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

Despite the presence of various multiple-occupancy clusters, compared to the smaller stem-loops modeled in the 5′ UTR, the frameshifting element models were less converged. The majority of variation between top models arises from different orientations of the 5′ "slippery sequence," which forms a weak stem-loop of variable orientation in Rosetta-FARFAR2 simulations. The convergence of the pseudoknot region excluding this 5´ slippery sequence was tighter, 8.7 Å, suggesting that lowenergy conformations of this pseudoknot element are captured by the current models. Models with both of the assumed secondary structures produced low energies; they contribute to distinct but similar clusters. We additionally generated over 20,000 FARFAR2 models for the dimerized frameshifting element. Modeling of the dimerized frameshifting element did not converge on a welldefined 3D structure (each of the 10 lowest energy conformations were 'singlets' with no other conformations discovered within 5 Å RMSD; Table 1 and Supplementary Figure 2 ). Fig. 4 presents models for structured regions of the 3′ UTR. The 3′ UTR includes a switch-like pseudoknot element on the 5′ end, a hypervariable region, and the stem-loop II-like motif (Fig. 4A ); all these regions have secondary structures built based on homology to other betacoronaviruses. 2, 17 The 3′ UTR pseudoknot along with its mutually exclusive stem-loop structure have suggested functional roles in viral RNA synthesis, with mutations that destabilize either the pseudoknot or the stem-loop structure proving inviable in related betacoronaviruses. 8 The structure of the stem-loop IIlike motif resembles that of an rRNA loop, leading to its proposed role in recruiting host translation machinery. 5 We generated over 10,000 models for the full 3´ UTR. Analogous to the models generated for the extended 5′ UTR, this simulation was far from reaching convergence (Table 1) , but suggested the possibility of stereotyped RNA 3D structure, perhaps recognized by factors such as the virus . CC-BY 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.04.14.041962 doi: bioRxiv preprint replicase (Supplementary Figure 3) . We additionally generated over 20,000 models for two large subregions of the 3´ UTR, the hyper-variable region and a region including the 3′ UTR pseudoknot (Supplementary Figure 3) . Here we built models for the 3′ UTR pseudoknot based on the secondary structure forming the pseudoknot P1Pk as depicted in Fig. 4A . Future simulations of this region should also include models based on the alternative secondary structure for this switch-like element forming a stem extending P0b. In both cases, we were not able to observe convergence; none of the top 10 lowest energy conformations were similar to each other or to any of the other top 400 lowest energy conformations, as evaluated by RMSD with cutoff 5 Å. This was expected: with nearly 150 nucleotides in the 3′ UTR pseudoknot and with the hyper-variable region having limited homology to prior structures, these two modeling cases were the largest sub-domains modeled in this report, and likely will remain challenging to model even with more computational power. Modeling of smaller sub-domains of the 3′ UTR and/or acquisition and incorporation of experimental data will likely be critical to model reasonable starting ensembles for virtual docking algo