaneously determined 10 kinds of ginsenoside in American ginseng by ultra-high-performance liquid chromatography and used them for identification and quality evaluation [13]. Brown et al. suggested that six major ginsenosides (Rg₁, Re, Rb₁, RC, Rb₂, and RD) may be used as the quality control index of GRR and PQR [14]. Based on the differences in the plots of ginsenoside components in the terahertz band, Kou et al. combined chromatography with principal component analysis to accurately distinguish GRR and PQR [15]. Wang et al. established a UHPLC-MS/MS method for the determination of PQR in GRR using ginsenoside F11 as a detection index [16]. Jiang et al. developed a multiplexed site-specific PCR method to identify GRR, PQR, and NRR, and the results showed that the specific bands of about 250, 500, and 1000 bp appeared in GRR, NRR, and PQR, respectively [17]. In addition, in China Pharmacopoeia, the quality control identification component of PR is platycodin D (C₅₇H₉₂O₂₈) [18]. These research studies have contributed to the identification and quality control of GRR, PQR, NRR, and PR to a certain extent. However, it is not difficult to discover that the current analytical approach is more inclined towards adopting a strategy of partially reflecting the whole; that is, the targeted identification studies have been conducted based on a single or a few chemical compounds with known molecular formulas and clear structures. The above-mentioned approach’s drawbacks are that (1) Chinese herbal medicine often contains thousands of compounds, which implies a complex system, so it is difficult to characterize the whole of Chinese herbal medicine only by a single or a few chemical components, (2) the analysis focuses on the chemical components with a known molecular formula and clear structure and fails to make reasonable and effective use of information of unknown components, and (3) in the digital era of Chinese medicine, what is particularly important is that current traditional identification methods fail to realize the digital identification and analysis of Chinese medicine at the individual level.

In view of the insufficiencies of current analytical methods, as well as the background of digitization and intellectualization of TCMs, this study aims to use UHPLC-QTOF-MS data for the characterization of chemical components to establish dynamic proprietary “matrix characteristics” of GRR, PQR, NRR, and PR in the form of a digital matrix so as to realize the informative and rapid non-targeted identification, at the individual level, of Chinese medicines, rather than focus on chemical components at the molecular level. Firstly, UHPLC-QTOF-MS was utilized to analyze GRR, PQR, NRR, and PR under unified analysis conditions. Secondly, Progenesis QI software (version 2.4.69) was used to digitize the mass spectrometry of GRR, PQR, NRR, and PR samples to obtain the digital characterization of chemical components [19]. Then, the algorithm flow of dynamic “matrix characteristics” of GRR, PQR, NRR, and PR was established using the Java programming language on the basis of digital characterization. Finally, the dynamic matrix characteristics were locked by a parameter control to realize the informative and rapid digital identification analysis of four TCMs.

2. Materials and Methods

2.1. Herbs and Reagent Materials

Eight batches of GRRs, 10 batches of PQRs, 10 batches of NRRs, 8 batches of PRs, 1 batch of GRRR, and 1 batch of SR were collected from the National Institutes for Food and Drug Control in China. All the medicinal materials were identified as genuine by the research team and met the requirements of 2020 China Pharmacopoeia. Detailed information such as batch number, year and so on about standard samples is shown in Table S1. Methanol (Lot: ED341-CN) and formic acid (L1670) were purchased from Honeywell Trading Co., Ltd. of Shanghai, China. Acetonitrile (Lot: 222372) was purchased from Thermo Fisher Scientific shier Technology Co., Ltd. of Shanghai, China, and ultrapure water (GB 19298) [20] was purchased from Watsons Food and Beverage Co., Ltd. of Guangzhou, China.

2.2. Sample Pretreatment

Sample pretreatment using a uniform ultrasonic extraction method: 1.00 g of powder of GRR, PQR, NRR and PR samples was accurately weighed and placed in a 50 mL conical flask; then, 25.00 mL of methanol was accurately added by pipette for sonication for 0.50 h (power: 500 W, frequency: 40 kHz); finally, the samples were allowed to cool and filtered through 0.22 μm organic filtration membranes to obtain the analyzed samples [21]. Before UHPLC-QTOF-MS analysis, the samples were stored in a refrigerator at 4 °C.

2.3. UHPLC Analysis Conditions

Chromatographic separations were conducted on Waters Acquity^TM UHPLC instrumentation (Waters, Milford, MA, USA) and Waters Acquity UHPLC BEH−C₁₈ (2.1 mm × 100 mm, 1.7 μm) chromatographic column (lot: 186002352) (Waters, Milford, MA, USA). The column temperature and injection volume were 35 °C and 2.0 μL. The mobile phases were 0.1% formic acid in water (A phase)−acetonitrile (B phase), and the gradient elution conditions were as follows: 0–23 min, 5–95% B; 23–26 min, 95% B; 26–26.10 min, 95–5% B; and 26.01–30 min, 5% B [22].

2.4. Mass Spectrometry Conditions

Mass spectrometry was performed on a QTOF Synapt G2 system (Waters MS Technologies, Manchester, UK) equipped with an electrospray ionization (ESI) source [23]. Data were collected in the continuum mode and the positive-ion mode. The capillary and sampling cone voltage were set at 3.0 KV and 30 V, respectively. The scan range was from 100 Da to 1500 Da, with 0.2 s scan time. The source and desolvation temperatures were kept at 150 °C and 450 °C, respectively. The desolvation and cone gas flow rates were set at 600 L/h and 50 L/h, respectively.

2.5. Data Processing and Analysis

2.5.1. Data Processing

The mass spectrometry information of GRR, PQR, NRR, PR, and blank solvent was processed by Progenesis QI software (Version 2.4.69, Milford, MA, USA) [24]. The parameters were set as follows: High resolution mass spectrometer; Ionization polarity: Positive; Peak picking limits: Automatic; Retention time limits:1.00~26.00 min; High energy limits: based peak 0.2%. Based on the above parameters, we obtained the quantized data of each sample, including the retention time (Rt), the mass-to-charge ratio (m/z), and ionic strength (I). Further, the quantized data of ions was uniformly integrated and saved as a CSV file in the following form:

$$\mathrm{Blank} \mathrm{or} \mathrm{TCM}\text{-}\mathrm{DIC} (\mathrm{name})=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ t& m& i\end{array}\right]$$

In the above form of a digital matrix, [Rt-m/z-I] is the basic unit of the data array, which represents one chemical component in digital form, where Rt represents polarity of the chemical component, and m/z represents relative molecular mass of component or fragmentation ion, while the ionic strength reflects content to a certain degree. On the other hand, GRR, PQR, NRR and PR, as Chinese herbal medicines, contain thousands of chemical components. In other words, from the perspective of digitization of traditional Chinese medicines, four Chinese herbal medicines can all be represented by a digital matrix containing thousands of [Rt-m/z-I] units.

2.5.2. Identification Algorithm Flow

In short, the process of identification and analysis roughly includes two steps: one is the dynamic acquisition of “matrix characteristics” of GRR, PQR, NRR and PR; the other is non-targeted identification of the basic “matrix characteristics” that were locked by parameter threshold settings. The algorithm flow of dynamic acquisition of matrix characteristics is as follows:

Suppose the digital matrices of blank samples (B₁ and B₂) and two samples A₁ and A₂ of the same traditional Chinese medicine A in different batches are recorded as b₀, b₁, x₁ and x₂, respectively.

$${\mathrm{b}}_0=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_{\mathrm{b}0}& {{\displaystyle m}}_{\mathrm{b}0}& {{\displaystyle i}}_{\mathrm{b}0}\end{array}\right] {\mathrm{b}}_1=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_{\mathrm{b}1}& {{\displaystyle m}}_{\mathrm{b}1}& {{\displaystyle i}}_{\mathrm{b}1}\end{array}\right] x_1 =\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_1& {{\displaystyle m}}_1& {{\displaystyle i}}_1\end{array}\right] x_2 =\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_2& {{\displaystyle m}}_2& {{\displaystyle i}}_2\end{array}\right]$$

Firstly, the union of blank was defined as x₀; it can be expressed as

$$x_0={\mathrm{b}}_0\cup {\mathrm{b}}_1=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_{\mathrm{b}0}& {{\displaystyle m}}_{\mathrm{b}0}& {{\displaystyle i}}_{\mathrm{b}0}\end{array}\right]\cup \left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_{\mathrm{b}1}& {{\displaystyle m}}_{\mathrm{b}1}& {{\displaystyle i}}_{\mathrm{b}1}\end{array}\right]=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_0& {{\displaystyle m}}_0& {{\displaystyle i}}_0\end{array}\right]$$

At the same time, taking the deviation of Rt and m/z as control parameters. In addition, the deviation thresholds of Rt and m/z were recorded as DT(t), DT(m/z), respectively. Under the limited analysis conditions, for a unit in the x₀ and x₁ digital matrix, if |t₀ − t₁| > DT(t) or |m₀ − m₁| > DT(m/z), it meant that their units ([t₀ − m₀ − i₀] and [t₁ − m₁ − i₁]) in the digital matrix could not match each other, representing different chemical compositions in blank solvent and traditional Chinese medicine A₁, and were not common features. On the contrary, if |t₀ − t₁| ≤ DT(t) & |m₀ − m₁| ≤ DT(m/z), it is considered that [t₀ − m₀ − i₀] in digital matrix x₁ was equal to [t₁ − m₁ − i₁] in digital matrix x₂, representing the same digital characterization in blank solvent and traditional Chinese medicine A₁. Then, we took the “intersection data” that meets the DT(t) and DT(m/z) as the core data, and summarized it into a new digital matrix x₃, which is recorded as:

$$x_3=x_0\cap x_1=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_0& {{\displaystyle m}}_0& {{\displaystyle i}}_0\end{array}\right]\cap \left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_1& {{\displaystyle m}}_1& {{\displaystyle i}}_1\end{array}\right]=\left[\begin{array}{ccc}Rt& m/z& I\\ {{\displaystyle t}}_0& {{\displaystyle m}}_0& {{\displaystyle i}}_0\\ {{\displaystyle t}}_1& {{\displaystyle m}}_1& {{\displaystyle i}}_1\end{array}\right]$$

In the same way, we can obtain the digital matrix of x₄ that is the “intersection data” that meets the DT(t) and DT(m/z) for x₀ and x₂ digital matrices:

$$x_4=x_0\cap x_2=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_0& {{\displaystyle m}}_0& {{\displaystyle i}}_0\end{array}\right]\cap \left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_2& {{\displaystyle m}}_2& {{\displaystyle i}}_2\end{array}\right]=\left[\begin{array}{ccc}Rt& m/z& I\\ {{\displaystyle t}}_0& {{\displaystyle m}}_0& {{\displaystyle i}}_0\\ {{\displaystyle t}}_2& {{\displaystyle m}}_2& {{\displaystyle i}}_2\end{array}\right]$$

For two samples (A₁ and A₂) of same traditional Chinese medicine A in different batches, we can obtain the data matrix x_A1, x_A2 of A₁ and A₂ after blank removal:

$${x_A}_1=x_1-x_3=x_1-x_0\cap x_1 =\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_1& {{\displaystyle m}}_1& {{\displaystyle i}}_1\end{array}\right]-\left[\begin{array}{ccc}Rt& m/z& I\\ {{\displaystyle t}}_0& {{\displaystyle m}}_0& {{\displaystyle i}}_0\\ {{\displaystyle t}}_1& {{\displaystyle m}}_1& {{\displaystyle i}}_1\end{array}\right]=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_{A1}& {{\displaystyle m}}_{A1}& {{\displaystyle i}}_{A1}\end{array}\right]$$

$${x_A}_2=x_2-x_4=x_2-x_0 \cap x_2=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_2& {{\displaystyle m}}_2& {{\displaystyle i}}_2\end{array}\right]-\left[\begin{array}{ccc}Rt& m/z& I\\ {{\displaystyle t}}_0& {{\displaystyle m}}_0& {{\displaystyle i}}_0\\ {{\displaystyle t}}_2& {{\displaystyle m}}_2& {{\displaystyle i}}_2\end{array}\right]=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_{A2}& {{\displaystyle m}}_{A2}& {{\displaystyle i}}_{A2}\end{array}\right]$$

On the above basis, the core digital matrix x_A of Chinese medicine A can be obtained by taking “intersection data” through the same parameter control of DT(t), DT(m/z). It can be expressed as

$$x_A={x_A}_1\cap {x_A}_2=\left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_{A1}& {{\displaystyle m}}_{A1}& {{\displaystyle i}}_{A1}\end{array}\right]\cap \left[\begin{array}{ccc}Rt& m/z& I\\ \cdots & \cdots & \cdots \\ {{\displaystyle t}}_{A2}& {{\displaystyle m}}_{A2}& {{\displaystyle i}}_{A2}\end{array}\right]=\left[\begin{array}{ccc}Rt& m/z& I\\ {{\displaystyle t}}_{A1}& {{\displaystyle m}}_{A1}& {{\displaystyle i}}_{A1}\\ {{\displaystyle t}}_{A2}& {{\displaystyle m}}_{A2}& {{\displaystyle i}}_{A2}\end{array}\right]$$

Analogously, for the same traditional Chinese medicines of many different batches, their shared data were also obtained as the core digital matrix of these traditional Chinese medicines based on the above method. Furthermore, as shown in Figure 1, the ionic strength was used as a feature screening indicator for ranking, and the Top-N or top percentile units ([Rt-m/z-I]) were output by dynamically adjusting the output number to construct a new data matrix as “matrix characteristics” for traditional Chinese medicines. In summary, the “matrix characteristics” of GRR, PQR, NRR and PR can be dynamically controlled by parameter settings of DT(t), DT(m/z) and output number (n) or percentage.

On the other hand, for the unknown TCMs, UHPLC-QTOF-MS analysis was carried out under the same conditions to obtain mass spectrometry information. After digital quantization, the data matrix of unknown traditional Chinese medicine was also obtained based on characterization of Rt, m/z, and I. Based on the obtained “matrix characteristics” of GRR, PQR, NRR, and PR, and the [Rt-m/z-I] in “matrix characteristics” as the matching unit, the digital matrix of unknown traditional Chinese medicine was compared and matched in turn according to the Rt. If and only if the threshold conditions of selected parameters DT(t), DT(m/z), and output number (n) were met, the matching was considered successful. In addition, the number of successfully matched units and that in the Chinese herbal “matrix characteristics” were denoted as X and Y, respectively. Then, we defined the matching credibility (M) as:

$$\mathrm{M}={\displaystyle \frac{X}{Y}}\times 100\%$$

By matching unknown samples with the four Chinese medicines GRR, PQR, NRR, and PR, the corresponding matching credibility can be obtained and summarized into a new array [M_GRR, M_PQR, M_NRR, M_PR]. Based on the Top-N algorithm, the Chinese herbal medicine with relatively high matching credibility can be obtained to realize rapid and non-targeted identification.

To summarize, the above algorithm can extract shared ion information of different batches of the same Chinese medicines and sort them according to the ionic strength from the largest to the smallest, finally using the Top-N ions matrix as the “matrix characteristics” to match the quantized data of test samples and feedback matching credibility. In this paper, we used the mass spectrometry information of six samples from different batches of GRR, PQR, NRR, and PR to extract the core digital matrix and “matrix characteristics” of four Chinese herbal medicines for identification analysis. In the process of obtaining “matrix characteristics”, the DT(t) and DT(m/z) were all set to different values. Then, we used the mass spectrometry data to extract the “matrix characteristics” for internal verification. At the same time, additional samples were taken and re-sampled for analysis for external validation.

3. Results

3.1. Results of UHPLC-QTOF-MS Analysis

Under the unified sample processing and experimental conditions, we obtained the base-peak chromatograms of GRR, PQR, NRR, and PR.

As shown in Figure 2, GRR, PQR, NRR, and PR showed different base-peak chromatograms under the same detection conditions, which means that essentially, these four herbs have different chemical compositions and can provide a basis for establishing “matrix characteristics” for four herbs based on their chemical compositions and further digital identification analysis [21]. In addition, for small batches of the four herbs, identification can be achieved based only on the spectrogram comparison, but if there are many samples, a spectrogram comparison based only on the manual is inefficient, so it is necessary to combine it with information technology to carry out digital identification and improve the efficiencies of identification and analysis.

3.2. The Results of Digital Quantization Processing

In digital quantization processing, the mass spectrometry information of three methanol samples was taken as “blank”. They contained 3608, 3948, and 3457 [Rt-m/z-I] units, respectively, after being converted into a data matrix [21]. The mean value was 3671 with an RSD of 6.85%. For GRR, PQR, NRR, and PR samples that were used to extract “matrix characteristics”, the number of [Rt-m/z-I] units of the four Chinese medicines is shown in

Table 1. The number of [Rt-m/z-I] units in GRR, PQR, NRR, and PR samples.

TCM	Batch	Number	RSD	TCM	Batch	Number	RSD
PR	20141101	5287	14.5%	GRR	20111001	7683	3.6%
	20141102	5496			20111002	7909
	20161201	5669			20121101	8519
	20161202	5736			20121102	8025
	20211301	7320			20171201	8203
	20211302	7146			20171202	8257
NRR	20110801	8145	4.6%	PQR	20060801	7127	8.4%
	20110802	7898			20060802	7005
	20140901	7787			20130901	6413
	20140902	7507			20130902	6590
	20181001	7121			20181001	7892
	20181002	7577			20181002	7769

.

As shown in Table 1, the number of [Rt-m/z-I] units in GRR and NRR samples exceeded 7000, with RSD values of 3.6% and 4.6%. Meanwhile, the number of PQR and PR sample units was more than 5000, with RSD values of 8.4% and 14.5%, respectively. The above results showed that the mass spectrometry information for samples from different batches (different years or producing areas) of the same TCMs differed, and there were different chemical components at the molecular level. For example, the number of units in the PR samples of the 20211301 batch was significantly higher than that in the PR samples of the 20141101 batch. This may be due to the difference in chemical composition caused by different producing areas or storage time, which is in line with the actual situation. However, this difference in different batches of the same Chinese medicines did not affect our research, mainly because we relied on the digital characterization of chemical components at the molecular level by mass spectrometry to extract common data as a core digital matrix and constructed the “matrix characteristics” for TCMs based on the core digital matrix. This academic thought focused on the common chemical components of the same Chinese herbal medicines from different batches (different points of origin, different years) rather than the different chemical components. In addition, the digital characterization of mass spectrometry information was used instead of identifying compounds [22].

3.3. Results of Algorithm Programming

Based on algorithm flow, with static files as the background storage, we used Java to write the programmed commands of the dynamic acquisition of “matrix characteristics” and the matching algorithm compared with unknown Chinese medicine. At the same time, we used Node.js to build the front end and realize the digital identification analysis relying on “matrix characteristics” of chemical components [25]. The whole framework is shown in Figure 3.

3.4. Results of Identification Analysis and Verification

On the basis of algorithm programming, we realized the recognition analysis and validation of GRR, PQR, NRR, and PR by adjusting the matching parameters DT(t), DT(m/z), and output number (n) of “matrix characteristics” compared to the digital matrix of unknown Chinese medicines. In order to ensure the normal operation of the algorithm program, we first carried out internal verification for recognition analyses of PR (batch: 20211301) and GRR (batch: 20121101) that were used to extract “matrix characteristics”. When DT(t) = 0.05, DT(m/z) = 0.01, and output number n = 200, both PR and GRR have matching credibility of more than 96.00% compared with their stored “matrix characteristics”, which had a high degree of recognition. The specific results can be found in

Table 2. The matching credibility (M) of PR and GRR compared with matrix characteristics.

TCM	Match Units	Number of Units in “Matrix Characteristics”	TCM	M(%)
PR	194	200	PR	97.00
	29	200	GRR	14.50
	8	200	PQR	4.00
	7	200	NRR	3.50
GRR	200	200	GRR	100.00
	139	200	PQR	69.50
	131	200	NRR	65.50
	55	200	PR	27.50

Deviation Settings: DT(t) = 0.05, DT(m/z) = 0.01, n = 200 [22].

. Further nonparametric tests (data non-normality) showed that the M of PR and GRR compared to their own “matrix characteristics” was significantly different from the M of PR and GRR compared to non-self “matrix characteristics” (p = 0.046 < 0.050). Further, when the DT(t), DT(m/z), and output number n were set to 0.05, 0.005, and 10%, the matching credibility of NRR (20181002) compared with “matrix characteristics” of NRR, GRR, PQR, and PR were 99.31%, 67.94%, 66.18%, and 26.49%, respectively. In addition, the matching credibility of PQR (20181001) compared with “matrix characteristics” of PQR was more than 95.90% when DT(t), DT(m/z) was set to 0.05, 0.005. The output numbers n were set to 10% and 20%, respectively, which was significantly higher than the maximum matching credibility of 75.24% for GRR ranked second in the above situation. Internal verification proved that the matching identification algorithm routine based on “matrix characteristics” and UHPLC-QTOF-MS is reasonable and feasible.

During external verification, we entrusted other researchers in the laboratory to randomly select samples (X₁, X₂, Y₁, Y₂ that we did not know beforehand) from four Chinese medicines to re-prepare samples and analyze them by UHPLC-QTOF-MS for identification analysis based on “matrix characteristics”. At the same time, two other Chinese medicines—Ginseng Radix et Rhizoma Rubra (GRRR, batch: 20180601) and Scrophulariae Radix (SR, batch: 20211101)—that do not belong to these four Chinese medicines were selected by sample preparation personnel for identification analysis. The results showed that based on the matching identification of “matrix characteristics”, the matching credibility of X₁ and GRR was as high as 92.00% when the DT(t), DT(m/z), and output number n were set to 0.05, 0.001, and 200. When DT(t) and output number were constant, DT(m/z) was changed to 0.01, 0.02, and 0.05, the matching credibility of X₁ and GRR was all more than 97.00%, which was significantly higher than the second-ranked M = 70.00% of X₁ and PQR. So, we finally identified X₁ as GRR through many times of parameter controls. After checking with the sample preparation personnel, Chinese medicine X₁ was GRR with the batch number 20121103. In the same way, we identified X₂ as NRR with M = 98.0%, which was higher than M = 73.75% compared with PQR when DT(t) = 0.05, DT(m/z) = 0.01, and n = 400, which was in line with the actual situation under which the staff selected NRR with batch number 20181003 for sample preparation. The matching credibility of Y₁ and PR was 95.50%, which was more than five times that (17.00%) of the second-ranked GRR when DT(t) = 0.05, DT(m/z) = 0.005, and n = 400, while the matching credibility of Y₂ and PQR was 98.60%, which was higher than that (77.00%) of Y₂ compared with the second-ranked GRR when DT(t) = 0.05, DT(m/z) = 0.01, and n = 500. The matching information of Y₁ and Y₂ is detailed in Tables S2 and S3. A nonparametric test combining Table 1 and Table 2 showed that the M of Y₁ and Y₂ compared to PR and PQR “matrix characteristics” was significantly different from the M of Y₁ and Y₂ compared to other “matrix characteristics” (p < 0.050). Finally, we identified Y₁ and Y₂ as PR and PQR, respectively, in line with the actual situation under which the researchers selected PR (Y₁, 20211303) and PQR (Y₂, 20181003) for sample preparation.

In addition, we analyzed the GRR, PQR, NRR, and PR samples that do not belong to the origins of herbs used to extract “matrix characteristics”. The results showed that the matching credibility of samples compared with their own “matrix characteristics” was no less than 92.00% when DT(t) = 0.05, DT(m/z) = 0.05, and n = 100, which was at least 20 percentage points higher than that of the Chinese medicine with the second highest match credibility. The results can be observed in Table S4. Similarly, non-parametric tests showed significant differences (p = 0.004). They also proved that the “matrix characteristics” of four TCMs have certain applicability and can be used for external identification and analysis.

When DT(t), DT(m/z), and output number n were set to 0.05, 0.001, and 200, the matching credibility of GRRR and GRR was 29.00%, which was the highest among the four traditional Chinese medicines. Meanwhile, the matching credibility of X₁ and GRR was 92.00%. Further, the matching credibility of GRRR and GRR was 35.50% when we turned up DT(m/z) to 0.05, with the matching credibility of X₁ and GRR being 97.50%. All the above differences indicated that it is difficult for GRRR to match the “matrix characteristics” of NRR, GRR, PQR, and PR. As for SR, When the DT(m/z) = 0.05 and DT(t) = 0.05 were constant, and the number of output was 100, 200, 30%, 60%, and 100%, respectively, among the four herbs, all SR had the highest matching credibility with PR, but all of the matching credibility value were also lower than 5.10%. If we set the parameters to DT(m/z) = 0.1, DT(t) = 0.1, and output number n= 500, we obtained the max matching credibility 9.20% of SR compared with PR. In the case where the threshold value of DT(m/z), DT(t) = 0.1 exceeds reality, the above extremely low matching credibility just explained that it is impracticable for SR to match the “matrix characteristics” of NRR, GRR, PQR, and PR. In summary, based on UHPLC-QTOF-MS and “matrix characteristics”, identification analysis can be quickly and accurately realized.

4. Discussion

4.1. Present Situation and Digitization of TCM Identification

Currently, the mainstream method and strategy for the identification analysis of TCMs is to adopt the exclusive analysis method for specific TCMs failing to make reasonable and effective use of the information of unknown components. On the other hand, the traditional identification methods can not meet the current demand for rapid inspection. With the advent of information and the digital era, where is the identification analysis of traditional Chinese medicine going? The identification analysis of TCM can develop in the direction of digital non-targeted identification at the individual level of TCM, which is conducive to the breakthrough of rapid detection technology and meets the needs of rapid identification analysis at present [26,27,28].

4.2. Discussion of Research Ideas and Strengths

In the background of the digitization of TCMs, we carried out this study in order to realize the rapid digital identification of NRR, GRR, PQR, and PR. To this end, for Chinese medicines, we put forward the concept of “matrix characteristics”, which is a variable digital matrix for a specific Chinese medicine identity. In this study, we relied on UHPLC-QTOF-MS data to obtain matrix characteristics of four traditional Chinese medicines, i.e., by extracting common data from different batches of the same Chinese medicine and representing it in the form of a digital matrix that contains retention time (Rt), mass-to-charge ratio (m/z) and ionic strength (I) [21,22]. Moreover, the research was not focused on the fragmentation pathways of chemical components, nor the identification analysis of specific chemical components. However, it was to realize the digital identification of Chinese medicines at the individual level based on the quantized characterization of chemical components. It eliminates the necessity of identifying the chemical structures. The characterization of holistic composition allows for more faithful identification.

The advantage of identification analysis based on matrix characteristics is that it does not need to identify the specific compound structure and molecular formula, and it can incorporate all the obtained quantized data into the digital matrix to characterize the individual Chinese medicine. The method can fully use the information of unknown components to make the analysis results more reasonable and reliable. For example, under the mass spectrometry analysis of GRR, it is easy to identify the chemical composition Ginsenoside Rg₁ (C₄₂H₇₂O₁₄, m/z = 823.4820, [M+Na]⁺), Ginsenoside F₁ (C₃₆H₆₂O₉, m/z = 677.4033, [M+K]⁺), Ginsenoside Rh₄ (C₃₆H₆₀O₈, m/z = 621.4392, [M+H]⁺), and Coumaric acid (C₉H₈O₃, m/z = 165.0536, [M+H]⁺), etc. However, there were thousands of unknown components in GRR, and the proportion of known components was too small. It was difficult to characterize GRR only by known components. To look at it another way, unknown components, like known components, can be represented by the unit of [Rt-m/z-I], which provides conditions for us to use unknown components. In addition, compared with traditional identification methods, the digital identification analysis of “matrix characteristics” based on chemical composition can greatly improve the analytical efficiency and conserve workforce and financial resources.

As we all know, saponins are highly susceptible to in-source fragmentation in positive polarity [29], leading to more than one precursor ion for the same chemical compound. Indeed, if these precursor ions are selected as [Rt-m/z-I] units simultaneously, the number of significance signals is less than the number of output “n”. However, from the data analysis perspective, the characterization data of these precursor ions can also be regarded as “characteristic ions information”. Moreover, the second screening will be carried out to acquire the “matrix characteristics” according to ionic strength order. Many other types of chemical components are also screened, in addition to saponin components. Therefore, their influence on identification analysis is limited. In addition, we can adjust the ion output number of the “matrix characteristics” many times and obtain the final recognition result according to many comparisons.

In addition, compared to DNA molecular technology, this study can greatly improve the efficiency of analysis and save time. Moreover, DNA molecules are at risk of degradation and are susceptible to breakage during sample preparation. On the other hand, relevant researchers also used a digital identification strategy to identify traditional Chinese medicines based on UHPLC-QTOF-MS. However, their research only retained the mass spectrometry information with an abundance greater than 1.0 × 10⁴, which may cause data loss and false positive results, particularly unfavorable for identifying homologous herbal medicines [23,30]. In addition, in their research, the m/z relative error of 10 ppm was adopted uniformly, which was not conducive to the matching identification of small molecular compounds. However, in this paper, we used absolute error for match identification. The absolute error of m/z can be as low as 10 mDa while retaining a threshold setting window that has wide applicability, which improves matching accuracy and allows for appropriate parameters to be set as needed. Moreover, to retain as much mass spectrometry information as possible in the TCMs, we uniformly eliminated blanks rather than setting limits on abundance.

4.3. Optimization of UHPLC-QTOF-MS Analysis

UHPLC-QTOF-MS is widely used in the research field of traditional Chinese medicine. it can provide us with massive data information that reflects chemical components [14,30]. In UHPLC-QTOF-MS analysis of NRR, GRR, PQR, and PR, we first examined the mass spectral information of positive and negative ions. It was found that the mass spectrometry information of four herbal medicines in positive ion mode was significantly more than that in negative ion mode conditions. The ion intensities in positive ion mode were generally higher. Therefore, it was finally determined that the samples were tested under positive ion conditions. Further, we explored the sample pretreatment methods such as extraction solvent methanol, ethyl acetate and dichloromethane under ultrasound or heating reflux based on the principle of obtaining more mass spectrometry data information and convenient operation. The results found that the mass spectrometry information obtained when methanol was used as extraction solvent was significantly higher than that when dichloromethane and ethyl acetate were used as solvents. Moreover, there was no difference between ultrasonic extraction and heat reflux treatments. So, we finally chose ultrasonic extraction (power: 500 W, frequency: 40 kHz) with methanol as the solvent for 30 min as the sample pretreatment method. In addition, we investigated the mass spectrometry information of four TCMs when the collision energies were 10 V~40 V, 10 V~60 V, and 10 V~80 V, respectively. It was clear that the mass spectrometry had the most abundant data information, with the collision energy being 10 V~40 V. This universal condition ensures the accuracy and repeatability of data collection in different periods.

4.4. Sample Selection Principles

The samples’ representativeness, diversity, and compliance were critical in constructing the “matrix characteristics” of the four herbs. Therefore, we should consider authentic medicinal herbs first. Moreover, each sample batch must comply with the national pharmacopoeia standards. In addition, the chemical composition of the same Chinese medicine differs from different origins and periods. From the diversity perspective, we need to obtain samples from different origins and periods. On the other hand, the inclusion of multiple batches provides robustness to the results. However, there is always a limit to the number of samples that can be analyzed. More samples may affect the efficiency of the analysis and the results. We initially proposed at least three origins (including daodi producing areas), with at least two batches per origin (at different times). If more samples are available, ensure the accuracy of data collection. Its can be controlled by methodology and chemical reference standards, such as Ginsenoside Rg1.

4.5. Parameter Selection, Setting and Suggestion

Setting a parameter threshold is very important in digital identification analysis. In this study, we set the parameters of “matrix characteristics” acquisition separately from matching identification of unknown herbal medicines. This separation design can temporarily store “matrix characteristics” and use them as a benchmark for matching identification, thus avoiding recalculating “matrix characteristics” when the parameters change and improving the efficiency of matching identification.

DT(t) is the deviation threshold of retention time that reflects the polarity of chemical components, and DT(m/z) is the deviation threshold of m/z that reflects the molecular weight of chemical components. In addition, the output number n can control the number of [Rt-m/z-I] units in the digital matrix by sorting the ionic strength (I). So DT(t), DT(m/z), and output number n were selected as control parameters. In the above process, ionic strength was not taken as an important control parameter, mainly because ionic strength reflected the content of components to a certain extent, and even the same Chinese medicines varied greatly in different batches, so it was difficult to control the deviation threshold of ionic strength. On the contrary, DT(t) and DT(m/z) are relatively stable. Further, under the unified detection conditions, we found that DT(t) is often in the range of 0.01~0.05 and DT(m/z) is often in the range of 0.001~0.05, which fluctuates due to the instrument state. Considering the actual situation and referring to the relevant literature, we set the DT(t), DT(m/z) in obtaining “matrix characteristics” to 0.05, which is also a value that we suggest setting [23,31]. As for the parameters of matching identification, we set DT(t) to 0.05 and DT(m/z) to the range of 0.001~0.05, and we recommend that DT(m/z) should be set preferentially to 0.001 or 0.01 to improve the matching accuracy [21,22]. In addition, we suggest that the output number n can be set in the range of 200~1000 preferentially. Of course, this is not absolute, and one can also try to change multiple parameters to assist identification.

4.6. Research Shortcomings and Prospects

This study proposed “matrix characteristics” for the digital non-targeted identification of GRR, PQR, NRR, and PR. However it still has some limitations: although the samples used in this study came from different origins, batches and years, the total sample size was small, and further samples need to be added for subsequent analysis. Fortunately, the shared ion extraction algorithmic process for the same traditional Chinese medicine in this study has a wide range of applications, and the amount of data can be expanded at any time. In addition, the UHPLC-QTOF-MS instrument is expensive and difficult to popularize. It may be used as a complementary means of identification or as a preliminary digital screen. Moreover, identification analysis based on matrix characteristics of chemical composition may become an important means of quality control of TCMs. It is possible to establish an identification database that contains many TCMs based on their “matrix characteristics” and conduct identification analysis according to the differences between different “matrix characteristics” of TCMs.

On the other hand, theoretically, the “matrix characteristics” constructed based on chemical compositions also apply to identifying other herbal medicines. However, more herbs were not studied in this study and could be further explored subsequently. In the future, we can also try to construct “matrix characteristics” based on the different chemical compositions of different Chinese medicines to identify Chinese medicines.

5. Conclusions

In this paper, the unified UHPLC-QTOF-MS method was used to analyze GRR, PQR, NRR, and PR. After quantization, the common data of different batches of the same Chinese medicines were extracted to establish “matrix characteristics” for non-targeted identification. The research confirmed that the identification of four traditional Chinese medicines can be successfully realized based on UHPLC-QTOF-MS and “matrix characteristics”. It has important reference significance for developing non-targeted identification of GRR at the individual level of TCM based on UHPLC-QTOF-MS and “matrix characteristics”.