Molecular Dynamics Simulations Identify Tractable Lead-like Phenyl-Piperazine Scaffolds as eIF4A1 ATP-competitive Inhibitors

Abstract

eIF4A1 is an ATP-dependent RNA helicase whose overexpression and activity have been tightly linked to oncogenesis in a number of malignancies. An understanding of the complex kinetics and conformational changes of this translational enzyme is necessary to map out all targetable binding sites and develop novel, chemically tractable inhibitors. We herein present a comprehensive quantitative analysis of eIF4A1 conformational changes using protein-ligand docking, homology modeling, and extended molecular dynamics simulations. Through this, we report the discovery of a novel, biochemically active phenyl-piperazine pharmacophore, which is predicted to target the ATP-binding site and may serve as the starting point for medicinal chemistry optimization efforts. This is the first such report of an ATP-competitive inhibitor for eiF4A1, which is predicted to bind in the nucleotide cleft. Our novel interdisciplinary pipeline serves as a framework for future drug discovery efforts for targeting eiF4A1 and other proteins with complex kinetics.

Figure 1

1FUU eIF4A1 homology model. (A) ATP is docked in the open state of the 1FUU homology model. ATP initially binds in the N-terminal domain with contacts between the adenine moiety and motifs Q and I. The interdomain distance (Pro231 to Ile246) in the open conformation is measured to be 39.26 angstroms. (B) 2D representation of ATP docked in the N-terminal domain of the open-conformation eiF4A1 model with hydrogen bonds to a conserved glutamine residue and several residues in the interdomain linker. (C) After an extended molecular dynamic simulation, the 1FUU homology model adopts a closed conformation, with the interdomain distance between Pro231 and Asp247 reduced to 27.15 Å. (D) ATP retains hydrogen bonds to the conserved glutamine residue in the N-terminal domain; however, terminal phosphates of the ATP extend to the C-terminal domain with key interactions occurring between arginine residues in motif IV.

Figure 2

Interdomain distance during MD simulations of the 1FUU model. Extended MD simulations of the 1FUU homology model were run for 3 ms in both the presence and absence of ATP. With ATP bound in the N-terminal domain, interdomain closure is quickly observed within the first 10 ns. The system equilibrates to an energy minimum at 60 ns with domain closure and is sustained for the entirety of the 3 ms simulation, the system converging at an interdomain distance of around 27 Å. Without ATP, the system remains open with an interdomain distance around 35 Å.

Figure 3

2J0S homology model analysis. ATP was removed from the active site of the 2J0S homology model and a molecular dynamics simulation was run for 1 ms. The protein quickly adopted an open conformation with an increase in interdomain distance from 22 to 27 Å. Docking ATP back into the model in the final frame and running the simulation for an additional 1 ms, the system closes again and converges at an energy minimum with an average interdomain distance of 23 Å.

Figure 4

Distribution of HYBRID scores by the small-molecule database. Top 25% of scores were selected for output so there is a sharp right-sided decline for each curve. Two visible peaks in blue and red represent 2J0S and 1FUU, with HYBRID from the 2J0S models trending toward better, more negative docking scores. For reference, ATP scores on average −7 in the 1FUU model, while it scores on average −9 in the 2J0S model. On average, compounds filtered from the eMolecules Blockbuster library scored better, likely because they are larger and more drug-like versus smaller lead-like compounds in the other libraries.

Figure 5

2D representation of phenyl-piperazine hit compounds: (A) UM127, (B) UM139, (C) UM162, and (D) UM167.

Figure 6

Molecular dynamics analysis of top hits. Interdomain distance analysis of UM127 in the 1FUU (A) and 2J0S (B). UM127 causes large conformational changes in both models, which leads to partial domain closure. (C) The model predicts that UM127 promotes intralinker contacts. UM139 in the 1FUU (D) and 2J0S (E) leads to full domain closure in the 1FUU model and maintains domain closure in the 2J0S model. (F) Binding post of UM139 shows similar contacts seen with ATP including bonding with Arg365 and Arg368.

Figure 7

Molecular dynamics analysis of UM167. UM167 was the most potent hit identified in the biochemical screen. MD simulation in the 1FUU model (A) shows rapid domain closure with sustained interactions with Arg368, conserved glutamine Gln60, and a glutamic acid residue Glu244 in the interdomain linker (B). (C) UM167 also maintains domain closure in the 2J0S model. (D) Binding poses reveal that the UM167 binding leads to domain closure with multiple interdomain and protein–ligand contacts.

Figure 8

Binding hypothesis of UM167 in the human eIF4A1 model (5ZC9). UM167 binds in the interdomain cleft of eIF4A1 with key interactions with conserved glutamines Gln59 and Gln77, active site Lys81, and motif VI Arg362 (A, B). Molecular dynamic simulations in the 5ZC9 model were stable and UM167 locked the model in the closed conformation with RNA also remaining in the interdomain cleft.