Development And Application Of New Fragment-Based Pesiticde Design Approaches

Pesticide is an important mean of production to ensure food security.It is reported that the crop loss in China can be reduced about 100 billion kilograms under the usage of pesticides.Traditional pesticides often meet the problems of low efficiency and high environmental risks in use.Therefore,the development of more potent green pesticides is an inevitable trend in agricultural of China.Fragment based drug discovery（FBDD）strategy has shown significant advantages in speeding up the pesticide discovery.FBDD strategy uses fragment molecules with molecular weight less than 300 as the starting structure for the discovery of pesticide lead compounds,and exhibits a series of advantages,such as high hit rate,good molecular novelty,and promising physicochemical properties.However,because fragments exhibit weak binding affinity to target protein,how to accurately and efficiently obtain the interaction information between protein and fragment is a key scientific issue in fragment-based pesticide discovery.In this study,we developed a series of new computational approaches,involving fragment library generaion,protein-fragment binding affinity calculation and fragment optimization,to improve the accuracy and efficiency of protein-fragment interaction prediction,and therefore accelerate the discovery of pesticide lead compounds.Understanding protein-fragment interaction is of great significance to the construction of fragment libraries and the hit rate of fragment screening.In Chapter 2,we collected the reported crystal structures of protein-fragment complexes from the PDB database,systematically analyzed the protein-fragment interaction,and summarized the structural and physicochemical property of fragment hits as well as the characteristics of protein-fragment interaction.The results showed that a fragment hit often has a ring structure and is polar.Their properties meet the requirements of heavy atom number between 9～21,hydrogen bond acceptor≤4,hydrogen bond donor≤3,Log P between-2-4,ring number between1～3,rotatable bond≤5,TPSA between 40-120?².In addition,it was found that the fragment hits interacted with the target protein mainly through hydrogen bonds and other polar interactions.They preferentially bound to the vicinity of charged amino acid residues,and showed huge buried surface area.The activity of fragments is greatly affected by their mode of action and the type of target.A specific fragment library should be established to improve the hit rate of screening.The prediction of protein-fragment binding affinity plays an important role in fragment virtual screening and binding mode prediction.In Chapter 3,we used the analyzed protein-fragment interaction features to build artificial intelligence prediction models.A series of machine learning and deep learning models are generated,such as random forest,support vector machine,deep neural network and convolutional neural network.Among them,the convolution neural network model showed the best prediction ability.In the evaluation of protein-fragment binding affinity,it exhibited a Pearson correlation coefficient R value of 0.66 and a standard deviation SD value of 1.70,surpassing the commonly used molecular docking software Autodock vina and fragment docking software SEED.Our scoring function also showed strong ranking ability on different drugs or pesticide that bind to the same target,with a Spearman correlation coefficient SP value of 0.63,a Kendall correlation coefficientvalue of 0.54,and a prediction index PI value of 0.61.These results indicated that our model had a strong predictive ability for protein-fragment binding affinity.Fragment optimization is an important step in the obtain of lead compounds,and protein flexible pose a major challenge.In Chapter 4,we constructed a fragment dynamic growing workflow for the o ptimization of fragment hit to lead compounds.Using the fragment rules obtained from the analysis re sults,we constructed a series of fragment libraries derived from commonly used drug,pesticide,and k inase inhibitor and hot-spot binding fragments.We developed a fragment dynamic growing strategy th at protein flexibility can be fully considered by using molecular dynamic simulations sampling.This m ethod exhibits high accuracy（88.5%）in discovering more potent compounds with the same scaffold.Using this strategy,we successfully discovered a TRK kinase inhibitor YT9.It showed a great activity improvement(IC₅₀^WT=2.7nM,IC₅₀^G595R=0.94nM,IC₅₀^G667C=1.4nM,IC₅₀^F589L=0.44nM)than approved drug larotrectinib(IC₅₀^WT=9.4nM,IC₅₀^G595R=65.4nM,IC₅₀^G667C=44.3nM,IC₅₀^F589L=31.1nM),and can overcome multiple drug resistance induced by mutations.Not only that,we have also built the ACFIS2 web server to perform fragment-based drug discovery for users freely at http://chemyang.ccnu.edu.c n/ccb/server/ACFIS2.Abscisic acid（ABA）is an important endogenous hormone in plants,and dimeric members of ABA receptor PYL family proteins play a key role in seed germination.However,the relationship between the stability of PYL dimers and seed germination remains unclear.In Chapter 5,we designed a dimer stabilizer DBSA targeting the binding pocket at the PYL dimer interface using fragment scoring function and fragment dynamic growth strategy developed by us.Compared to ABA,DBSA showed a huge receptor affinity increased by about ten-fold.It can reverse ABA-induced seed germination inhibition,and it did not induce the expression of ABA-activating genes.X-ray crystallography showed that DBSA targets a new pocket in the PYL dimer interface.SEC-MALLS results suggested that DBSA showed no effect on the dimeric state of PYR1.Molecular dynamic simulations revealed that DBSA stabilize PYL dimers through hydrogen bonding networks.Our results demonstrated the great potential of PYL dimer stabilization in preventing ABA induced seed germination inhibition.