De novo 3D models of SARS-CoV-2 RNA elements and small-molecule-binding RNAs to guide drug discovery

Ramya Rangan; Andrew M. Watkins; Wipapat Kladwang; Rhiju Das

In advance of detailed experimental structural characterization, computational predictions for the 3D structural conformations adopted by conserved RNA elements may aid the search for RNA-targeting antivirals. Representative conformations from these RNA molecules' structural ensembles can serve as starting points for virtual screening of small-molecules drug candidates. For example, in prior work by Stelzer, et al., 6 virtual screening of a library of compounds against an ensemble of modeled RNA structures led to the de novo discovery of a set of small molecules that bound a structured element in HIV-1 (the transactivation response element, TAR) with high affinity. Such work motivates our modeling of not just a single 'native' structure but an ensemble of states for SARS-CoV-2 RNA regions. As with HIV-1 TAR, many of the SARS-CoV-2 elements are unlikely to adopt a single conformation but instead may sample conformations from a heterogeneous ensemble. Furthermore, transitions among these conformations may be implicated in the viral life cycle, as they change longrange contacts with other RNA elements or form interactions with viral and host proteins at different steps of replication, translation, and packaging. A possible therapeutic strategy is therefore to find drugs that stabilize an RNA element in a particular conformation incompatible with conformational changes and/or changing interactions with biological partners at different stages of the complete viral replication cycle. Consistent with this hypothesis, prior genetic selection and mutagenesis experiments stabilizing single folds for stem loops in the 5′ UTR and the pseudoknot in the 3′ UTR demonstrate that changes to these RNA elements' structural ensembles can prove lethal for viral replication. [7] [8] [9] Here, we provide de novo modeled structure ensembles for key RNA elements in the SARS-CoV-2 genome obtained from Rosetta's protocol for Fragment Assembly of RNA with Full-Atom Refinement, version 2 (FARFAR2). 10 These structures include de novo models for stem loops 1 to 7 (SL1-7) in the extended 5′ UTR, the frameshifting element and its dimerized form, the 3′ UTR pseudoknot, and the 3′ UTR hypervariable region, along with homology models of SL2 and s2m. The use of Rosetta FARFAR2 is motivated by extensive testing: FARFAR2 has been benchmarked on all community-wide RNA-Puzzle modeling challenges to date, [11] [12] [13] achieving accurate prediction of complex 3D RNA folds for ligand-binding riboswitches and aptamers, and producing models with 3-14 Å RMSD across six additional recent blind modeling challenges. 10 In addition to providing structural ensembles for SARS-CoV-2 RNA elements, we provide analogous FARFAR2 de novo and homology models for 10 riboswitch aptamers, in the hopes of providing a benchmark dataset for virtual screening approaches that make use of computational RNA models.