Antitumor Active Ingredients Identification From Zanthoxylum Essential Oil Based On Composition-activity Relationship

Pepper is a kind of medicinal and edible plant and mainly contains kinds ofvolatile oil. It has been reported that the volatile oil of pepper has antitumor activities.In this study, we analyzed the components of Zanthoxylum bungeanum volatile oiland measured its anti-tumor activity. The composition-activity relationship (CAR)models between components and anti-tumor activity were established to predict theanti-tumor activity according to the components. Based on the CAR models, theactive components were identified to benefit the new drug discovery. The main workis as follows:1. Volatile oil was extracted from Zanthoxylum bungeanum using steamdistillation. The extraction technology was optimized based on response surfacemethodology. The optimum condition was determined as follows: adding8timesamount of water, soaking for2.5h, extracting for4.7h. Under these conditions,39batches of volatile oil were extracted with yields ranging from0.54%to4.52%(g/g,volatile oil). The yields were different from different batches of volatile oil.2. GC-MS was used to analyze the components of39batches of volatile oil.And23characteristic peaks were analyzed qualitatively and quantified by the internalstandard method (internal standard substance, n-tridecane). The inhibitory activity ofvolatile oil against Hela cells were measured by the MTT assay with the inhibitionratios ranging from0.457to0.839. It showed that there are obvious differences incomponent content and anti-tumor activity of different bathes of volatile oil.3. BP neural network (BPNN), support vector regression machine (SVR) andgeneral regression neural network (GRNN) were used to establish the CAR models inwhich the input variable is relative area of23characteristic peaks and the outputvariable is the inhibition ratio of39batches of volatile oil against Hela cells. Themodel parameters were optimized by k-fold cross validation, grid search (GS), geneticalgorithm (GA) and particle swarm optimization (PSO) respectively. The training andprediction accuracies of the models were evaluated by root mean square error(RMSE), relative standard error (RSE) and correlation coefficient (R). Finally, GRNNwas the super model whose RMSE, RSE and R of test data were0.013,0.022and 0.974, respectively, followed by PSO-SVR whose RMSE, RSE and R of test datawere0.029,0.049and0.967, respectively.4.9potential significant anti-tumor active components were identified by thesuper models, GRNN and PSO-SVR, combined with mean impact value (MIV).These components are Limonene, α-Terpineol, γ-Terpinene,(-)-4-Terpineol,β-Linalool, β-Myrcene, Eucalyptol, Sabinene and trans-β-Ocimene of which5components have been identified with anti-tumor activities and the other componentsare the main components of several traditional Chinese medicine with anti-tumoractivities. It indicated that the method could efficiently identify active componentsfrom traditional Chinese medicines.