Method for screening anti-inflammatory compounds by combining quantitative structure-function relationship and molecular action relationship

Abstract

The invention relates to the technical field of functional ingredient development, in particular to an anti-inflammatory medicinal and edible food capable of inhibiting cyclooxygenase-2 (COX-2) activity, and the anti-inflammatory medicinal and edible food can effectively improve the inflammatory condition. Potential anti-inflammatory active ingredients in food are screened out by constructing a COX-2 inhibitor database and applying a quantitative structure-activity relationship (QSAR) model and a molecular docking technology. Through in-vitro experimental verification and molecular dynamics simulation, a COX-2 inhibitor is successfully identified as an anti-inflammatory candidate drug. The research results provide a scientific basis for the development of anti-inflammatory drugs and functional foods.

Description

Technical Field

[0001] The invention belongs to the technical field of functional ingredient development, and specifically relates to a method for screening cyclooxygenase-2 (COX-2) inhibitors for anti-inflammatory purposes. Background Art

[0002] According to the Global Burden of Disease Study, inflammatory diseases occupy an important position in global health losses. Inflammatory diseases have become one of the major threats to the health of the global population. With the acceleration of urbanization and aging and changes in lifestyles, major chronic non-communicable diseases including infectious diseases, autoimmune diseases, malignant tumors, cardiovascular diseases, diabetes, etc. all fall into the category of inflammatory diseases. These diseases are spreading around the world, seriously endangering human health and sustainable socio-economic development. Studies have shown that malignant tumor tissues contain high concentrations of inflammatory cells, and tumors usually form at sites of chronic inflammation. Nonsteroidal anti-inflammatory drugs are the most widely used drugs for controlling inflammation, but they are accompanied by side effects such as gastrointestinal discomfort.

[0003] In recent years, the emergence of selective COX-2 inhibitors has brought convenience to people, which can not only effectively control the continuation of inflammation, but also reduce the generation of gastrointestinal side effects. COX-2 inhibitors reduce the production of prostaglandins G2 and H2 by inhibiting the activity of arachidonic acid, thereby mediating the occurrence of inflammatory reactions. While the more effective selective cyclooxygenase-2 inhibitors on the market, such as celecoxib, meloxicam, and nimesulide, these drugs more or less also produce some side effects. Therefore, if a fast and efficient screening method can be used to screen out natural, healthy and effective anti-inflammatory substances from food, this will be of great help to the regulation and treatment of inflammation, pain, etc. Summary of the invention

[0004] The purpose of the present invention is to establish a COX-2 inhibitor molecular database to combine the QSAR model with molecular docking to screen potential COX-2 inhibitor candidates, and to obtain a COX-2 inhibitor for anti-inflammatory purposes through in vitro experimental verification.

[0005] The product and preparation method of the present invention are as follows:

[0006] A COX-2 inhibitor for anti-inflammatory purposes, characterized in that it comprises the following steps: (1) establishing a database of COX-2 inhibitors with medicinal and edible properties, the database comprising 135 compounds that inhibit COX-2; (2) using four descriptor pools including MOE, Mordred, ChemoPy and Rdkit, calculating the canonical SMILES of (1) molecules to generate molecular descriptors for capturing molecular structures and properties and comparing them; (3) using three machine learning methods including random forest (RF), support vector machine (SVM) and linear discriminant analysis (LDA), establishing a qualitative structure-activity relationship (QSAR) model capable of screening COX-2 inhibitors based on the descriptors generated by (2), and selecting the optimal model by comparing the accuracy of the models; (4) using celecoxib as a positive control, obtaining the crystal structure of COX-2 (PDBID: 3LN1) from the PDB database (http: / / www.rcsb.org / ), and using it as a protein target, and using MOE to identify the COX-2 inhibitors. In 2015, molecular docking was performed to verify the results of model prediction. Gromacs was used to perform 100 ns molecular dynamics simulations on COX-2 and the screened compounds to verify the binding stability. The COX-2 inhibitory activity of the compounds was further determined by fluorescence using a COX-2 kit; (5) Potential COX-2 inhibitors were discovered through model screening and experimental verification.

[0007] The RDkit descriptor has the best modeling effect. RDKit calculates three molecular fingerprints to characterize the molecular structure of the compound: MACCS fingerprint (166 bits), ECFP4 fingerprint (1024 bits) and Avalon fingerprint (1024 bits), and the prediction is more robust.

[0008] The best model is the RF model with an accuracy of 0.9411

[0009] The experimental conditions are as follows: first, prepare the working solution and configure each working solution according to the corresponding proportion, including COX-2Assay Buffer, COX-2Cofactor working solution, COX-2 working solution and COX-2Substrate working solution. Add 75μL COX-2Assay Buffer, 5μL COX-2Cofactor working solution and 5μL working solution (except the blank group) to the 96-well black plate, mix well and add 5μL of the sample to be tested or the positive control (celecoxib), add 5μL of sample solution to the blank control group and the 100% enzyme group, mix well and incubate at 37°C for 10 minutes. Then, add 5μL COX-2Probe and 5μL COX-2Substrate to start the reaction. Finally, after incubation at 37°C for 5-20 minutes, read the relative fluorescence unit (RFU) using an excitation wavelength of 560nm and an emission wavelength of 590nm. Each group of samples is tested in 3 groups in duplicate. The calculation formula of COX-2 inhibition rate is as follows:

[0010]

[0011] Wherein, RFU100% enzyme activity control, RFU blank control, and RFU sample are 100% enzyme activity control fluorescence unit, blank control fluorescence unit, and sample fluorescence unit, respectively.

[0012] The isorientin, resveratrol, caffeic acid and cyanidin have COX-2 inhibitory activity and can be used as COX-2 inhibitory candidates. 50 They are 42.07μM, 19.73μM, 295.6μM, and 15.62μM. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a process flow chart of the present invention;

[0014] Figure 2 is the importance map of the RDkit descriptor;

[0015] Figure 3 is the accuracy plot of the model;

[0016] Figure 4 is the molecular docking diagram;

[0017] Figure 5 This is a molecular dynamics simulation diagram. DETAILED DESCRIPTION

[0018] The specific embodiments are further described below in conjunction with the accompanying drawings.

[0019] Example 1

[0020] (1) A database of 135 COX-2 inhibitors with medicinal and edible properties was established by searching the literature and databases, and the canonical SMILES of these compounds were collected.

[0021] (2) The canonical SMILES of the compounds collected in step (1) were calculated using MOE and Python software to generate MOE, ChemoPy, Mordred, and RDkit descriptors. The nearZeroVar function of the caret package in the R language can effectively remove variables with extremely small or close to zero variance. The findCorrelation function helps to identify and exclude highly correlated variables. The variable screening and optimization are achieved by adjusting the correlation coefficient cutoff value, and the correlation coefficient of the descriptor is removed. Figure 2 After optimization, the 20 descriptors with the highest contribution rate are retained and principal component analysis is performed on them to achieve the purpose of dimensionality reduction.

[0022] (3) Based on the descriptors generated in (2), the training set and the test set are divided into a ratio of 75%:25%. The training set is fitted with three models: RF, SVM, and LDA and trained. The test set is used to evaluate the generalization ability of the final model and compare its accuracy. Figure 3 It can be seen that after calculation, the RF model of Rdkit has the highest accuracy, which is 0.9411.

[0023] (4) Further based on the IC values ​​of 135 COX-2 inhibitors 50 The quantitative model was established using four algorithms: SVM, RF, PLS, and XGBoot regression, and the R of the test set was calculated. 2 The R value of the XGBoost regression algorithm is calculated based on the root mean square error and the predictive ability of the model. 2 and RMSE performed the best, which were 0.9582 and 0.01823 respectively.

[0024] (5) Using celecoxib as a positive control, the crystal structure of COX-2 (PDBID: 3LN1) was obtained from the PDB database (http: / / www.rcsb.org / ). All water molecules and ions were removed from the retrieved crystal structure using MOE 2015 software, and the structure was prepared using the Quickprep function. Then, the energy of the small molecule ligand was minimized (the parameters were set to Gradient: 0.1, force field: Amber99). Figure 4 The results of molecular docking based on the active sites showed that isorientin, cyanidin, resveratrol, and caffeic acid were screened out.

[0025] (6) The compounds screened out according to (5) were subjected to fluorescence measurement at an emission wavelength of 560 nm and an emission wavelength of 590 nm to determine the COX-2 inhibitory activity of these compounds.

[0026] (7) The best candidate for COX-2 inhibitory activity was found to be cyanidin, IC 50 It is 15.62μM.

[0027] (8) GROMACS was used to perform molecular dynamics simulations on the protein structure and the compound with the best inhibitory activity, cyanidin. The receptor was dehydrated and hydrogenated to calculate its force field to create a topology file. The force field of the ligand was calculated separately. At the same time, a simulated water box was constructed, solvent was added, balanced charges were added, energy was minimized, and isothermal, isochoric, and isothermal and isobaric equilibrium was achieved before starting a 100 ns dynamics simulation. Figure 5 It can be seen that after dynamic simulation, the receptor and ligand are stably bound, and the root mean square deviation (RMSD) is less than 0.1nm

[0028] Example 2

[0029] (1) An angiotensin-converting enzyme database was established by reviewing literature and Pubchem. The constructed database includes 101 inhibitors and 101 non-inhibitors. By reviewing their SMILES and generating four descriptors, MOE, Mordred, Chemopy, and Rdkit, a comparative analysis of the descriptor pool was performed.

[0030] (2) The descriptors were preprocessed using the nearZeroVar function, findCorrelation function, and PCA analysis in the R language. The nearZeroVar function in the caret package was used to filter out descriptors with small changes, and the findCorrelation function was used to delete descriptors with large correlations. In order to reduce the dimensionality of the data, PCA analysis was used to eliminate molecular descriptors with low contribution rates.

[0031] (3) Based on the descriptors generated in (2), the training set and test set were divided into 3:1, and ten-fold cross validation was performed. The training set was fitted with four models: RF, SVM, LDA, and kNN, and trained. The test set was used to evaluate the generalization ability of the final model. The model was run 1000 times, the average accuracy was calculated, and the model with the highest average accuracy was selected. The calculation showed that the average accuracy of the model established by the Rdkit descriptor using the RF algorithm was 0.9540±0.0299, which performed best.

[0032] (4) Based on the model obtained in (3), eight food-borne compounds, including apigenin, caffeic acid, catechin, rosmarinic acid, quercetin, chlorogenic acid, gallic acid, and ferulic acid, were predicted and screened in FOODB. Using lisinopril as a positive control, the crystal structure of angiotensin-converting enzyme (PDB: 1O8A) was obtained from the PDB database (http: / / www.rcsb.org / ) and used as a protein target. Molecular docking was performed by MOE to verify the results of the model prediction.

[0033] (5) To further verify it, fluorescence was measured at an excitation wavelength of 340 nm and an emission wavelength of 455 nm using an ELISA reader to obtain the angiotensin converting enzyme inhibitory activity. The eight compounds were found to inhibit angiotensin converting enzyme to varying degrees. Among them, chlorogenic acid was the compound with the best inhibitory effect in the in vitro enzyme inhibition experiment, with an IC of 50 The value is 0.12mM.

[0034] (6) GROMACS software was used to perform molecular dynamics simulation of chlorogenic acid, the compound with the best inhibitory effect. Acpype was used to apply the GAFF2 force field to the ligand, and the addition of solvent, charge balance, energy minimization, isothermal and isovolumetric equilibrium, isothermal and isobaric equilibrium, etc. were performed on it. A simulation was then performed for 10 ns, within which the compound was in a stable binding state.

[0035] Example 3

[0036] (1) Based on literature review, a dataset of edible and medicinal adenosine deaminase inhibitors was established, which included 100 adenosine deaminase inhibitors and 100 non-inhibitor compounds. The Canonical SMILES of the compounds were searched in Pubchem, and MOE, Rdkit, Mordred, and Chemopy descriptors were generated for them using MOE and Python software.

[0037] (2) The descriptors generated in (1) were optimized, and the correlation of the descriptors was calculated using the Pearson correlation coefficient test method. The descriptors with correlation coefficients greater than 0.96 and those with low contribution rates were deleted. Finally, the 20 descriptors with the highest contribution rates were retained and principal component analysis was performed on them.

[0038] (3) Based on the optimal descriptor generated by (2), Chemopy constructed three qualitative structure-activity models (QSAR) for screening adenosine deaminase inhibitors: RF, SVM, and LDA. The collected data set was divided into training set and test set at a ratio of 75%:25%. In order to avoid the contingency between the training set and the test set, the model was repeatedly trained 10,000 times, and all the test sets and training sets were randomly assigned. The accuracy of each model was evaluated by comparing the corresponding area under the ROC curve. It was found that the RF model had the best model fitting effect, and its ROC curve area was 0.9648, so Chemopy's RF model performed best.

[0039] (4) Based on the optimal model screened out in (3), foodborne compounds with adenosine deaminase inhibitory effect were screened and predicted from the FOODB database. The crystal structure of adenosine deaminase (PDB: 1VFL) was downloaded from the PDB database (http: / / www.rcsb.org / ). The receptor was dehydrated, hydrogenated and energy minimized in MOE. Molecular docking was performed based on morin, sedge biflavonoids, kaempferol, rosmarinic acid, vitexin and the positive control pentostatin screened out from FOODB. The docking free energies of the six substances with adenosine deaminase were all less than -7.5 KJ / mol, showing a good binding effect.

[0040] (5) To further verify it, the content of inosine was determined by high performance liquid chromatography, and its chromatographic value was recorded at 248nm. 70% ethanol was used as the control group to obtain the adenosine deaminase inhibitory activity. All six compounds were able to inhibit the production of inosine to varying degrees. Among them, the compound with the best inhibitory effect in the in vitro enzyme inhibition experiment was aconitum biflavonoids, with an IC 50 The value is 12.87 μM.

[0041] (6) GROMACS software was used to perform molecular dynamics simulation of the compound with the best inhibitory effect, aconitum biflavonoids. Acpype was used to apply the GAFF2 force field to the ligand, and the addition of solvent, charge balance, energy minimization, isothermal and isobaric equilibrium, etc. were performed on it. A simulation was then performed for 10 ns, within which the compound was in a stable binding state.

Claims

1. A cyclooxygenase-2 (COX-2) inhibitor for anti-inflammatory purposes, characterized in that: The following steps are involved: (1) A database of COX-2 inhibitors with medicinal and edible properties was established, including 135 compounds that inhibit COX-2. (2) Four descriptor pools, including MOE, Mordred, ChemoPy, and Rdkit, were used to calculate (1) the canonical SMILES of molecules to generate molecular descriptors for capturing molecular structure and properties and compare them. (3) Four machine learning methods, including random forest (RF), support vector machine (SVM), and linear discriminant analysis (LDA), were used to establish qualitative structure-activity relationship (QSAR) models for screening COX-2 inhibitors based on the descriptors generated by (2), and the accuracy of the models was compared to select the optimal model. (4) Celecoxib was used as a positive control, and the crystal structure of COX-2 (PDB ID: 3LN1) was obtained from the PDB database (http: / / www.rcsb.org / ), and it was used as a protein target. MOE was used to identify COX-2 inhibitors. In 2015, molecular docking was performed to verify the results of model prediction. Gromacs was used to perform 100 ns molecular dynamics simulations on COX-2 and the screened compounds to verify the binding stability. The COX-2 inhibitory activity of the compounds was further determined by fluorescence. (5) Potential COX-2 inhibitors were discovered through model screening and experimental verification.

2. A COX-2 inhibitor for anti-inflammatory purposes according to claim 1, characterized in that: The RDkit descriptor has the best modeling effect. RDKit calculates three molecular fingerprints to characterize the molecular structure of the compound: MACCS fingerprint (166 bits), ECFP4 fingerprint (1024 bits) and Avalon fingerprint (1024 bits), and the prediction is more robust.

3. A COX-2 inhibitor for anti-inflammatory purposes according to claim 1, characterized in that: The best model is the RF model with an accuracy of 0.9411.

4. A COX-2 inhibitor for anti-inflammatory according to claim 1, wherein the experimental conditions are as follows: first, a working solution is prepared and each working solution is configured according to a corresponding proportion, including COX-2Assay Buffer, COX-2Cofactor working solution, COX-2 working solution and COX-2Substrate working solution; 75 μL COX-2AssayBuffer, 5 μL COX-2Cofactor working solution and 5 μL working solution (except the blank group) are added to a 96-well black plate, and after mixing evenly, 5 μL of the sample to be tested or the positive control (celecoxib) is added, and 5 μL of the sample solution is added to the blank control group and the 100% enzyme group, and the mixture is fully mixed and incubated at 37°C for 10 minutes. Then, 5 μL COX-2 Probe and 5 μL COX-2 Substrate were added to start the reaction; after incubation at 37°C for 5-20 minutes, the relative fluorescence unit (RFU) was read using an excitation wavelength of 560 nm and an emission wavelength of 590 nm. Each group of samples was tested in duplicate for 3 times, and the COX-2 inhibition rate was calculated as follows: Wherein, RFU100% enzyme activity control, RFU blank control, and RFU sample are 100% enzyme activity control fluorescence unit, blank control fluorescence unit, and sample fluorescence unit, respectively.

5. A COX-2 inhibitor for anti-inflammatory purposes according to claim 1, characterized in that: The candidate substances with COX-2 inhibitory activity are isorientin, cyanidin, resveratrol, and caffeic acid, wherein IC 50 They are 42.07μM, 19.73μM, 295.6μM, and 15.62μM.

Images