Modelagem molecular aliada ao aprendizado de máquina na busca por assinaturas de resistência a herbicidas em Acetolactato sintases (ALSs)

Abstract

Approximately 60% of pesticides used on plant crops are herbicides aimed at eliminating weeds. The enzyme ALS or acetohydroxy acid synthase (AHAS; EC 2.2.1.6) is targeted for inhibition by five classes of herbicides and is involved in the pathway of branched-chain amino acid biosynthesis (valine, leucine and isoleucine). Continuous exposure of these herbicides to crops has led to the evolution of weeds of herbicide resistant biotypes. These biotypes generally present target-site resistance, with point mutations being documented in several species. The development of plant crops resistant to current herbicides (contributing to only weeds being affected) and the development of new herbicides has become extremely necessary. In search of resistance signatures to two types of herbicides (sulfonylureas and imidazolinones) in ALS, here we link machine learning to molecular modeling data of enzymes with the presence of the inhibitor, with and without mutations. Molecular dynamics simulations, and other structural bioinformatics techniques were taken to attribute selection techniques used in machine learning in order to better discern attributes of ALSs that separate resistant and susceptible. The results suggest that the mechanism of gain or not of resistance to herbicides with mutations is linked to changes in both dynamics, network of contacts and energy profile in the protein-ligand complex. For the sulfonylurea (SU), the alterations in these attributes suggest a greater restoration of the competitive component of the inhibitor in the protein in relation to imidazolinone (IMI), in line with the greater inhibitory component itself reported in the literature for the SU compared to the IMI. In enzymes with the imidazolinone inhibitor, resistance seems to have a direct relationship with allosteric modifications, structurally modifying the cofactors region. In enzymes with the sulfonylurea inhibitor, the resistance pattern suggests a strong relationship with the loss of affinity for the ligand. The results obtained here can contribute to the elucidation of new paths for the sustainable theme of weeds, crops and herbicides. These same results also demonstrate that using machine learning to find patterns amidst a diversity of data returned by modeling and molecular dynamics is a readily applicable and effective strategy