Computational Identification of Tractable Drug-like Inhibitors of eIF4A1

Targeted signaling inhibitors for hematologic malignancies often lead to limited clinical efficacy due to the outgrowth of subpopulations with alternative pathways independent of the drug target. The eIF4F complex responsible for translation initiation is a convergence point for many cancer-promoting signaling pathways and its inhibition leads to decreased expression of key oncoproteins and apoptosis. Lymphomas and leukemias show particular dependence on constitutive eIF4F activation. Indeed, natural compounds targeting the eIF4F enzymatic component, eIF4A1, demonstrate activities both in vitro and in vivo against lymphoma and leukemia model systems, among other tumor types. eIF4A1 is a noteworthy target for hematologic malignancies based on the finding that BCR stimulation leads to increased mRNA translation primary CLL patient samples. Additionally, eIF4F components eIF4A1 and eIF4G1 had increased expression upon IgM-induced BCR activation. The natural compound silvestrol is a potent inhibitor of eIF4A1, results in cancer cell cytotoxicity, and has an established therapeutic window in vivo. Silvestrol shows potent antitumor activity against 924 pan-cancer tumor cell lines with 830/924 (90%) sensitive at IC50 <100nM with lymphoma and leukemia cell lines being particularly sensitive. Silvestrol and other natural compounds, however, lack core drug-like properties and synthetic tractability. To discover new, specific and tractable inhibitors of eIF4A1 that are more drug-like, we have constructed several molecular models that we used to virtually screen more than 20 million compounds. eIF4A1 is the founding member of the DEAD-box RNA helicases, which include its paralogs eIF4A2 (91% amino-acid identity with eIF4A1) and eIF4A3 (60% identity). All DEAD-box helicases contain two RecA-like domains separated by a flexible linker. The cleft between these domains is lined with helicase motifs that mediate nucleotide binding and hydrolysis. In an absence of RNA or nucleotide, eIF4A proteins adopt diffuse open conformations; binding of RNA and ATP triggers transition to a more stable closed state. Modeling small-molecule interactions in the nucleotide cleft of eIF4A1 therefore assesses ability of molecules to lock eIF4A1 in a conformation unable to cycle through ATPase and helicase activities. Although no experimentally derived structure of human eIF4A1 co-crystalized with ATP exists, crystal structures of other DEAD-box family members with similar motifs permit detailed studies of nucleotide and ligand-binding and the development of homology models. We have used four available high-resolution crystal structures to build models predicting interactions of small molecules in the interdomain nucleotide-binding cleft. We identified nucleotide binding-site residues and accurately reproduced ATP interactions for all four models (derived from PDB: 2J0S, 1FUU, 2VSO, 2DB3). We then performed all-atom explicit-water molecular dynamics (MD) simulations for 500-700 ns to study conformational dynamics and atomic interactions of ATP-bound and ATP-unbound states. Yeast eif4A crystal structure (PDB:1FUU) in the open state, for example, illustrated closure of the two RecA domains upon ATP binding. As expected, ATP makes strong interactions with the N-terminal, while phosphate groups extend to the C-terminal interacting with arginines, bringing the two RecA domains together. In contrast, we did not observe domain closure in the same simulation with 1FUU without ATP bound. We also assessed 2J0S, a crystal structure of human closed eIF4A3 bound to ANP. ATP docked to this active site followed by 500 ns MD held the protein in the closed state with several interdomain interactions. Upon nucleotide removal, marked RecA separation occurred. We observed similar domain closure and opening for PDB: 2VSO and 2DB3. Pooling these results, we constructed two homology models of human eIF4A1 with both open (2VSO, 1FUU) and closed conformations (2J0S) as structural templates. We therefore have developed accurate and novel in silico models of eIF4A1 highly useful in assessing interactions of small-molecule ATPase inhibitors, with focus on the ATP-binding cleft. Disclosures No relevant conflicts of interest to declare.