Crystal structure of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) frameshifting pseudoknot

Abstract

SARS-CoV-2 produces two long viral protein precursors from one open reading frame using a highly conserved RNA pseudoknot that enhances programmed −1 ribosomal frameshifting. The 1.3 Å-resolution X-ray structure of the pseudoknot reveals three coaxially stacked helices buttressed by idiosyncratic base triples from loop residues. This structure represents a frameshift-stimulating state that must be deformed by the ribosome and exhibits base-triple-adjacent pockets that could be targeted by future small-molecule therapeutics.

INTRODUCTION

The connection between programmed −1 ribosomal frameshifting and pseudoknots was recognized in a coronavirus more than 30 yr ago (Brierley et al. 1989). Coronaviruses use frameshifting pseudoknots to regulate the ratio of translation of the precursor proteins Orf1a and Orf1b, which are produced from the same transcript and proteolytically processed into smaller proteins (for review, see V'Kovski et al. 2021). Frameshifting is essential for other viruses, notably retroviruses (Farabaugh 1996), and frameshifting RNA structures have been proposed as therapeutic targets because altering frameshifting inhibits viral infectivity (Brierley and Dos Ramos 2006; Plant et al. 2010).

Programmed −1 ribosomal frameshifting requires at least three mRNA features—a slippery sequence, a spacer, and a stable RNA structure, often a pseudoknot (Brierley et al. 1989; Farabaugh 1996). The slippery sequence is 5′ X XXY YYZ (X = any nucleotide, Y = A or U, and Z ≠ G). In SARS-CoV-2, this sequence is 5′ U UUA AAC, nucleotides (nts) 13,462–8 (in the SARS-CoV-2 reference genome, used throughout). Frameshifting from the UUA AAC to the UUU AAA frame eludes the stop codon at 13,481. While the slippery sequence alone results in low levels of frameshifting, the presence of a pseudoknot at a critical distance downstream from the slippery sequence greatly stimulates frameshifting (for review, see Farabaugh 1996). In addition to these core elements, upstream structures (Cho et al. 2013) and conformational heterogeneity (Omar et al. 2021; Rangan et al. 2021; Schlick et al. 2021) have been proposed to influence frameshifting.

Previously, the SARS frameshifting pseudoknot was proposed to consist of three helices or paired elements (P1–P3) connected by loops, a composition found across coronaviruses (Baranov et al. 2005; Plant et al. 2005). There is considerable variation among the predicted secondary structures of the SARS-CoV-2 genome (Lan et al. 2020; Manfredonia et al. 2020; Wacker et al. 2020; Huston et al. 2021; for review, see Kelly et al. 2021), which may reflect the multiple roles RNA plays in the virus life cycle. Recently, a 6 Å resolution cryo-electron microscopy (cryo-EM) structure of the SARS-CoV-2 frameshifting pseudoknot was determined, revealing a Y-shaped arrangement in which P3 branches away from a coaxial stack formed by P1 and P2 (Zhang et al. 2021). The construct used for that study encompassed sequences 5′ to the pseudoknot and thus that cryo-EM conformation may represent one adopted by the RNA before upstream nucleotides become sequestered inside the translating ribosome, which was in turn revealed by cryo-EM of a ribosome translating prior to the pseudoknot (Bhatt et al. 2021). Here, we report the crystal structure of an RNA lacking extensions 5′ to the pseudoknot, thereby more closely corresponding to the roadblock that the translating ribosome encounters. Our structure reveals all three helical elements stacked coaxially, with the three loops of the pseudoknot forming a previously undescribed stack of four base triples. Our high-resolution structure reveals potential binding sites for ligands to modulate frameshifting and inhibit virus infectivity.

RESULTS

Overall structure of the frameshifting pseudoknot

Previous studies showed that mutations in P2 of the SARS-CoV-2 pseudoknot modulate frameshifting (Plant et al. 2010), whereas mutations in P3 are tolerated; indeed, P3 is not strictly required (Plant et al. 2005). To reduce dimerization (Ishimaru et al. 2013) and obtain high-quality crystals (Supplemental Fig. S1A,B), we replaced sequences at the distal end of P3 (residues 13,514–13,522) with a base-paired segment capped with a GAAA tetraloop (Fig. 1A). The structure was solved at 2.1 Å resolution by the single-wavelength anomalous dispersion (SAD) method using data from an iridium derivative. This structure was then refined against a native data set extending to 1.3 Å resolution (Materials and Methods and Table 1). The experimental electron density map was of excellent quality (Supplemental Fig. S1C,D) and allowed complete tracing of the RNA, revealing a ∼87 Å-long coaxial stack (Fig. 1).

FIGURE 1.

Structure of the SARS-CoV-2 frameshifting pseudoknot. (A) Secondary structure. Arrows denote chain connectivity; Leontis and Westhof (2001) symbols, non-Watson–Crick base pairs. Numbering corresponds to the SARS-CoV-2 reference genome (NC_045512.2), except for residues 39–46 (outlined letters) introduced to facilitate crystallization. (B) Cartoon representation of the 1.3 Å-resolution structure of the pseudoknot, colored as in A. Nucleotides altered for crystallization are gray. Pink spheres denote chain connectivity in lieu of U13534, which is disordered. For alternate conformers, bases are in pastel shades. Red, lime, light purple, and purple spheres represent water molecules, Mg²⁺, Na⁺, and K⁺, respectively. Cobalt hexammine is colored blue.

TABLE 1.

Summary of crystallographic statistics

In our structure, three loops, L1–L3, connect the P1 and P2 helices. These helices and loops correspond to those in canonical H-type pseudoknots (Aalberts and Hodas 2005). L1 and L3 traverse the major and minor grooves of the stacked helices, respectively. While some pseudoknots lack L2, the SARS-CoV-2 frameshifting pseudoknot has a 2-nt L2. One of the L2 residues (G13493) forms a base triple with the L1 residue U13485 and the L3 residue A13537. This base triple brings together the three loops and extends base stacking between P1 and P2. P3 is formally an insertion into L3. As there is no gap between the 3′ end of P1 (residue 13,503) and the 5′ end of P3, there is no alteration of the helical pitch at the P1–P3 coaxial junction. Although appearing approximately straight from one viewpoint (Fig. 1B), the RNA bends ∼35° end to end (Supplemental Fig. S1C).

Noncanonical interactions produce idiosyncratic pockets

Our structure shows that the core of the SARS-CoV-2 frameshifting pseudoknot is formed by a stack of four consecutive base triples, in which nucleotides in P1, P2, L1, L2, and L3 are plaited together (Fig. 2). This base-triple stack was not predicted by previous models of the viral RNA element (Wacker et al. 2020; Bhatt et al. 2021; Roman et al. 2021; Zhang et al. 2021). In addition to the U13485•G13493•A13537 base triple formed by residues from the three loops, G13486 from L1 inserts into the major groove of the C13492•G13538 base pair at the bottom of P2 (Fig. 2B,C) and A13535 and C13536 from L3 insert into the minor grooves of the P1 pairs A13483•U13496, and G13484•C13495, respectively (Fig. 2D,E). The minor groove location of A13535 produces a type III A-minor interaction (Nissen et al. 2001). The high resolution of our structure (mean coordinate precision ∼0.13 Å, Table 1) reveals that well-ordered water molecules are integral to all base triples and extend the hydrogen bond network of the nucleobases of each triple with the exception of A13483•U13496•A13535 (Fig. 2E). For the G13486• C13492•G13538 triple, two water molecules hydrogen bond with one another and bridge the O6 of G13538 to the pro-R_P nonbridging phosphate oxygen (NBPO) of C13492 (Fig. 2B). For the U13485•G13493•A13537 triple, single water bridges the O6 of G13493 to the pro-R_P NBPO of U13485 (Fig. 2C). Finally, for the G13484• C13495•C13536 triple, a water molecule hydrogen bonds to the N3 of C13536 and bridges the sugar edge of G13484 via both the N3 and 2′-OH (Fig. 2D). Overall, this set of unique base-triple interactions, supported by ordered water molecules, gives rise to a highly structured, continuously stacked connection between the pseudoknotted P1 and P2 helices.

FIGURE 2.

Four base triples produce uninterrupted base stacking between P1 and P2. (A) Network of hydrogen-bonding interactions (black dashes) of the four base triples spanning the P1–P2 junction. Only water molecules bridging nucleobases are shown. Alternate conformers of G13538 and G13539 are light blue. (B) Axial view of the G13486•C13492•G13538 triple. Blue mesh shows a portion of the 2|F_o|–|F_c| simulated annealing-omit map contoured at 1σ. Hydrogen-bonding distances (Å) are indicated. (C) Axial view of the U13485•G13493•A13537 triple. The ribose of U13494 (whose base is extruded into solvent) extends the hydrogen-bonding network. (D) Axial view of the G13484•C13495•C13536 triple. The phosphate of U13494 extends the hydrogen-bonding network. (E) Axial view of the A13483•U13496•A13535 triple.

The SARS-CoV-2 frameshifting pseudoknot structure also exhibits idiosyncratic backbone and metal ion interactions (Fig. 3). In a canonical ribose zipper (Cate et al. 1996), the backbones of two strands approach closely, forming two reciprocal sets of hydrogen bonds between adjacent ribose 2′-OH's and the sugar edge of a nucleobase. The P1–L2–L3 junction is overwound, such that the 2′-OHs of C13495 and C13536 hydrogen bond (2.9 Å), but the O2 of the former is distant (3.6 Å) from the ribose of the latter (Figs. 2D, 3A). While that interaction is similar to one tier of a ribose zipper, in the tier above, extrusion of U13494 (Fig. 2C) and the sheared pairing of A13537 allow the Watson–Crick face of the purine to hydrogen bond to two 2′-OH groups from L2 (Fig. 3A). The geometry of this “snagged zipper” results in cross-strand stacking of A13537 on C13495. Moreover, the conformation results in a broadened major groove, in which a cobalt(III)-hexammine ion (presumably replacing a physiologic hexahydrated Mg²⁺) bridges the Hoogsteen face of G13484 and G13493 above it (Fig. 3B). Similarly, at the P1–P3 junction, a hydrated Mg²⁺ and K⁺ bridge the major groove edges of four tiers in a broadened major groove, hydrogen bonding to or coordinating with the Hoogsteen faces of G13503 from P1 and U13504, G13531, and U13532 from P3 (Fig. 3C). In addition to interactions of hydrated Mg²⁺, our structure contains an example of asymmetric µ-oxo chelation by the two physiologic cations Mg²⁺ and K⁺. In this arrangement (Fig. 3C) the divalent cation coordinates (2 Å) six oxygen ligands, one of which is shared with the monovalent ion, which coordinates it at a characteristic, longer distance (2.8–3.2 Å). The asymmetric coordination likely results from the mixed ions finely counterbalancing the local electrostatic potential of the noncanonical RNA structure.

FIGURE 3.

Ribose and ion interactions stabilize pseudoknot. (A) Interactions between the 2′-OH groups of G13493 and U13494 with the A13537 nucleobase are shown, along with the ribose zipper hydrogen bonding between 2′-OH groups of C13495 and C13536. (B) Interactions between a cobalt(III) hexammine with the O6 and N7 of G13493 and G13484. (C) Interactions among a hydrated Mg²⁺, K⁺, and four tiers of base pairs within the major groove. Distances (Å) are shown for the oxygen ligand shared between the Mg²⁺ and K⁺ in a µ-oxo arrangement. Hydrogen bonds between the hydrated Mg²⁺ and base pairs are colored black, K⁺-ligand bonds are represented by dotted red lines, and the µ-oxo bond involving K⁺ is represented by a thick dotted line. Yellow mesh shows a portion of the anomalous difference Fourier synthesis calculated with 1.771 Å X-ray data (Table 1), contoured at 3 σ.

Requirements for frameshifting efficiency in vitro

To understand how specific residues contribute to frameshifting, we performed in vitro frameshifting assays. As upstream motifs may affect frameshifting (Cho et al. 2013), we used an ∼0.6 kB portion of the SARS-CoV-2 genome surrounding the frameshifting pseudoknot (nts 13,202–800). In this assay, a Renilla luciferase in the 0 frame precedes the frameshifting element and a firefly luciferase in the −1 frame (see Materials and Methods). To reduce the effects of protein misfolding on luciferase activity, a foot-and-mouth disease virus 2A peptide is placed between each luciferase and the RNA element (de Felipe et al. 1999). Consequently, the Renilla luciferase, viral protein products encoded by the frameshifting element, and firefly luciferase are translated as separate protein products. Thus, coupled transcription and translation in vitro allows for measurement of luciferase activity and quantification of frameshifting (see Materials and Methods).

In addition to point mutations to assess the function of various pseudoknot residues, we tested a series of controls. These included inserting a UAA stop codon at the 3′ end of the Renilla luciferase (5′ control), inserting a UAA stop codon at the 5′ end of the firefly luciferase (3′ control), mutating the slippery site from UUUAAAC to CCGAAAC (SS control), and mutating the slippery site from UUUAAAC to CCGAAA (SSF control). The latter control places the firefly luciferase in the −2 frame such that the luciferase is out of frame even in the case of −1 frameshifting with the mutated slippery site. In each of these cases, frameshifting was greatly reduced compared to WT, which we report normalized to WT frameshifting (Table 2 and Materials and Methods). The 5′ control showed higher levels of readthrough than the 3′ control (0.067 ± 0.004 vs. 0.003 ± 0.001; P = 0.0002), suggesting that termination is read through more often in 0 frame than in the −1 frame. Frameshifting in the SS control was also greater than in the SSF control (0.032 ± 0.003 vs. 0.002 ± 0.0004; P = 0.0008), consistent with low levels of frameshifting despite the lack of a canonical slippery site, which is then further reduced when the firefly luciferase is in the −2 frame. Mutations were then tested to evaluate the essentiality of residues to frameshifting, as observed here and in other structures that include the SARS-CoV-2 frameshifting pseudoknot (Bhatt et al. 2021; Roman et al. 2021; Zhang et al. 2021). As summarized in Table 2, these include mutations prior to the slippery site and throughout the pseudoknot. (For comparison, the Rous sarcoma virus frameshifting element and CCR5 mRNA were also tested along with the same controls for each RNA motif. Frameshifting for the RSV frameshifting pseudoknot was 9.0% ± 1.6%, and the CCR5 motif showed no frameshifting [data not shown].)

TABLE 2.

Frameshifting efficiencies of pseudoknot variants

Mutations in and around L1 were relatively well tolerated, as G13484A was not statistically different from WT (0.84 ± 0.04; P = 0.15), and G13486A (0.57 ± 0.02) and G13486U (0.54 ± 0.06) were approximately twofold less efficient at frameshifting than WT, a statistically significant difference (P = 3 × 10⁻⁵ and P = 2 × 10⁻⁵, respectively). However, A13483G greatly reduced frameshifting (0.03 ± 0.005, P = 1 × 10⁻⁸), suggesting that a wobble pair is not tolerated at this position. Interestingly, U13485C frameshifted poorly (0.08 ± 0.04, P = 2 × 10⁻⁸) while U13485A was slightly enhanced compared to WT (1.08 ± 0.03, P = 0.04). Mutation of the relatively disordered residue C13487U also significantly elevated frameshifting compared to WT (1.34 ± 0.05, P = 0.006).

Mutations in and around L2 were more detrimental, as frameshifting efficiencies for mutations C13492U, G13493A, and G13493C were greatly reduced (values between 0.02 and 0.08, P = 1 × 10⁻⁸). Likewise, mutations to C13495U (0.25 ± 0.01) and U13496C (0.17 ± 0.04) also reduced frameshifting (P = 3 × 10⁻⁸ and 4 × 10⁻⁷, respectively), albeit to a lesser extent. As observed for L1 residue U13485, the U13494A mutation was tolerated and more efficient at frameshifting than WT (1.36 ± 0.04, P = 0.008) while the U13494C mutation was not (0.15 ± 0.01, P = 1 × 10⁻⁸).

Mutations in and around L3 were also largely detrimental to frameshifting. That is, A13535G (0.23 ± 0.03) and A13537G (0.30 ± 0.05) were moderately reduced (P = 4 × 10⁻⁸ and 2 × 10⁻⁸, respectively), and G13538C was further reduced compared to WT (0.06 ± 0.01; P = 1 × 10⁻⁸). In contrast, mutation of nearby disordered linker residue U13534A (0.86 ± 0.14) was not significantly different from WT (P = 0.8), and U13534C was reduced by less than twofold (0.67 ± 0.03, P = 0.002). Likewise, the naturally occurring mutation C13536U was not significantly different from WT (1.18 ± 0.10, P = 0.07).

Mutations made by Piccirilli and coworkers (Roman et al. 2021) for crystallization (U13474G and U13542G) were not significantly different from the WT plasmid (1.14 ± 0.03 and 1.20 ± 0.23; P = 0.1 and 0.3, respectively), suggesting that these interactions are not essential or inhibitory to frameshifting. Likewise, the naturally occurring variation observed in SARS-CoV-1 (A13533C) was not inhibitory to frameshifting (1.56 ± 0.03) and was significantly greater than WT (P = 0.0001). Mutations made upstream of the slippery site (U13460A, U13460C, U13461A, U13461C), which may be part of a stem–loop in the extended pseudoknot structure (Zhang et al. 2021), led to 32%–65% more frameshifting (Table 2), which was significantly greater than WT (0.14 < P < 0.049).

DISCUSSION

The frameshifting pseudoknots of the SARS-CoV-2 and SARS viruses are highly conserved, differing by a single nucleotide (A13533 vs. C, respectively). The SARS sequence would disrupt the U13504•A13533 pair, alter stacking of P1 and P3, and potentially increase flexibility in this region (Fig. 1). However, recently published small-angle X-ray scattering studies suggest that the two RNAs adopt similar shapes in solution (Kelly et al. 2020) and the frameshifting efficiency of A13533C is similar to or slightly better than WT (Table 2). This residue participates in a base triple in one high-resolution structure of the pseudoknot (Roman et al. 2021), which may be partly promoted in crystals by the U13474G mutation. As neither U13474G nor A13533C reduces frameshifting, the exact participation of the G13475•U13504•A13533 triple in frameshifting remains to be determined. In that structure (Roman et al. 2021), L3 residues U13534, A13535, and C13536 are disordered and the L1 pairing register has been displaced by one base stack such that G13486 interacts with G13493 and U13485 interacts with U13494. The latter pairing, which was also modeled in the ribosome-bound pseudoknot (Bhatt et al. 2021), is consistent with frameshifting assays (Table 2), as mutation of either residue to an A is tolerated by forming U–A or A–U pairs, while mutation to a C reduces frameshifting by disrupting the U–U pair. As U13494 is extruded in the structure here (Fig. 2C,D), disruption of frameshifting in U13494C could result from pairing with G13484, which would not occur in U13494A. Effects from mutating U13485, which is involved in a triple in the structure presented here (Fig. 2C), could be partly explained by the disruption of hydrogen bonds by U13485C. Accommodation of U13485A is less clear but would potentially involve interactions between A13485 and G13493. Nonetheless, frameshifting efficiencies of U13494 and U13485 variants are more simply explained by their pairing seen in other structures (Bhatt et al. 2021; Roman et al. 2021). In contrast, the lack of a strong effect on frameshifting by G13484A (Table 2), which introduces an A–C mismatch, is not well explained, as the residue is paired with C13495 here and in all reported structures (Bhatt et al. 2021; Roman et al. 2021; Zhang et al. 2021).

Natural variations of the SARS-CoV-2 pseudoknot have been documented (Neupane et al. 2020) and largely consist of substitutions that would preserve base pairing within paired regions (Supplemental Fig. S3); however, variations are observed at or near the P1–P2 junction. In general, frameshifting efficiencies for L2 variants are the most reduced in vitro, while L1 are the least reduced and L3 are intermediate (Table 2). The C13536U transition should retain interactions in the base triple with G13484•C13495 (Fig. 2D) as the N4 of C13536 is not involved. Consistent with this is the lack of an effect on frameshifting in this mutant (Table 2), which is similar to previous findings (Neupane et al. 2020) and consistent with the lack of interaction or disorder of C13536 in other structures (Bhatt et al. 2021; Roman et al. 2021). Both the C13492U and G13486A mutations would alter the base triple involving G13538 (Fig. 2B), and the G13486A mutation has been shown to affect frameshifting by approximately threefold in HEK293T cells (Bhatt et al. 2021). We find a modest reduction (approximately twofold) in frameshifting at this residue in vitro, either when mutated to an A or a C (Table 2). While mutation of this residue consistently reduces frameshifting by at least twofold, the residue is not as essential to frameshifting as many other residues throughout the pseudoknot, raising the possibility that the partially melted RPS3-bound state (Bhatt et al. 2021) occurs after frameshifting.

Comparison of our structure to those of other viral frameshifting pseudoknots shows how the basic H-type pseudoknot scaffold can be elaborated (Supplemental Fig. S4). The beet western yellow virus (BWYV) pseudoknot comprises only two helices. Two base triples, C8•G12•C26 and A24•G7•C14 form at the junction of its P1–P2 junction, and A25 (which would correspond to the extruded U13494 of the SARS-CoV-2 pseudoknot) interacts extensively with the triples. Mutation of any of the base triple residues at the BWYV P1–P2 junction resulted in the loss of frameshifting (Kim et al. 1999). In addition to triples at the P1–P2 junction (Supplemental Fig. S4A), loop 3 residues form triples with base pairs at the 5′ end of P1. Such triples are absent in the SARS-CoV-2 pseudoknot structure here but are observed in the structure by Piccirilli and coworkers (Roman et al. 2021) in the form of the G13475•U13504•A13533 triple. The murine leukemia virus recoding element (Houck-Loomis et al. 2011) contains a single base triple, which is located at the P1–P2 junction (Supplemental Fig. S4B). The coronavirus frameshifting pseudoknots, in addition to having more tertiary interactions than retroviral elements, also contain the third helix P3 (Plant et al. 2005), which our structure shows can stack coaxially on P1 (Fig. 1). P3 could participate in different interactions in the presence of upstream sequences such as those in the cryo-EM structure of the free RNA (Zhang et al. 2021). Recent chemical probing data (Lan et al. 2020; Manfredonia et al. 2020) provide evidence supporting the conformational plasticity of this coronavirus genomic region. The variation among the four structures now available of the frameshifting pseudoknot further highlights this motif's plasticity.

The SARS-CoV-2 pseudoknot is thought to fold prior to frameshifting; it therefore must be melted for translation of Orf1b. Prior to encountering the ribosome, the structure of the pseudoknot is likely dynamic. This can be inferred by comparing our structure to that adopted (Zhang et al. 2021) by the pseudoknot when 5′ sequences are present (Fig. 4) and is supported by in silico modeling (Omar et al. 2021; Rangan et al. 2021; Schlick et al. 2021), chemical probing (Lan et al. 2020; Manfredonia et al. 2020; Huston et al. 2021), and NMR (Wacker et al. 2020). On the ribosome, in which the pseudoknot is further bent, the interaction of ribosomal protein uS3 with G13486 appears to melt the G13486•C13492•G13538 triple, possibly destabilizing the pseudoknot and facilitating movement in the −1 frame (Bhatt et al. 2021), although the timing of G13486 melting with respect to frameshifting is still unclear. If the G13486•C13492•G13538 triple is not formed, either due to conformational plasticity or lack of pseudoknot refolding after ribosome passage, less frameshifting occurs (Table 2), and Orf1a translation proceeds until the UAA stop codon at U13481. A similar argument can be made for the alternative G13486•G13493•A13537 triple observed elsewhere (Roman et al. 2021). While mutation of G13486 only affected frameshifting by approximately twofold, mutation of either C13492 or G13538 eliminated frameshifting entirely (Table 2). Consequently, stability of the pseudoknot is initially required for frameshifting, but its conformational flexibility is necessary for translation to proceed. Multiple competing states may also give rise to fractional readthrough, analogous to riboswitch control of transcription termination (Hua et al. 2020).

FIGURE 4.

Model for pseudoknot plasticity during translation. Prior to encountering the ribosome, the frameshifting pseudoknot adopts multiple conformations. The pseudoknot structure determined by cryo-EM (Zhang et al. 2021) contains an additional helix (red) from nts 13,459–73, which includes the upstream slippery site and is absent from the coaxially stacked pseudoknot structure presented herein. As the ribosome proceeds past the slippery site, the fully stacked conformation may be favored.

Our structure suggests locations in the SARS-CoV-2 frameshifting pseudoknot that could bind to therapeutics. The pockets adjacent to G13486 could be targeted with small molecules, locking the nucleotide in place, enhancing the stability of the RNA, and possibly preventing key ribosome interactions. Alternatively, small molecules targeting a more open state could reduce pseudoknot stability and reduce frameshifting, potentially by destabilizing C13492•G13538, the mutation of which reduced frameshifting by more than 15-fold (Table 2). Sites in the pseudoknot at which three strands meet offer the highest information-content pockets (Warner et al. 2018), with the goal of disrupting residues (e.g., in L2) found to be most essential to frameshifting (Table 2). For example, the region between P1 and L3, although disordered, would be a relatively high information target that could modulate P1 stability. Moreover, cations at sites within the P2 major groove (Fig. 3B) and P1/P3 interface (Fig. 3C) could also be replaced with more specific ligands. Compounds have been identified that inhibit frameshifting in SARS and SARS-CoV-2 in cells (Park et al. 2011; Kelly et al. 2020; Bhatt et al. 2021; Sun et al. 2021). Future studies must reevaluate these compounds and others in light of our high-resolution state and aim to identify small molecules that modulate frameshifting.

MATERIALS AND METHODS

RNA preparation

DNA oligonucleotides (Supplemental Table S1) from Integrated DNA Technologies or Eurofins Genomics were resuspended in DEPC-treated water and used without further purification. Variant plasmids were prepared using the Ambion Site-Directed Mutagenesis Kit. For transcriptions, PCR templates were amplified from plasmids that encode a T7 promoter, an 81-nt hammerhead ribozyme (Ferré-D'Amaré et al. 1998), and the pseudoknot. The reverse PCR primers carried two 5′-terminal ribose 2′-methoxy substitutions (Kao et al. 1999). RNAs transcribed from PCR templates (as previously described; Jones and Ferré-D'Amaré 2014), were purified by denaturing gel electrophoresis and recovered from excised gel bands by electroelution. Eluted RNAs were concentrated in 10,000 molecular-weight cutoff centrifugal concentrators (EMD Millipore), washed successively with 0.5 M KCl and DEPC-treated water (three times), concentrated to 10–20 g/L, and filtered through 0.1 µm centrifugal filters. RNAs were stored at −20°C.

Crystallization and diffraction data collection

RNAs at 2–8 g/L in 25 mM HEPES-KOH and pH 7.4, 150 mM KCl were incubated at 95°C for 2 min and placed on ice for >5 min. MgCl₂ was then added to 10 mM and RNAs incubated at 37°C for 15 min, after which they were incubated at room temperature until crystallization setup. Hanging drops were prepared by mixing 1 µL of 4–6 g/L folded RNA with 1 µL of a reservoir solution comprised of 34%–40% (v/v) 2-methyl-2,4-pentanediol (MPD), 40 mM sodium cacodylate, pH 5.5, 20 mM MgCl₂, and 1 mM cobalt hexammine and equilibrated by vapor diffusion against a 0.5 mL reservoir at 21°C. Crystals grew in 2–3 wk as clusters with maximum dimensions 400 × 50 × 50 µm³. For the iridium derivative, crystals were transferred to 3 µL of 2 mM iridium hexammine, 40% v/v MPD, 40 mM sodium cacodylate, pH 5.5, 30 mM MgCl₂, 25 mM HEPES-KOH, pH 7.4, and 150 mM KCl, and equilibrated for 1 h at 21°C over a 0.5 mL reservoir of the same solution with cobalt hexammine instead of iridium hexammine. Crystals were mechanically separated, transferred into the same solution without iridium hexammine briefly, mounted on nylon loops, and flash frozen by plunging into liquid nitrogen. For the native data set, crystals were transferred from the same growth conditions above to 3 µL of 1.5% v/v dimethyl sulfoxide, 40% v/v MPD, 40 mM sodium cacodylate, pH 5.5, 30 mM MgCl₂, 25 mM HEPES-KOH, pH 7.4, and 150 mM KCl, and equilibrated for 7 d at 21°C over a 0.5 mL reservoir of the same composition. Crystals were mounted on nylon loops and flash frozen. Diffraction data were collected in rotation mode at 100K at beamlines 24-ID-C and 24-ID-E of the Advanced Photon Source, Argonne National Laboratory, and reduced with XDS (Kabsch 2010). For the high-energy native data set, the first 450 frames (90° total rotation) from two well-diffracting crystals were merged with AIMLESS (Evans and Murshudov 2013) in CCP4 (Winn et al. 2011). A low-energy data set was collected with 1.771 Å X-radiation from one of the same crystals (Table 1).

Structure determination and refinement

Two iridium sites were identified by Autosol in Phenix (Adams et al. 2010), yielding an initial mean overall figure of merit of 0.418. After density modification, experimental maps (Supplemental Fig. S1C) allowed near-complete chain training in Coot (Emsley and Cowtan 2004) aided by RCrane (Keating and Pyle 2012). Initial rounds of the model building were interspersed with refinement in Phenix using simulated annealing, energy, real space, and individual atomic B-factor refinement. For the later stages, simulated annealing was no longer used, and, in the last stages, TLS refinement was enabled (Table 1). In addition to the two strongest iridium hexammine sites, one other site with a weak anomalous signal was evident in the anomalous difference Fourier synthesis. However, iridium hexammine or cobalt hexammine built at this weak site refined to high B-factors (∼190 Ų); therefore, the site was modeled as K⁺. For other ion and water sites, features were modeled and refined initially as water. After refinement, waters lacking electron density at 1.0 σ in the σA-weighed 2|F_o|–|F_c| map were removed. Sites with positive residual |F_o|–|F_c| features were modeled as Mg²⁺. For one site, the residual feature was still not satisfied and was modeled as K⁺. At three sites, positive residual |F_o|–|F_c| features are present at 4.0–5.0 σ, suggesting alternative conformations, but were not built. These are the iridium hexammine adjacent to G13493, as well as the phosphates of A13526 and U13529. Though C13487 has a high B-factor (Supplemental Fig. S2A), a small negative |F_o|–|F_c| feature suggests it to be in the syn glycosidic conformation.

For the native data set, the model above was used for molecular replacement with Phaser (McCoy et al. 2007), after removing all ions and residues C13487, U13525, and A13526 (TFZ score and LLG of 38.9 and 4276, respectively). Initial rounds of refinement used simulated annealing, energy, real space, and isotropic B-factor refinement and focused on building the most ordered ions and waters. Later stages of refinement included refinement of occupancies for residues with alternative conformations, and occupancies of ions at special positions were fixed to 0.5. For C13487, G46, C13523, A13533, U13534, G13538, and G13539, two alternative conformers were built. For C13487, the nucleobase is disordered and two conformations were built. For G46 and C13523, residual density was present for the phosphate and C5′/C4′ atoms, so alternative conformers were built. Similar residual density was observed for G13538 and G13539, so alternative conformers were built as well. Along with the preceding residue A13533, the phosphate of U13534 is present in two conformers while the ribose and nucleobase are not modeled. A single cobalt hexammine ion (Fig. 3B) was placed owing to the strong anomalous feature at that site. Fully or partially hydrated Mg²⁺ was placed at sites at which octahedral coordination was apparent and H₂O–Mg²⁺ distances were ∼2.1 Å. For one Mg²⁺ ligand, residual density was observed adjacent to the water consistent with a solvent molecule. For this ligand, two atoms of an MPD were modeled. K⁺ was placed at two sites for which anomalous peaks were present in the low-energy data set. Na⁺ was placed at sites for which residual density was present after refinement with water but no anomalous peak was present in the low-energy data set. In the final stages of refinement, individual anisotropic B-factor refinement was used for all atoms except waters, resulting in a 2% decrease in R_free. Structure figures were prepared with PyMOL (Schrödinger, LLC).

Size-exclusion chromatography

An amount of 50 µL aliquots of RNA were folded at 2 g/L or 8 g/L and injected in their entirety onto a Superdex 200 10/300 GL column (Cytiva) run at 0.75 mL/min. The matching RNA folding buffer and mobile phase were 25 mM HEPES-KOH, pH 7.4, 150 mM KCl, and 10 mM MgCl₂.

Dual-luciferase frameshifting assay

A portion of the SARS-CoV-2 genome surrounding the frameshifting pseudoknot (nts 13,202–800, NC_045512.2) was Gibson cloned (New England Biolabs) into pSGDLucV3.0 (Addgene plasmid #119760) linearized with BglII and PspXI (New England Biolabs). Site-directed mutagenesis was used to obtain variants of this plasmid (Table 2), which were diluted to 0.2 g/L. In vitro translation reactions were performed using the TnT T7-Coupled Rabbit Reticulocyte Lysate System (Promega) in 10 µL reactions including 0.1 µg plasmid per reaction and otherwise following the manufacturer's protocol (i.e., 90 min incubations at 30°C). Luciferase activities were measured using the Dual-Luciferase Reporter Assay System (Promega) in a Tecan Infinite F200 plate reader. Triplicate aliquots of 2 µL of the translation reaction were diluted 10-fold with 1× passive lysis buffer and transferred to white flat 96-well microplates (Greiner). Luciferase reactions were initiated by the addition of 50 µL of each luciferase reagent prior to recording luminosity measurements. Frameshifting values were calculated by dividing the luminosity of firefly luciferase by the luminosity of Renilla luciferase for each well, and technical replicates of frameshifting were averaged. At least three independent experiments were conducted for each variant, and frameshifting efficiencies were calculated by dividing by frameshifting of the WT plasmid performed during the same experiment and were averaged for all independent experiments (Table 2). The average amount of frameshifting of the WT plasmid was 34.8% ± 3.6% (s.d., n = 8, independent experiments) with a range of 30.9% ± 0.4% to 40.4% ± 1.2% (s.d., n = 3, technical replicates). For statistical tests, P values were calculated by comparing frameshifting values via one-tailed or two-tailed Student's t-tests.

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.

MEET THE FIRST AUTHOR

Christopher P. Jones.

Meet the First Author(s) is a new editorial feature within RNA, in which the first author(s) of research-based papers in each issue have the opportunity to introduce themselves and their work to readers of RNA and the RNA research community. Christopher Jones is the first author of this paper, “Crystal structure of the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) frameshifting pseudoknot.” Chris is a postdoctoral researcher in the laboratory of Adrian Ferré-D'Amaré at the National Heart, Lung, and Blood Institute of the NIH. Chris's research focuses on RNA structural biology and RNA–small-molecule interactions.

What are the major results described in your paper and how do they impact this branch of the field?

SARS-CoV-2 uses a frameshifting pseudoknot to control the ratio of translation of the two major nonstructural polyproteins of SARS-CoV-2. I present a model for the SARS-CoV-2 frameshifting pseudoknot from X-ray data that is among the highest resolution data ever reported for an RNA of this size. I also use in vitro frameshifting data to evaluate the importance of interactions observed in the structure. In light of other structures determined for this motif, this RNA is dynamic, which is likely important for function.

What led you to study RNA or this aspect of RNA science?

Frameshifting pseudoknots are excellent targets for small molecules, and many groups are actively trying to identify ways to inhibit this particular RNA motif, given its essentiality to coronavirus replication. Evaluating frameshift-modulating therapeutics greatly benefits from structural data, especially at high resolution as presented here. As many viruses use frameshifting to control aspects of their replication, this particular motif was an attractive model beyond its therapeutic potential.

What are some of the landmark moments that provoked your interest in science or your development as a scientist?

Personally, there is nothing quite like the first time you stare at a freshly phased electron density map, when the structure is there but—at the modest resolutions mostly achieved for RNA—ill-defined and amorphous, yet tangible. The twinkle of results just out of reach is the dragon we chase and the sudden revelation of data provided by a piping hot map is quite the hit.

If you were able to give one piece of advice to your younger self, what would that be?

Spend more time with deliberately idle hands. My work has been quite fruitful during the 5-d-on-5-d-off laboratory schedule as it dedicates time for deeper reflection and analysis.

Are there specific individuals or groups who have influenced your philosophy or approach to science?

Robert Pirsig's Zen and the Art of Motorcycle Maintenance describes approaches to problem solving that influenced me tremendously in high school. While outdated, The Sheltering Sky (Paul Bowles) introduced me to a type of mindfulness that is important to my approach. But by far, the students and post-bacs that I have mentored have taught me far more about the variety and diversity of approaches one can take to perform and enjoy experiments.