Discovery of Anti-inflammatory Ingredients in Chinese Herbal Formula Kouyanqing Granule based on Relevance Analysis between Chemical Characters and Biological Effects

Kouyanqing Granule (KYQG) is a traditional Chinese herbal formula composed of Flos lonicerae (FL), Radix scrophulariae (RS), Radix ophiopogonis (RO), Radix asparagi (RA), and Radix et rhizoma glycyrrhizae (RG). In contrast with the typical method of separating and then biologicalily testing the components individually, this study was designed to establish an approach in order to define the core bioactive ingredients of the anti-inflammatory effects of KYQG based on the relevance analysis between chemical characters and biological effects. Eleven KYQG samples with different ingredients were prepared by changing the ratios of the 5 herbs. Thirty-eight ingredients in KYQG were identified using Ultra-fast liquid chromatography-Diode array detector-Quadrupole-Time-of-flight-Tandem mass spectrometry (UFLC-DAD-Q-TOF-MS/MS) technology. Human oral keratinocytes (HOK) were cultured for 24 hours with 5% of Cigarette smoke extract (CSE) to induce inflammation stress. Interleukin-1β (IL-1β), interleukin-6 (IL-6), interleukin-8 (IL-8), and tumour necrosis factor-α (TNF-α) were evaluated after treatment with the eleven KYQG samples. Grey relational analysis(GRA), Pearson’s correlations (PCC), and partial least-squares (PLS) were utilized to evaluate the contribution of each ingredient. The results indicated that KYQG significantly reduced interleukin-1β, interleukin-6, interleukin-8, and tumour necrosis factor-α levels, in which lysine, γ-aminobutyric acid, chelidonic acid, tyrosine, harpagide, neochlorogenic acid, chlorogenic acid, cryptochlorogenic acid, isoquercitrin, luteolin-7-o-glucoside, 3,4-dicaffeoylquinic acid, 3,5-dicaffeoylquinic acid, angoroside C, harpagoside, cinnamic acid, and ruscogenin play a vital role.

Traditional Chinese medicine (TCM) has made eminent contributions to human health in East Asia over the past thousand years, owning the advantages that their effectiveness testified directly on numerous individuals initially in ancient times. Recently, TCM has drawn increasing interest worldwide for its complementary and alternative therapies compared with Western medicine and its remarkable ability to address various difficult diseases [1][2][3] . Herbal formulas are developed from combinations of various herbs at appropriate doses guided by TCM theory. These formulas contain multiple ingredients and affect numerous targets 4 . Bioactive ingredients in herbal formulas are responsible for the therapeutic effects and are essential to improve quality standards 5,6 . However, the bioactive ingredients in most herbal formulas are unclear. Current quality control methods for herbal formulas in China 1-5%), dexamethasone (Dex, 1-100 μ M), Niuhuangjiedu Pian (NP, 10-100 μ g/mL), KYQG of original formula (S6, 2.2-555.6 μ g/mL), and KYQG S1-S11 samples (55.6 μ g/mL) exhibited no cytotoxic activity in 24-hour cultures ( Table 3)

Bioactive ingredients on anti-inflammatory effects in KYQG.
Correlations among thirty-eight independent variables (identified ingredients) and four dependent variables (anti-inflammatory effects) were analysed to assess the contribution of each ingredient utilizing grey relational analysis (GRA), principal component analysis (PCA), and partial least-squares (PLS) methods (Table 6). Calculated results of IL-1β: Harpagide (P12), angoroside In addition, a dynamic KYQG ingredient-effect relevance bubble chart and a dynamic KYQG herb-effect relevance bubble chart based on the computed results are provided as supplementary files (KYQG ingredient-effect relevance bubble chart S1, KYQG herb-effect relevance bubble chart S2) to illustrate a clear picture of the anti-inflammatory effects of the 38 identified ingredients and 5 herbs.

Discussion
Oral mucosa keratinocytes comprise the main component of the oral mucosa epithelial cells, acting as a major barrier for oral diseases. After stimulation, oral mucosa keratinocytes produce cytokines that are involved in inflammatory processes. In addition, CSE contains a variety of harmful chemicals and causes inflammatory disorders in keratinocytes or epithelial cells, thus inducing the overexpression of pro-inflammatory cytokines, such as tumour necrosis factor-α (TNF-α ), interleukin (IL)-1, IL-6, and IL-8 [26][27][28] . Keratinocytes were treated with 5% CSE to generate an in vitro inflammation model in the present study. Some recent evidence suggests that TNF-α , IL-8, IL-6, and IL-1β are notable mediators of inflammation, playing a prominent part in the development and progression of inflammatory disorders 29 . TNF-α induces the production of inflammatory mediators (e.g., IL-1, IL-6, etc.), promoting inflammatory cell migration. IL-8 possesses diverse functions, participating in the activation of neutrophils and the chemotaxis of neutrophils, T cells, and basophils. IL-6 regulates the proliferation, differentiation, and migration of inflammatory cells. IL-1β induces the expression of adhesion molecules and enhances leukocyte migration [30][31][32] . Recurrent aphthous ulcers, oral mucositis, oral lichen planus and other oral inflammatory disorders are generally associated with parasecretion of various pro-inflammatory cytokines (e.g., TNF-α , IL-1β , IL-6, IL-8, etc.) [33][34][35][36][37] . Thus, the modulation of pro-inflammatory cytokine production is conducive to recovery from oral inflammatory disorders. The KYQG S6 (original formula)

Figure 2. UFLC-Q-TOF-MS/MS characters (A) and cluster analysis results (B) of eleven KYQG samples.
Twenty-five ingredients were confirmed by the retention time. The MS data of the reference substances and thirteen ingredients were characterized by MS data and related studies. Given the rescaled distance of 5, eleven KYQG samples could be divided into eight classes as follows: S2 and S4 belonged to one class, S3 and S5 belonged to one class, S9 and S10 belonged to one class, and the remainder of the samples constituted their own class.
results revealed that the alleviation of oral inflammatory disorders by the Chinese herbal formula KYQG is potentially related to the regulation of TNF-α , IL-8, IL-6, and IL-1β .
Mathematically, the relevance analysis of UFLC-DAD-Q-TOF-MS/MS characters and bioactive effects is an assessment of correlations between variables. Numerous mathematical statistical methodologies are available, and every method has its own theoretical basis and scope of application. Thus, the combination of several suitable methods is necessary to comprehensively and truly reflect the information from the experimental data. GRA's basic ideology is to estimate the associated degree of factors by studying the change in their similarity. The method is applicable to problems with complicated interrelationships among multiple factors and variables and is feasible in multiple attribute decision 38,39 . With the GRA method, the degree of association between each ingredient and bioactive effects is visually presented. However, multicollinearity problems can occur among the 38 original independent variables only using GRA. Thus, PCA and PLS were employed. PCA can reassemble the original variables into several mutually independent variables, and then some of the latter variables may be taken to reflect the original variables as much as possible 40 . In this study, five principal components were extracted from the original variables, contributing to 94.36% of the total variance. The scores between five principal components and the original variables were computed for their conversion (Table 7). PLS regression initiated by S. Wold and C. Albano is a novel multivariate statistical approach with the merits of overcoming the multicollinearity problem among the independent variables, discriminating between system signals and noises, and explaining dependent variables. Furthermore, the technique can integrate three analytical approaches: multiple linear regression, PCA, and typical correlation analysis. Overall, the combination of GRA, PLS, and PCA methods are helpful to reveal the relevance of ingredients and bioactive effects.
Harpagide (P12), angoroside C (P23), harpagoside (P27), and cinnamic acid (P28) are the bioactive components in RS that exert anti-inflammatory activities as evidenced by numerous in vivo and in vitro reported experiments.  Table 3. Cell viability. Data are shown as the mean ± SD, n = 3. S1-S11 represented the eleven KYQG samples. Cell viability in HOK after treatment with CSE, Dex, NP, KYQG (the original formula) and eleven KYQG samples (S1-S11) for   Harpagoside (P27) prevents IL-1β production by RAW 264.7 48 . Harpagide (P12) possesses anti-inflammatory activity in carrageenan-induced inflammation in vivo 49 . Angoroside C (P23) exhibits activity in the regulation of nitric oxide (NO) in LPS-stimulated macrophages 50 . Cinnamic acid (P28) reduces LPS-stimulated IL-6 production in macrophages. These 4 ingredients exhibit a close relevance to anti-inflammatory effects and are also the active ingredients in KYQG. Steroidal saponins are the main bioactive compounds in RO that perform anti-inflammatory, anti-aging, anti-tumour, and immunomodulatory activities. Ruscogenin (P37) is a steroidal saponin. Previous research has The control group and model group received the same volume of RPMI-1640 medium for the treatment. Each bar represents the contents of IL-1β , TNF-α , IL-8, and IL-6 as the mean ± SD, n = 3. * P < 0.05 and ** P < 0.01 vs. control group, # P < 0.05 and ## P < 0.01 vs. model group.
In addition, several saponins exerted negative effects on the following anti-inflammatory parameters: macranthoidin B (P29) and macranthoside A (P30) against IL-1β ; saponin 2 (P26) against IL-8 and TNF-α ; dipsacoside B (P31) against IL-6; and ruscogenin (P37) against IL-8). These saponins may exert activities other than the anti-inflammatory parameters detected in the present study, and the interaction among these saponins and other bioactive ingredients should also be investigated. However, the specific pharmacological mechanisms need further exploration and verification.

Methods
Preparation of samples. The KYQG samples were prepared using a five-factor, eleven-level uniform design, in which eleven KYQG samples were prepared by changing the contents of 5 herbs. Sample 6 (S6) was the original KYQG formula ratio. For each sample, the 5 herbs were decocted twice with water, and the liquids were mixed together and filtered. After concentrating the filtrate, ethanol was added. The mixture was stirred thoroughly and    . The DAD scanned from 190 nm to 400 nm. Electrospray ionization (ESI) source was operated in the positive ion mode with the following settings: ion spray voltage = 5500 V, source gas 1 = 55 psi, source gas 2 = 55 psi, ion source heater temperature = 550 °C, curtain gas setting = 35 psi, and collision gas pressure = 10 psi. Nitrogen was used as a nebulizer and auxiliary gas. The ingredients in KQYG were identified or tentatively characterized according to the mass spectrum data, related literature, and reference substances. To assess the difference among the prepared eleven KYQG samples, the relative peak areas of identified ingredients calibrated by psoralen (internal standard) were entered into SPSS 19.0 (SPSS, Inc., Chicago, IL, USA) to conduct a cluster analysis. Between-group linkage as the amalgamation rule was applied in this study, whereas Pearson's correlation was employed to calculate distance.

Evaluation of anti-inflammatory activity in vitro. Cell cultures. Human oral keratinocytes (HOK,
Guangzhou Jenniobio Biotechnology Company, Guangzhou, China), were cultured in RPMI-1640 medium supplemented with 10% foetal bovine serum, 100 U/mL penicillin, and 100 μ g/mL streptomycin. All cells were grown in a 37 °C incubator (5% CO 2 , 95% air) and passaged at approximately 90% confluence. All assays using HOK       Table 9. Dimensionless data of the 38 relative peak areas in the eleven KYQG samples. S1-S11 represented the eleven KYQG samples. P1-P38 represent 38 identified ingredients in KYQG and were regarded as the independent ingredient variables. P1-P38 were processed to be dimensionless by the equalization method for the calculation.
were performed within the 10th passage in 24-or 96-well plates. Preparation of cigarette smoke extract. CSE was prepared as described previously [58][59][60] with the following modifications. Two cigarettes were combusted to produce 300 mL smoke, which was then sucked into a syringe containing 10 mL RPMI-1640 medium. After fully dissolved, the resulting suspension was filtered through a 0.22-μ m filter, and this solution was defined as 100% CSE. Then, the desired concentrations of CSE (1%, 2%, 3%, 4%, 5%, 10%, 15%, 20%) were obtained by further diluting 100% CSE with RPMI-1640 medium within 30 min. Cell Viability Assay. HOK (5 × 10 3 cells/well) were cultured for 24 hours in 96-well plates and treated with various concentrations of CSE, Dex, NP, eleven KYQG samples for another 24 hours. Then, the MTT assay was conducted to assess the cell cytotoxicity. Cytokine measurement by ELISA. HOK (2 × 10 5 cells/well) were cultured for 24 hours in 24-well plates and then serum-starved overnight. For experiments with KYQG S6 (the original formula), the cells were divided randomly into 8 groups as follows: control, model, Dex (1, 10 μ M), NP (10 μ g/mL), and KYQG S6 (5.6, 55.6, 555.6 μ g/mL). For experiments with all the eleven samples, the cells were divided randomly into the following 15 groups: control, model, Dex (1 μ M), NP (10 μ g/mL), and KYQG S1-S11 samples (55.6 μ g/mL). Different groups were pre-incubated with drugs or RPMI-1640 medium for 1 hour followed by an additional 24 hours with 5% of CSE (tested to exhibit no cytotoxic activity by MTT assay) or the same volume of RPMI-1640 medium. TNF-α , IL-8, IL-6 and IL-1β levels in the cell culture supernatants were determined with ELISA kits.
Mathematical methods of relevance analysis. Data dimensionless. Relative peak areas of identified ingredients were regarded as independent variables (Tables 8 and 9), and values of anti-inflammatory effects were regarded as dependent variables (Tables 10 and 11). Negative effect values (the smaller the better) were converted to positive values by taking the reciprocal. All of the relative peak areas and effect values were then processed to be dimensionless by the equalization method for the relevance analysis. Grey relational analysis (GRA). GRA is a method of measuring the degree of influence between each given comparative series and the reference series in a system 61 , and this degree of influence is referred to the grey relational degree (GRD). In this study, ingredient variables were taken as the comparative series, and the effect values (TNF-α , IL-8, IL-6 and IL-1β ) were the reference series. GRD was calculated with a distinguishing coefficient of 0.5 61 Table 11. Dimensionless data of the effect parameters after switching from negative to positive. S1-S11 represented the eleven KYQG samples. Effect values of switched parameters were processed to be dimensionless by an equalization method for the calculation.
variables, which are mutual independent principal components. The regression equations between new principal components and effect values (TNF-α , IL-8, IL-6 and IL-1β ) were established with a stepwise regression method. Once a regression equation was formed (P < 0.05, IL-6 and IL-1β ), the original independent variables would replace the new principal components to establish another proper multilinear regression equation. Regression coefficients (RCs) were used to estimating the effect contribution of each ingredient. When no available equations (P > 0.05) were formed (IL-8 and TNF-α ), Pearson's correlation coefficients (PCCs) between the original independent variables and effect values were computed directly. In the present study, the regression equation between extracted components and IL-1β was used, as follows: Y IL-1β = 0.066 + 0.003 × Component4-0.001 × Componen t1 (P < 0.05). After replacing the extracted components with the original variables, we obtain the following equation: Y IL-1β = 0.066 + 3.51 × 10 −4 P1 + 6.10 × 10 −4 P2 + ……+ 7.91 × 10 −4 P38. The regression equation between extracted components and IL-6 is noted as follows: Y IL-6 = 0.401 + 0.013 × Component4+ 0.008 × Component2 (P < 0.05). After replacing the extracted components with the original variables, we obtain the following equation: Y IL-6 = 0.401 + 3.84 × 10 −3 P1 + 4.15 × 10 −3 P2+ ……+ 1.78 × 10 −3 P38. Partial least-squares (PLS) regression. PLS regression effectively settles the multicollinearity problem among variables and is suitable for situations in which the number of observations is less than the number of variables 63 . In this study, the standardized regression coefficients (RCs) in PLS regression between the independent variables and effect values (TNF-α , IL-8, IL-6 and IL-1β ) were calculated to evaluate every ingredient's effect contribution. All of the processes above were implemented in SPSS 19.0.