Utilization of AlphaFold models for drug discovery: Feasibility and challenges. Histone deacetylase 11 as a case study

by Selleckchem | Jan 04 2024

Histone deacetylase 11 (HDAC11), an enzyme that cleaves acyl groups from acylated lysine residues, is the sole member of class IV of HDAC family with no reported crystal structure so far. The catalytic domain of HDAC11 shares low sequence identity with other HDAC isoforms which complicates the conventional template-based homology modeling. AlphaFold is a neural network machine learning approach for predicting the 3D structures of proteins with atomic accuracy even in absence of similar structures. However, the structures predicted by AlphaFold are missing small molecules as ligands and cofactors. In our study, we first optimized the HDAC11 AlphaFold model by adding the catalytic zinc ion followed by assessment of the usability of the model by docking of the selective inhibitor FT895. Minimization of the optimized model in presence of transplanted inhibitors, which have been described as HDAC11 inhibitors, was performed. Four complexes were generated and proved to be stable using three replicas of 50 ns MD simulations and were successfully utilized for docking of the selective inhibitors FT895, MIR002 and SIS17. For SIS17, The most reasonable pose was selected based on structural comparison between HDAC6, HDAC8 and the HDAC11 optimized AlphaFold model. The manually optimized HDAC11 model is thus able to explain the binding behavior of known HDAC11 inhibitors and can be used for further structure-based optimization.

Comments:

That's an impressive study! Your approach to optimizing the HDAC11 AlphaFold model by incorporating the catalytic zinc ion and then assessing its usability by docking selective inhibitors like FT895, MIR002, and SIS17 is quite thorough.

The fact that HDAC11 lacks a reported crystal structure and has low sequence identity with other HDAC isoforms makes it a challenging protein to model conventionally. Utilizing AlphaFold for structure prediction despite these challenges showcases the power of machine learning-based approaches in predicting protein structures accurately, even in cases where template-based homology modeling might struggle.

Your methodology of verifying the stability of the generated complexes through multiple replicas of molecular dynamics simulations and then using these structures for docking known HDAC11 inhibitors like FT895, MIR002, and SIS17 is a comprehensive way to validate the model's utility in explaining binding behaviors. Additionally, the structural comparison between HDAC6, HDAC8, and the manually optimized HDAC11 model to select the most reasonable pose for SIS17 demonstrates a thoughtful approach to understanding inhibitor binding specificity.

The successful explanation of the binding behavior of known HDAC11 inhibitors using your optimized model paves the way for further structure-based optimization and rational design of new HDAC11 inhibitors. This work could be instrumental in drug discovery efforts targeting HDAC11.