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

With these caveats and metrics of convergence in mind, a description of models from the extended 5′ UTR, frameshifting element, and 3′ UTR follows. Fig. 1 presents models for the stem loops that make up the extended 5´ UTR. Also called the 5′proximal region, this region extends the 5´ UTR by ~100 residues to bracket potential structures that involve the beginning of the coding region. The assumed secondary structure depicted in Fig. 1A is largely based on previous dissection of betacoronavirus secondary structures by several groups. 2, 17 More specifically, secondary structures for SL1-5 in the 5' UTR are based on homology to prior betacoronaviruses, where these conserved stems have been confirmed through genetic experiments and sequence alignments in related betacoronaviruses. (We note for non-coronavirus researchers here that SL1, SL2, SL4, and SL5 have also been termed SLI, SLII, SLIII, and SLIV in an important set of studies. [18] [19] [20] Some of these stems have structural preferences that have proven critical to the viral life cycle based on genetic experiments in other betacoronaviruses; these preferences have the potential to be altered through the binding of small-molecule drugs. Prior genetic selection experiments have demonstrated a preference for mutations that destabilize SL1. The lower part of this stem must unpair to allow for the formation of a long-range RNA contact between the 5′ and 3′ UTRs. 9 The stem must also presumably unfold to enable cap-dependent initiation of translation by the human ribosome. The loop in SL2 has sequence features across betacoronaviruses consistent with a U-turn conformation, and mutations that disrupt this structure are not viable, leading to a loss of subgenomic RNA synthesis. 7 SL3 presents the transcription regulation sequence of the leader (TRS-L), which must be available to base-pair with TRS-B binding partners in the negative-strand viral genome to facilitate sub-genomic RNA synthesis. 21 SL4 has been shown to be functionally important, and also harbors an upstream open reading frame (uORF) across many betacoronaviruses. Here computational secondary structure modeling returned an additional stem immediately 3′ of SL4, and we have modeled them together as SL4ab. SL5 is a well-established domain that, in SARS-related viruses, has a long stem elaborated with a 4-way junction; this element has been proposed to harbor packaging signals, and it harbors the AUG start codon for the genome's first gene product, the ORF1a/b polyprotein. Stems SL6 and SL7, downstream of the start codon, were modeled through computational secondary structure prediction. SL6 and SL7, computationally predicted to occur downstream of the AUG start codon, are included as potentially analogous to stems discovered to be important in bovine coronavirus but not yet established in SARS-related viruses. 22 We first produced models for the full extended 5′ UTR (over 10,000 FARFAR2 models), with the topscoring structure depicted in Supplementary Figure 1 . With the current level of sampling, this and other lowest energy structures appear only once amongst our models (Table 1) , reducing any confidence that these models capture lowest energy conformations in reality. A construct of this size would require substantially more models to achieve sufficient modeling convergence. Nevertheless, the top-scoring models suggest the potential for compact RNA structures for the 5´ UTR mediated by potential tertiary contacts between stem