Discovery of quality markers in the rhizome of Atractylodes chinensis using GC–MS fingerprint and network pharmacology

2.2.1 Test conditions for GC–MS

The analysis was carried out using a gas chromatography–mass spectrometry system (GC–MS 7890A-5975C, Agilent, Santa Clara, CA, USA) and the NIST14.L mass spectrometry search database. The chromatographic column was HP-5 MS (Agilent, Santa Clara, CA, USA, 30 m × 0.25 mm × 0.25 µm) with the following parameters: injection port temperature of 280 °C; the carrier gas was high purity helium (He); injection volume of 1 μL; the split ratio of 10:1 (v:v); flow rate of 1 mL·min⁻¹; programmed ramp-up: initial temperature of 100 °C; ramp-up to 135 °C at 10 °C·min⁻¹; and ramp-up to 250 °C at 0.5 °C·min⁻¹. An electron ionization (EI) source was used with an ionization energy of 70 eV, an ion source temperature of 230 °C, a quadrupole temperature of 150 °C, full wavelength scanning, and a solvent delay period of 3 min.

2.2.2 Preparation of samples

All samples were ground to powder and screened with a 50-mesh sieve. Then, 1.00 g of RAC powder was suspended in a cone bottle with 5 mL of n-hexane solution. The sample solution was obtained after ultrasonic extraction for 30 min. After making up the lost weight with n-hexane, the solution was filtered through a 0.22 μm microporous membrane.

2.2.3 Methodological validation

The developed GC–MS method for qualitative analysis of RAC was validated in terms of precision, stability, and repeatability under the optimized conditions described above. Six consecutive injections of a sample solution were performed to verify the accuracy. Six independent replicates of the same sample solution were prepared to measure repeatability. Stability was determined by testing the same sample solutions under the same conditions at 0, 2, 4, 6, 12, and 24 h. The RSD values of the RPA and RRT of the common peaks were calculated for each test.

2.2.4 Data analysis

The RAC fingerprint was established by GC. Combining the reference substance and the GC–MS NIST14.L database, the common peaks with large peak areas and good resolution in the fingerprint were identified, and the characteristic spectrum of RAC was determined. The “Chromatographic Fingerprint Similarity Evaluation System for Traditional Chinese Medicine” (2012 version) was used to evaluate the similarity of the GC fingerprint of RAC to determine its overall quality. The software SIMCA-P 14.1 (Umetrics, Umeå, Sweden) was used to perform HCA, PCA, and OPLS-DA. SPSS 26.0 (SPSS Inc., Chicago, IL, USA) was used to calculate the variance and correlation analysis. The bioinformatics online website (https://www.bioinformatics.com.cn/) was used to create the box diagram.

2.3.1 Screening of RAC targets

The Canonical SMILES format of compounds in RAC was searched in the PubChem database (https://pubchem.ncbi.nlm.nih.gov/), and imported to the Swiss Target Prediction (https://www.swisstargetprediction.ch/), to predict potential targets for the compound. Remove duplicates and targets with a “Probability” of “0” and construct compounds–targets network diagram using Cytoscape 3.9.0 software.

2.3.2 Construction of PPI network

The protein–protein interaction (PPI) network plays a key role in predicting the function of target proteins and the drug ability of molecules, which handle a wide range of biological processes, including cell-to-cell interactions and metabolic and developmental control. The STRING database (https://cn.string-db.org/) was used to analyze the interaction between each target using default parameters, with species selection “Homo sapiens” and the “minimum required interaction score” ≥ 0.700. Remove unrelated independent targets from the PPI network. The PPI network was analyzed using Cytoscape’s built-in plug-in, and targets above mean degree centrality (DC), betweenness centrality (BC), and closeness centrality (CC) were selected as core targets. These are the primary targets for the chemical components in RAC to exert their pharmacological effects. DC is equal to the number of links to a node and reflects how often a node interacts with others. BC represents the number of shortest paths through a given node. The higher the BC value, the more powerful the node is at controlling other nodes in a network. CC represents the average shortest path from one node to all other nodes. The higher the CC of a node, the easier it is for other nodes to reach it and for it to reach out to other nodes.

2.3.3 KEGG pathway enrichment analysis

The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was used to identify key genes and pathways involved in RAC. The DAVID 6.8 database (https://david.ncifcrf.gov/) was used for KEGG pathway enrichment analysis of RAC targets, and p ≤ 0.05 was used as the screening criterion. Meanwhile, fold enrichment was introduced to demonstrate the significance of the pathway. The greater the fold enrichment, the greater the gene enrichment. Finally, the pathways were visualized using an online bioinformatics website.

2.3.4 Compounds–targets–pathways network

The “active ingredients–targets–pathways” network was constructed using the Cytoscape 3.9.0 software. Nodes in the network represent components, core targets, or core pathways, and edges represent interactions between them. Moreover, the network was analyzed using three thresholds, CC, BC, and CC, to screen the components, targets, and pathways that contributed most to the network.

3.1.1 Establishment of the GC fingerprint of RAC

The GC fingerprints of the 10 batch samples are shown in Fig. 2. The similarity analysis results were obtained by importing the chromatogram data into the “Chinese Medicine Chromatographic Fingerprint Similarity Evaluation System ”. The fingerprint contained 20 characteristic peaks that were used to calculate the similarity indices. The similarity analysis results in Table S1 showed that the similarity was greater than 0.923, indicating that each sample had a good consistency and stable quality. By comparing the mass spectra obtained from the analyses with those from NIST14.L, reference standards, and published reports, we were able to identify 16 of the 20 common ingredients (Table 2). As shown in Table S2, the relative percentage content of 16 chemical components in batches of RAC exceeded 70% (70.98%–89.96%). These 16 components were considered potential Q-markers and were used in the subsequent analysis.

Close

Fig. 2

GC chromatographic fingerprint of 10 batches of the rhizome of Atractylodes chinensis (RAC). (A) Local magnification of retention time of 4–10 min in GC fingerprint, a total of 9 common peaks in this interval. (B) The GC global fingerprint, a total of 9 common peaks were labeled in retention time intervals of 24 to 46 min. Detailed information on the common compounds is shown in Table 2.

Export to PPT

3.1.2 Validation of the GC fingerprint method

For the precision study, the relative retention time (RRT) and relative peak area (RPA) of peak 20 (atractylodin) were chosen as the reference, and the RRT and RPA of the 20 common peaks of each sample were calculated. The relative standard deviation (RSD) of the RPA of each common peak were < 5.00%, and the RRT of each common peak was < 0.30% (Table 3). In the repeatability test, the RSD of RPA for each peak were < 5.00%, and the RSD of RRT were < 0.20%. For the stability test, the sample solution was measured at 0, 2, 4, 6, 12, and 24 h after preparation, and then the RRT and RPA were calculated. The RSD of RRT was < 0.40%. The RSD of RPA was < 5.00%. The above results indicate that the precision and repeatability of the GC method were satisfactory. Additionally, the RAC sample solution was stable within 24 h.

3.2.1 Target network analysis of RAC

A total of 135 targets of 16 constituents (Table S3) were obtained in Swiss Target Prediction. Based on these data, the ingredients–targets network was constructed, as shown in Fig. 3A. The network had 150 nodes and 268 interactions, indicating that RAC exerted pharmacological effects via multiple components and targets. In particular, the number of gene targets corresponding to atractylon, hinesol, β-eudesmol, elemol, atractylodin, and valencene were 54, 53, 29, 26, 25, and 17, respectively, indicating that the above components are closely related to the pharmacological activity of RAC.

Close

Fig. 3

Target network analysis of the rhizome of Atractylodes chinensis (RAC). (A) “Ingredients–targets” network consisting of 16 components in RAC and 135 targets. (The hexagon is the ingredient and the circle is the target.) (B) PPI network and interaction diagram of core targets. The PPI network consisting of 135 targets was further analyzed and 29 core targets were identified based on the thresholds of DC ≥ 3.80, BC ≥ 339.24, and CC ≥ 0.08. (The size and color were correlated to the degrees of targets in network: the big size and deep color with blue means high degree of this target.).

Export to PPT

The PPI network was created using the STRING database to analyze the potential relationship between 135 targets. After removing an unconnected node, the PPI network contained 135 nodes and 224 edges, with an average node degree of 3.32 and an average local clustering coefficient of 0.528, p < 1.0 × 10^-16. Based on the threshold values for DC ≥ 3.80, BC ≥ 339.24, and CC ≥ 0.08, 29 core targets were identified (Fig. 3B). The above targets are critical for RAC to exert pharmacological effects. At the same time, the targets were found to interact with each other, indicating that RAC exerts its drug effect through the synergistic effect of multiple targets.

3.2.2 Pathway enrichment and components–targets–pathways network analysis

To further elucidate the mechanism and efficacy of RAC, we analyzed 29 key target genes interacting with the KEGG pathway using the data extracted from the DAVID database. There were 23 pathways with p < 0.05, as shown in Fig. 4A. These pathways can be divided into five broad categories: cellular processes, metabolism, environmental information processing, organismal systems, and human diseases. Metabolism includes steroid hormone biosynthesis and arachidonic acid metabolism. Environmental information processing includes neuroactive ligand-receptor interaction and calcium signaling, VEGF signaling, and cAMP signaling pathways. Organismal systems include the PPAR signaling pathway, serotonergic synapse, estrogen signaling pathway, ovarian steroidogenesis, adipocytokine signaling pathway, bile secretion, inflammatory mediator regulation of TRP channels, and C-type lectin receptor signaling pathway. In addition, pathway enrichment analysis shows that RAC is closely related to human diseases including insulin resistance, diabetic cardiomyopathy, Alzheimer’s disease, and cancer.

Close

Fig. 4

Network pharmacology diagrams of RAC: (A) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis involved in 29 core targets. (The bubble size represents the number of enriched genes in the pathway, and the larger the bubble, the more enriched genes there are; the color of the bubble represents the significance of enrichment, i.e., p-value, and the darker the color, the more significant the enrichment of the pathway.) (B) Active components–targets–pathways network consisting of 16 components, 29 core targets and 23 pathways. (The hexagon is the active ingredient, the circle is the target, and the rectangle is the pathway.).

Export to PPT

A comprehensive “components-targets-pathways” network was built by entering 16 components, 29 core targets, and 23 pathways into Cytoscape to investigate the relationship between active components and pathways in RAC. As shown in Fig. 4B, the network diagram consisted of 68 nodes and 177 edges, indicating the intricate correlations between different compounds and targets. Based on the threshold values of DC ≥ 5.13, BC ≥ 143.52, and CC ≥ 0.35, atractylon (DC = 14.00, BC = 803.79, and CC = 0.45), hinesol (DC = 12.00, BC = 614.95, and CC = 0.42), and β-eudesmol (DC = 7.00, BC = 199.83, and CC = 0.35) as the primary bioactive ingredients in RAC, thereby designating them as candidate Q-markers. These ingredients act principally on PPARA, PIK3CA, PTGS2, CYP19A1, AR, ESR1, CYP2C19, and RXRA, affecting primary pathways such as neurogenesis, cancer, calcium signaling, and PPAR signaling pathways.

According to the above research, the candidate Q-marker screened by network pharmacology does not include atractylodin, which is the only index component recorded in Chinese pharmacopoeia for quantitative detection. Numerous studies have demonstrated that atractylodin is unsuitable for use as a test index (Liu et al., 2022; Zhao et al., 2022). The rhizome of A. lancea, which traditionally believed to be of superior quality, is deficient in atractylodin, rendering it inadequate for meeting the standards of pharmacopoeia. Conversely, the rhizome of A. chinensis considered to be of inferior quality, often contains more atractylodin (Zhang et al., 2010, Chen et al., 2013). The present study also established that the quality of Cangzhu was not highly correlated with atractylodin.

3.4.1 HCA and PCA analysis

HCA, also known as unsupervised learning or exploratory data analysis, deconstructs a limited data set into a finite, discrete set of “natural” hidden data structures (Cao et al., 2019). In this study, HCA was performed on 18 batches of RAC and 10 batches of RAJ using SIMCA software and RPA of three components as variables. Based on the results of cluster analysis displayed in Fig. 5A. It can be concluded that all samples can be roughly divided into two categories, with RAJs comprising cluster I and wild and cultivated RACs comprising cluster II.

Close

Fig. 5

Chemometrics analysis of the rhizome of Atractylodes chinensis (RAC) samples. (A) Hierarchical cluster analysis (HCA). All samples can be roughly divided into two categories, with RAJs comprising cluster I and wild and cultivated RACs comprising cluster II. (B) Principal component analysis (PCA). (C) Orthogonal partial least squares–discriminant analysis (OPLS-DA). (D) Variable importance of the projection (VIP) values. The VIP values of atractylon, β-eudesmol, and hinesol were 1.48, 0.80, and 0.42, respectively.

Export to PPT

Through data analysis and visualization, PCA can reduce data by geometrically projecting it onto smaller dimensions known as principal components (PCs) (Tan et al., 2020). The goal is to find the best summary of date using a small number of PCs. To evaluate and distinguish RAJ and RAC with two different growth patterns, we used PCA analysis with standard data consisting of three components. The PCA′s two-dimensional score plot is shown in Fig. 5B. The first two PCs accounted for 74.50% and 18.10% of the variance, respectively. The tasted samples were divided into two groups, which are similar to the HCA results. Orthogonal partial least squares–discriminant analysis (OPLS-DA) is a supervised discriminant analysis statistical method, which is different from the PAC method. In the model, the first two components described 100.0% of the variation in X (R²X = 1.000) and 87.6% of the variation in Y (R²Y = 0.876); Q² was 0.837. These results demonstrate the appropriateness of the model. The OPLS-DA analysis yielded the same results as the PCA analysis (Fig. 5C). Moreover, the separation between the two sample types was more apparent in the OPLS-DA model.

Variable Importance in Projection (VIP) scores estimate the importance of each variable in the projection used in a partial least-squares (PLS) model and are often used for variable selection (Zhou et al., 2022). In a given model, a variable with a VIP score close to or greater than 1 may be considered significant. As shown in Fig. 5D, the VIP values of atractylon, β-eudesmol, and hinesol were 1.48, 0.80, and 0.42, respectively. Thus, atractylon and β-eudesmol were identified as the main substances causing RAC variability from various origins. The results showed that multiple components, rather than a single component, influenced the variability of RAC quality.

The above results show that the three Q-markers can clearly distinguish RAJ and RAC, which is important for identifying RAC and its counterfeit products. However, there is still overlap, which may indicate there is a separation trend between cultivated and wild RAC, but this is not obvious. In addition, we found that the spots representing cultivated RAC is more concentrated, whiles the distribution of the spots of wild RAC is more scattered, suggesting that the quality of artificially grown RAC is more stable. To further differentiate RAC samples with two different growth modes, the contents of components were analyzed using a box diagram and an ANOVA.

3.4.2 Content difference and correlation analysis

The HCA and PCA demonstrated that the 28 batches of “Cangzhu” were divided into two groups, indicating that the three index components from various sources had different compositions. Therefore, ANOVA was used to analyze variations in Q-markers content. In addition, the Pearson correlation (PC) was used to investigate the relationships between the three components of RAC.

Box plot is an analytical method that uses interval scale measurement data. It is mainly used to indicate if the distribution is skewed and if there are outliers in the data set (Williamson et al., 1989). In this study, a two-by-two comparison of wild RAC, cultivated RAC and RAJ was performed by box plots to analyze the differences between RAC and RAJ and the differences between RACs of two different growth patterns. As shown in Fig. 6, the relative percentages of atractylon, hinesol, and β-eudesmol in the RAC in artificial cultivation were 6.89 ± 2.82%, 2.51 ± 0.95%, and 8.64 ± 1.93%, respectively, whereas the relative percentages in the wild RAC were 5.97 ± 1.91%, 2.46 ± 1.63%, and 7.10 ± 3.28%, respectively. In RAJ, the relative percentages of the three ingredients were 22.17 ± 5.87%, 1.60 ± 1.19%, and 2.45 ± 1.12%, respectively. The two components of RAJ, atractylon, and β-eudesmol, differed significantly from RAC. The relative content of atractylon in RAJ was more than twice that of RAC, and the relative content of β-eudesmol was significantly lower than that of RAC, indicating a significant difference in the quality of the two “Cangzhu”. Additionally, the contents of the three components in RAC grown using two different methods were approximately the same, and the statistical results revealed that none of the three components in either of them had significant differences, even though wild RAC is traditionally considered to be of superior quality. Moreover, the box plot and PCA results revealed that the relative contents of the marker components in wild RAC samples varied significantly, especially the contents of hinesol, and β-eudesmol. This could suggest that the quality of wild RAC is more variable, while cultivated RAC is more homogeneous. This maybe because the natural environment of wild RAC is more volatile than the growth environment of cultivated RAC, which is more stable.

Close

Fig. 6

Variation analysis of relative percentage contents in 18 batches of the rhizome of Atractylodes chinensis (RAC) and 10 batches of Atractylodes japonica (RAJ). (A) atractylon. (B) hinesol. (C) β-eudesmol. *** p < 0.001, ** p < 0.01, * p < 0.05, no was considered no significant difference.

Export to PPT

Meanwhile, we examined the correlation among three components in 18 batches of RAC. The results are presented in Table 5, which demonstrate a highly significant positive correlation between hinesol and β-eudesmol in RAC, with a PC coefficient of 0.625 (p < 0.01). This finding is congruent with previous research, potentially due to the analogous chemical structures of hinesol and β-eudesmol, both of which possess two alicyclic rings and an isopropyl hydroxyl group, and consequently may have similar biochemical synthesis pathways (Wu et al., 2008). Overall, the RAC volatile oil fractions of hinesol and β-eudesmol manifest a synergistic variability and warrant further investigation.