P01-05

Prediction Method for Protein-Bound Conformation of Macrocycles

Shoya HAMAUE *, Isao YASUMATSU, Ayako MORITOMO, Mizuki TAKAHASHI, Hiroyuki HANZAWA

Modality Research Laboratories I, Research Function, R&D Division, Daiichi Sankyo Co., Ltd. 
( * E-mail: syoya.hamaue@daiichisankyo.com )

In recent years, so-called “Beyond Rule of 5 (bRo5)” modalities have been recognized in the pharmaceutical industry to possess new therapeutic potential. Examples of synthetic bRo5 modalities are peptides, nucleic acids, macrocycles, and heterobifunctional molecules such as Targeted Protein Degraders (TPDs). These modalities offer advantages in addressing medical needs that were difficult to achieve with small molecules, such as targeting Protein-Protein Interactions (PPIs) and achieving high target specificity. However, due to their large molecular weight and complexity, it is generally considered challenging to accurately predict their 3D structures, biochemical activities, and physicochemical properties using computational chemistry methods. Therefore, it is necessary to devise new computational methods to overcome these challenges and improve prediction accuracy.
In this study, we selected macrocycles as an example of bRo5 modalities. Macrocycles can adopt a vast number of conformations in solution, and thus computational methods for accurate prediction of the protein-bound conformations of macrocycles have not yet been fully established. Then, we aimed to establish a versatile computational approach for predicting their protein-bound conformation. To accomplish this objective, we evaluated the computational methods that can reproduce the bound conformation of crystal structures in a variety of cases. We curated 51 complex crystal structures from the Protein Data Bank (PDB), considering the diversity in the ring size of the macrocycles, the type of targets, and the pocket shapes. To thoroughly explore the extensive range of possible conformations of the macrocycles, we examined the conformational search conditions for it alone using three different modeling tools. As a result, we found it effective in predicting protein-bound conformations to explore expansion of the conformational space by using multiple modeling tools. Additionally, we established potential energy criteria to exclude unstable conformations from the explored conformational ensemble. Finally, we performed docking studies on the conformations (11~4,223) that are filtered from generated conformations (33~7,158) to verify whether they could bind to the target proteins with the predicted conformations.
As an outcome of this study, we successfully identified a set of conditions that accurately reproduced the protein-bound conformations for 46 out of 51 complexes. This methodology is expected to enable the prediction of protein-bound conformations of macrocycles without any previous knowledge of their bound form, regardless of ring size, target categories, and pocket shapes. In the future, we would like to investigate whether this methodology can be applied to other bRo5 modalities.