Integration of fingerprint-activity relationship and chemometric analysis to accurately screen Q-markers for the quality control of traditional Chinese medicine compound preparation: Jinlian Qingre granules as an example

2.2.1 Sample solution preparation

The sample was crushed into powder using a mortar. The sample powder (0.50 g) was accurately weighed, and 50 % (v/v) methanol aqueous solution (25 mL) was accurately measured in a 50 mL conical flask with cover ultrasound in a water bath (120 W, 40 k HZ) at room temperature for 30 min. Methanol aqueous solution (50 %) was used to compensate for reduced weight. The appropriate sample solution was placed in an EP tube and centrifuged at 10,000 rpm for 15 min. Then, the supernatant was injected into the instrument for further analysis.

2.2.2 Negative sample solution preparation

According to the prescription ratio and the process conditions of extraction, drying, and molding of JQG, negative samples lacking FT, AR, and SR were prepared. Negative sample solutions were prepared using the same method described above.

2.2.3 Standard solution preparation

Harpagide, amygdalin, mangiferin, 2″-O-beta-L-galactopyranosylorientin, orientin, veratric acid, vitexin, indirubin, indigo, anemarrhena saponin B II, catalpol, and harpagoside were accurately weighed and separately dissolved in methanol to get individual stock standard solutions. The stock standard solutions were appropriately diluted by methanol to prepare the mixed standard solution. A mixed standard solution consisting of all twelve reference standards above was prepared for UHPLC-HRMS analysis. In addition, a mixed standard solution composed of 31.8 μg/mL mangiferin, 118 μg/mL 2′′-O-beta-L-Galactopyranosylorientin, 188 μg/mL orientin, 28.6 μg/mL veratric acid, 55.6 μg/mL vitexin, and 12.0 μg/mL harpagoside was prepared for quantitative analysis. Then, it was diluted by methanol to obtain the serial working standard solutions.

2.4.1 HPLC conditions

HPLC analysis was conducted using a SHIMADZU HPLC system (SHIMADZU, Japan) consisting of an LC-20AT pump and an SPD-20A DAD detector. An Agilent Eclipse Plus C18 liquid chromatography column (4.6 × 250 mm, 5 µm) was utilized. The injection volume, flow rate, and column temperature were 10 µL, 1 mL/min, and 30 ℃, respectively. The mobile phases were acetonitrile (A) and a 0.1 % phosphoric acid aqueous solution(pH = 2.5) (B). The volume change of the mobile phase was as follows: 0–5 min, 97 % B; 5–15 min, 97–95 % B; 15–25 min, 95–90 % B; 25–91 min, and 90–68 % B. The detection wavelength was 270 nm in 0–18 min, 210 nm in 18–36 min, 270 nm in 36–70 min, and 210 nm at 70–91 min.

2.4.2 Methodological verification

Fingerprinting, stability, precision, and repeatability of the analysis method were examined to verify the applicability of HPLC. The precision was evaluated by conducting six replicates of the same sample solution and calculating the relative standard deviation (RSD) derived from the relative retention time (RRT), relative peak area (RPA) and relative peak height (RPH) of the common peak. Repeatability was assessed by injecting six solutions from the same batch prepared in parallel. Sample stability was determined by re-analyzing the identical sample solution at various time intervals within a 24-h period (0, 2, 4, 6, 8, 10, 12, and 24 h).

2.6.1 Sample preparation

Different proportions of JQG (DPJQGs, labeled N0 ∼ N20) samples were prepared to ensure variations in the chemical composition. The proportion of herbs in DPJQGs was determined using a uniform design and the Data Processing System (Table S2). The ratio of N0 was consistent with that of JQG. The extraction method employed for DPJQG preparation followed the guidelines outlined in ChP 2020. This involved fire-boiling the herbs twice with water, followed by filtration. Then, the resulting twice-combined-extraction decoction was concentrated under reduced pressure and spray-dried. Negative control samples were prepared using the same method. The sample solution of each DPJQG was prepared using the same method as in section “2.2.1”, and the common peak content was analyzed using the HPLC conditions described in section “2.4.1”. These samples were also accurately weighed separately and dissolved in DMEM to obtain samples for cell experiments.

2.6.2 Cell viability assay

The CCK8 assay was used to evaluate the RAW 264.7 cell viability. RAW 264.7 cells (3 × 10⁴ wells/100 μL) were inoculated into a 96-well plate. After 24 h, the old culture medium was removed, and different concentrations of JQG sample solution were added to continue cultivation for 24 h. After removing the old culture medium, 10 μL CCK8 and 90 μL DMEM were added and incubated at 37 ℃ for 2 h. Each well's optical density (OD) was measured at 450 nm using a microplate absorbance meter. Each drug concentration was repeated for threefold repetition, and all measurements were conducted in three independent experiments.

2.6.3 Cytokine measurement

LPS stimulates an inflammatory response in RAW 264.7 cells, and JQG suppresses the inflammatory response. Inflammatory cytokines were quantitatively analyzed using a reagent kit to evaluate the anti-inflammatory efficacy of JQG. The cells were inoculated into 96 well plates (2 × 10⁴ wells/100 μL) and treated with DPJQG in the presence of LPS. The cells treated with DMEM complete basic medium were used as the blank control. The cells treated with LPS alone were used as the model control group. The cells treated with LPS and aspirin (ASP) were used as the positive control group. Following a 24-h incubation period, TNF-α and IL-6 levels were analyzed in the cell supernatant per the protocols provided by the manufacturer.

2.7.1 PLSR

PLSR was established to screen anti-inflammatory active compounds, with common peak areas set as independent variables and inhibition rates of TNF-α and IL-6 as dependent variables. All modeling and analysis procedures were executed using SIMCA 14.1 software.

2.7.2 GA-BPNN

A GA-BPNN model was developed to depict the intricate and nonlinear correlation between 24 independent variables (24 common peak areas) and two dependent variables (inhibition rates of TNF-α and IL-6). Firstly, normalize the common peak areas of 21 samples and inhibition rates of TNF-α and IL-6. Then, use the common peak area as the input layer neuron, the inhibition rate as the output neuron, and select one layer from the middle layer neuron to establish a BP neural network. Through repeated experiments and optimization, the number of hidden layer neurons was ultimately determined to be 6, and the algorithm was trainlm. Calculate the MIV values of the inhibition rates of TNF-α and IL-6 for each common peak using the optimal GA-BP model, clarify the relative importance of each variable's impact on the network output, and identify potential active components.The GA-BPNN fitness was evaluated using statistical metrics, including root mean square error (MSE). All modeling and analysis procedures were performed using MATLAB 2019b.

2.8.1 Inhibition of TNF-α and IL-6

Six reference compounds (mangiferin, 2″-O-beta-L-galactopyranosylorientin, orientin, veratric acid, vitexin, and harpagoside) were dissolved in a 0.1 % DMSO solution and diluted with DMEM to an appropriate concentration to conduct anti-inflammatory experiments (measuring TNF-α and IL-6 release inhibition rates) and assess the rationality of determining the efficacy of compounds, following the procedure outlined in section “2.6.3”.

2.8.2 Molecular docking verification

Protein structure was downloaded from the RCSB PDB database (https://www.rcsb.org). The 3D structures of the chemical compounds were determined using PubChem (https://pubchem.ncbi.nlm.nih.gov). Finally, molecular docking was conducted using the potential targets and their respective constituents.

2.9.1 Chromatographic conditions, preparation of sample solutions, and standard solutions

The content analysis of chromatographic conditions was the same as in section ‘‘2.4.1″. Meanwhile, the sample and standard solution preparation was the same as in section ‘‘2.2.1″ and “2.2.3”.

2.9.2 Validation of the method

Method validation, including linearity, precision, repeatability, stability, and recovery rate, was performed to ascertain the suitability of the HPLC analysis method. The standard solutions were diluted with methanol at six different concentrations to establish standard curves. The linearity of the calibration curves was evaluated by plotting the peak area against the corresponding concentration of each standard. Subsequently, the regression equation, correlation coefficient, and linear range were determined based on the obtained curves. Limit of detection (LODs) and quantification (LOQs) were determined at signal-to-noise (S/N) ratios of 3:1 and 10:1, respectively. The precision was evaluated by conducting six replicates of the same sample solution and calculating the RSD derived from the peak area (PA) of the compound being measured. Both intra-day and inter-day precision were evaluated. The repeatability was assessed by calculating the RSD derived from the content of the compound measured in six parallel samples. The stability was assessed by calculating the RSD derived from the PA measured and re-analyzing the identical sample solution at various time intervals within a 24-h period (0, 2, 4, 6, 8, 10, 12, and 24 h). The recoveries were calculated using the concentration (100 %, equivalent to 1.0 times the original amounts in samples), and the concentrations were analyzed six times.

3.2.1 Optimization of HPLC conditions

The optimization of HPLC conditions, encompassing the selection of the chromatographic column, mobile phase, flow rate, detection wavelength, and column temperature, was conducted systematically to attain a desirable separation and improved analytical efficiency. Section “2.4.1” provides detailed information regarding chromatographic conditions. During the establishment of a fingerprint, the characteristic peak distributions at various wavelengths and time intervals are compared to identify distinctive peaks, leading to the selection of different detection wavelengths at different time intervals (Chen et al., 2023).

3.2.2 Methodology validation

P17 (orientin) was selected as the reference peak due to its superior peak shape and large peak area. Furthermore, the RSD of the RRT, RPA and RPH for the other 23 common peaks were calculated to assess precision, stability, and repeatability. The precision test results indicated that the RSD values for RRT, RPA, and RPH were below 1.12 %, 1.94 %, and 1.57 %, respectively. Similarly, the repeatability test results demonstrated that RRT, RPA, and RPH were less than 1.05 %, 2.98 %, and 1.38 %, respectively. The stability test results demonstrated that the RSD values for RRT, RPA, and RPH of each chromatographic peak were below 1.41 %, 2.84 %, and 2.14 %, respectively. These findings suggest that the method is accurate and reliable, making it suitable for establishing HPLC fingerprints for sample analysis.

3.2.3 Establishment of HPLC fingerprint

HPLC analysis of 20 (S1–S20) batches of JQG samples yielded the fingerprints of JQG, and HPLC data were collected. The peak matching was performed using the “TCM Chromatographic Fingerprint Similarity Evaluation System (Version 2012)” software. The reference fingerprint was the S1 sample, and the “time width” was 0.1 min. Twenty-four peaks (P1-P24) were identified as common peaks. JQG quality was assessed by evaluating similarity. Twenty-four common peaks were calibrated, and their similarity was calculated. Table 2 presents the results. The obtained similarity values between the 20 batches of JQG and the reference fingerprint exceeded 0.97, indicating that the JQG preparation exhibited uniformity, stability, and controllability. However, the RSD of RPA for each common peak ranged from 5.44 % to 29.39 %, indicating that despite the significant consistency observed among various batches of JQG, discernible differences persisted. Integrating fingerprint analysis with chemometric techniques is imperative for identifying the compounds responsible for these variations.

Based on the RT of the peak and the photodiode array detector (PDA) spectrogram, 6 of the 11 compounds identified by reference substance comparison in UHPLC-HRMS analysis, could be detected under optimized HPLC chromatographic conditions, namely P15 mangiferin from the minister drug AR, P16 2′'-O-beta-L-galactopyranosylorientin, P17 orientin, P19 veratridic acid, P20 vitexin from monarch drug FT, and P23 harpagoside from the adjuvant drug SR. Fig. 2A displays the HPLC chromatograms of references, JQG, and negative samples. Fig. 2B indicates the fingerprints of 20 batches of JQG samples and their reference fingerprints. Fig. 2C depicts the UV spectrogram of the six references and the corresponding peaks in the sample.

Close

Fig. 2

A: Chromatograms: references (a), JQG (b), the negative samples that lack FT (c), the negative samples that lack AR (d), and the negative samples that lack SR (e). B: The fingerprints of 20 batches of JQG samples. C: Comparison of the spectrogram of six compounds. 15. mangiferin, 16. 2′′-O-beta-L-galactopyranosylorientin, 17. orientin, 19. veratric acid, 20. vitexin, 23. harpagoside.

Export to PPT

3.3.1 HCA

Chromatographic common peaks were investigated using HCA to classify samples of JQG (Shi et al., 2023). Fig. 3A indicates that the 20 batches of JQG can be classified into different categories. At an Euclidean distance of 25, various batches of JQG can be classified into two categories: S1–S13 and S14–S20. Table S1 displays that each product category was prepared from a different batch number of extractum. The classification results were related to raw material production. The presence of different categories may lead to variations in quality due to disparities in raw materials. The results also suggest that products of similar quality can be grouped based on 24 common peaks.

Close

Fig. 3

HCA results (A). PCA score scatters plot (B). OPLS-DA score scatter plot (C). The OPLS-DA VIP results (D).

Export to PPT

3.3.2 PCA

PCA is currently the prevailing technique for unsupervised classification visualization. PCA effectively represents the data characteristics of the original variables while minimizing information loss by employing dimensionally reduced variables (Fan et al., 2022). For PCA model construction, 24 common peak areas from 20 batches of JQG samples were imported into SIMCA 14.1 software, and the score scatters plot of PCA was acquired. To predict discrimination in the model, the parameters R² and Q² were calculated. The result indicated the R² X and Q² were 0.828 and 0.513, respectively, indicating that the model had good prediction ability. Fig. 3B depicts establishing a two-dimensional PCA model in which all data points are contained within 95 % confidence intervals, and 20 batches of JQG can be separated into two distinct categories consistent with HCA. This suggests the presence of quality disparities among different batches of JQG.

3.3.3 OPLS-DA

OPLS-DA can promote sample separation and explore group differences while reducing sample dispersion within categories (Wang et al., 2023). Fig. 3C indicates that 20 batches of JQG are divided into two groups, and the results are consistent with those of PCA. The results indicate that JQG varies between different batches. The model underwent 200 substitution tests to assess its performance, resulting in interpretation rate parameters for the model (R²X and R²Y) and prediction parameters (Q²). The parameters were 0.569, 0.937, and 0.729, respectively, proving that the model was reliable. Furthermore, the VIP plot (Fig. 3D) effectively demonstrates the significant contribution of nine specific chemical constituents (P21, P15, P9, P18, P1, P13, P17, P6, P24, P20, P10, P16, and P14) to the categorization process. These are the key chemical ingredients responsible for the differences in JQG quality.

According to the above analysis, unsupervised and supervised pattern recognition approaches based on 24 common peaks could effectively distinguish JQG samples with quality differences, and 24 common peaks are of great significance for evaluating JQG quality.

3.4.1 Research on anti-inflammatory effects

3.4.1.2

3.4.1.2 Inhibition of inflammatory cytokines

The anti-inflammatory efficacy of DPJQG was assessed by quantifying TNF-α and IL-6 in cell supernatants. Fig. 4A and 5B illustrate that DPJQG can inhibit inflammatory cytokine production. The inflammatory cytokine content was significantly increased in the model group than in the control group (p < 0.01), whereas inflammatory cytokines were significantly suppressed in the aspirin group than in the model group (p < 0.01). Furthermore, different DPJQG samples have different inhibitory efficacies, indicating that their ratio composition influences the anti-inflammatory activity of DPJQGs. The dissimilarities in the chemical compositions of DPJQGs may account for the variations in their anti-inflammatory properties.

Close

Fig. 4

Effect of DPJQG samples on TNF-α levels (A) and IL-6 (B) in the supernatant of LPS-induced RAW 264.7 cells inflammation model (n = 3. data are expressed as the mean ± SEM. ^####P < 0.001 contrast with the control group, ****P < 0.001 contrast with the model group). Regression coefficients of 24 common peaks of JQG in the PLS models of TNF-α (C) and IL-6 (D). Variable Importance of 24 common peaks of JQG in the GA-BPNN models of TNF-α (E) and IL-6 (F).

Export to PPT

3.4.2 Determination of anti-inflammatory active ingredients in JQG

PLSR and GA-BPNN were utilized to analyze the association between common peaks and anti-inflammatory activity in DPJQG and screen anti-inflammatory active compounds.

3.5.1 Inhibition of inflammatory cytokines

The cytotoxicity of six compounds on RAW 264.7 cells was evaluated to mitigate the influence of diminished activity on inflammatory factor production. Fig. S3 displays the observed cytotoxicity. TNF-α and IL-6 were quantitatively measured to evaluate the anti-inflammatory properties of each compound using the maximum non-toxic concentration of each compound as the therapeutic concentration. Fig. S4 indicates that each compound significantly inhibits inflammatory cytokine production. Therefore, six selected chemical components exhibited good anti-inflammatory activity.

3.5.2 Molecular docking studies

The docking analysis involved the examination of Q-markers with TNF-α (PDB ID: 1A8M) and IL-6 (PDB ID: 1ALU) using data from the PDB database, providing comprehensive information on proteins and their 3D structures (Li et al., 2023). PyRx software was used to minimize the energy of the isolated and identified compound molecular structure. When combined with AutoDock Tools software, the predicted target proteins are dehydrated, hydrogenated, added with atomic charges, set with atomic types, and stored as a backup for receptors. Then, AutoDock Vina in PyRx software was used to dock the identified compounds with receptor proteins, and the results are shown in Fig. 5A. Fig. 5A indicates that the absolute values of the docking scores are all greater than 5. The active ingredients obtained by screening have a strong binding capacity with anti-inflammatory-related proteins, thereby proving the reliability of the selected quality markers. Fig. 5B and 5C exhibit the binding diagrams of orientin with TNF-α and IL-6. This was consistent with previous report that mangiferin(Kumar et al., 2023), orientin, vitexin(Yu et al., 2022), veratric acid, 2′'-O-beta-L-galactopyranosylorientin(Wang et al., 2015; Liu et al., 2018) and amygdalin(Zeng et al., 2023) independently exhibited stronger anti-inflammatory effects and further confirmed the reliablility of these compounds as the main material basis for anti-inflammatory activity of JQG.

Close

Fig. 5

Docking scores of six compounds with TNF-α and IL-6 (A). Molecular docking analysis of orientin with TNF-α (B) and IL-6 (C).

Export to PPT

3.6.1 Methodology validation of quantitative analysis

Validation tests were conducted using JQG (sample S1) on HPLC. Table S4 depicts the specific results of precision, repeatability, and stability. The range of spike-recoveries was 95.69 %∼101.42 %, and Table S5 presents specific information. Table 3 illustrates the regression equations. The LOD and LOQ values are 0.063 to 0.125 and 0.250 to 0.500 μg/mL, respectively. The r values of the established linear equations were greater than 0.9998 within the corresponding concentration range. These results indicate that this method is highly accurate and can meet the requirements for content determination.

3.6.2 Calculation and robustness of RCF

Determining and calculating the RCF are key issues that must be addressed to promote and apply QAMS. RCF is a relatively fixed constant that must be established by determining all reference solutions for the testing components. However, redundant research in the field of QAMS is present, where identical research objects, components to be measured, and analysis methods are employed to repeatedly establish RCFs with different RS in different studies. This practice leads to the squandering of resources and goes against the original purpose of establishing the QAMS. Establishing a QAMS with different substances, such as RS, can quantify multiple components in the presence of only one type of RS.

RCF can be determined using a linear model, that incorporates six distinct concentrations of standard substances and their corresponding peak areas. RCF values were calculated employing one of the orientin, 2′'-O-beta-L-galactopyranosylorientin, and vitexin as the RS. Table S6 displays the calculation results. Subsequently, the established RCF was employed to calculate the content of the other substances to be tested (ST). ESM and QAMS results were compared by calculating the standard method difference (SMD) (Wu et al., 2021). The calculation formulas for RCF, content, and SMD are as follows:

where f_RS/ST refers to the RCF value of the analyte, A_RS and C_RS are the peak area and concentration of the RS (one of the 2′'-O-beta-L-galactopyranosylorientin, orientin, and vitexin), respectively; A_ST and C_ST are the peak area and concentration of the ST, respectively; C_ESM and C_QAMS represent the substance concentrations to be measured using ESM and QAMS methods, respectively.

3.6.3 Ruggedness test

The robustness of the RCF was investigated by changing the column temperature and flow rate, as well as using different apparatus and chromatographic columns. Tables S7–9 indicate that the established RCF has good robustness within the linear range.

3.6.4 Quantitative analysis of Q-marker in JQG

Table 4 indicates the specific contents of the six Q-markers in 20 batches of JQG, calculated using ESM and QAMS. Fig. 6A compares the two method results and calculates the SMD values, indicating that the difference between ESM and QAMS was not statistically significant. This indicates that the RCF value can be used to quantify the compounds in JQG.

Close

Fig. 6

A: SMD between QAMS using 2′′-O-beta-L-Galactopyranosylorientin as RS and ESM (a). SMD between QAMS using orientin as RS and ESM (b). SMD of QAMS using vitexin as RS and ESM (c). B: Cluster analysis of six Q-marker contents in JQG.

Export to PPT

The comparison was performed that the 4 batches sample results observed in this study with standard method in ChP 2020. The content of vitexin in S16-S19 JQG samples using the method in ChP 2020 was 0.685 mg/g, 0.656 mg/g, 0.626 mg/g, 0.648 mg/g, respectively. The difference was not statistically significant.

3.6.4 HCA based on the contents of six Q-markers

The HCA of 20 batches of JQG was conducted based on the contents of six Q-markers using SPSS Statistics 22.0 (Armonk, NY, USA). Fig. 6B indicates that the clustering results are the same as those in Fig. 3A. These results also suggest that products of similar quality can be grouped. Consequently, mangiferin, 2′'-O-beta-L-galactopyranosylorientin, orientin, veratric acid, vitexin, and harpagoside were identified. It can be used to distinguish between different JQG quality levels and assess JQG quality.